Sean Pitman on evolution of mitochondria

seanpit: Your false claims that I drew the line at 7-letter words is nothing but an obvious strawman since it’s quite clear that I’ve always drawn the line at 1000-character sequences. So you retract your claim that “If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words" (evolution as characterized by stepwise selectable steps)? Zachriel

See also the latest from James Tour: https://youtu.be/_zQXgJ-dXM4 seanpit

Zachriel,

A rotating ion pump could have been coopted for motility.

Not true. There are simply too many required modifications (that are not sequentially beneficial) for such a successful cooptation - as described in detail at: DetectingDesign.com/flagellum.html And, as far as your Word evolution algorithm, it shows an exponential decline in evolutionary potential, just like I said it would, with each increase in the minimum size. This particular program of yours actually supports my main argument! Your false claims that I drew the line at 7-letter words is nothing but an obvious strawman since it's quite clear that I've always drawn the line at 1000-character sequences. Your Phrase evolution algorithm, on the other hand, is based on template matching to any portion of a long sequence regardless of it's own meaning/function - which doesn't resemble the Darwinian mechanism of RM/NS. seanpit

seanpit: Improved meaning/function is based on context. That's fine. You said @36, “Select based on changes in beneficial function”. It's up to you to provide an operational definition. Mung: Correct me if I am wrong, but I thought you were the one with the word mutagenesis program. Yes, it was designed to test this claim:

Sean Pitman: If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” Turns out there are stepping stones.

Mung: What does your objective function look like? Words in the dictionary per this statement:

Sean Pitman: say you start with a short sequence, like a two or three-letter word that is defined or recognized as beneficial by a much larger system of function, such as a living cell or an English language system. Try evolving this short word, one letter at a time, into a longer and longer word or phrase. See how far you can go. Very quickly you will find yourself running into walls of non-beneficial function.

bill cole: I have looked at the explanations for interim functionality of flagellum proteins and find them very unconvincing. A rotating ion pump could have been coopted for motility. Zachriel

Zachriel

It’s thought that the components of the flagellum had other functions that were coopted for motility. In any case, there is ample evidence of complex biological morphological evolution, such as in vertebrates, so the argument about complexity is already contradicted by substantial evidence.

I have looked at the explanations for interim functionality of flagellum proteins and find them very unconvincing. In the case of this motor you have to get 20 to 30 proteins to fit together in form and function so highly sequence specific. There is no good explanation for this based on the laws of chemistry in physics. Why would an existing protein be just by chance a charge and form fit to build this highly specified molecular machine? You have at least 4^50000 of sequential space in the genome to organize. What mechanism do you propose? bill cole

Zachriel: Notably, Sean Pitman still can’t provide an operational definition of “beneficial function” with regards to word evolution. Correct me if I am wrong, but I thought you were the one with the word mutagenesis program. What does your objective function look like? Mung

Zachriel,

Did you want to try some examples? Which has more “beneficial function”? “king” or “gdafmg” “king” or “the king” “the king” or “the king shall” “the king shall” or “the king shall drink” “to be or not to be” or “To thine own self be true”

Improved meaning/function is based on context. What ultimate “goal” is being achieved? For biological evolution, the goal for the Darwinian mechanism is to produce enhanced survival/reproduction. The goal for the Darwinian mechanism is not to produce increased functional complexity – which is supposed to be produced as a sideline to the primary effort of enhancing the primary goal of survival/reproduction. So, when you ask, “Which has more beneficial function?”, the answer must be based on what the changes produce regarding some ultimate goal for the “organisms” in your population. Increasing sequence length, by itself, does not necessarily enhance function toward some particular goal. Keeping this concept in mind, what would be the beneficial difference between phrases like: “to be or not to be” or “To thine own self be true”? The first phrase is 18 characters in length while the second is 25 characters in length. Yet, what is the functional benefit in going from the 18-character sequence to the 25-character sequence here? It’s not obvious to me. What ultimate functional goal is enhanced here? The same thing is true regarding “phrases” marked as selectable in your Phrasenation program, such as: “give the first” “second hit or” “than that which” “me the cups” Upon what basis can you argue that any one of these “phrases” is more functionally beneficial compared to the others? Yet, your algorithm defines them all as selectable based only on their match to the target sequence Hamlet. Now, compare this to a bacterium gaining a flagellar motility system. The ultimate goal of survival/reproduction is clear because such a bacterium would gain superior access to food – and therefore improve its survival/reproductive advantage. If you wish to model the Darwinian mechanism you need to set up such a situation. seanpit

Notably, Sean Pitman still can’t provide an operational definition of “beneficial function” with regards to word evolution. Zachriel

Zachriel,

It’s thought that the components of the flagellum had other functions that were coopted for motility. In any case, there is ample evidence of complex biological morphological evolution, such as in vertebrates, so the argument about complexity is already contradicted by substantial evidence.

Oh please. Again, evolutionists always assume evolutionary ancestry when they see some homology here or there within subsystems of a very complex system. The problem as I’ve already mentioned, is that homologies can be and are often produced by intelligent design. Object oriented computer programming, for example, is extensively based on homologies within subsystems of more complex programming. Therefore, reference to homology, by itself, does not support a default assumption of a Darwinian origin vs. a common designer without a viable Darwinian mechanism. And, when it comes to highly complex systems, like the flagellar motility system, co-opting the required components to produce the subsystems along the proposed evolutionary pathway toward full flagellar motility is not remotely as simple as evolutionists, like Zachriel or Matzke imagine. It is not as simple as a single cut-and-paste translocation mutation. Many additional mutations would be required to get a protein, that was working as part of a different system of function, to then work as part of the new system of function. The fact of the matter is, because of this problem, there are simply no examples of evolution in action at this level of functional complexity. It just doesn't happen and cannot happen, via a Darwinian mechanism, for very good statistical reasons. Details of this problem are discussed on my website: DetectingDesign.com/flagellum.html seanpit

bill cole: Zachriel seems to be trying to simulate beneficial function with small words where in the biological world some of those words would have to be 500 characters or more. The proposed test of letterspace was made by Sean Pitman. As he can't provide an operational definition of "beneficial function", his proposed test is undefined. bill cole: Until it can move through the medium whats the benefit? It's thought that the components of the flagellum had other functions that were coopted for motility. In any case, there is ample evidence of complex biological morphological evolution, such as in vertebrates, so the argument about complexity is already contradicted by substantial evidence. Zachriel

Sean Zachriel

So, you see, there is an equivalent spectrum in real life. However, as one moves step-by-step up the ladder of functional complexity that exists within this spectrum, the average time needed for the Darwinian mechanism to achieve the next level increases exponentially… which is what Zachriel is having difficultly getting his mind around. He really is having trouble getting his mind around the reality of the truly “big numbers” involved with this situation for the Darwinian mechanism – and what they really mean.

I agree, in the debate you cited Nick was having trouble conceptualizing the magnitude of the sequential space. I believe this is why GA's fail without a target. Zachriel seems to be trying to simulate beneficial function with small words where in the biological world some of those words would have to be 500 characters or more. In the biological world beneficial function is a moving target as you evolve complexity. There is flexibility in the sequence of the first protein of the flagellum the following proteins get more and more specific to have beneficial function i.e. the bacteria can move though a medium. Until it can move through the medium whats the benefit? bill cole

bill cole: I will think about beneficial function but at first blush as you add complexity the definition may move i.e. a stand alone enzyme vs a nuclear protein that binds with 14 different proteins. We're looking for an operational definition of "beneficial function" appropriate to letter sequences. That way we can explore phrasespace and see if it behaves as seanpit has asserted. seanpit: You didn’t falsify my position regarding an exponential increase in the average time required to make additional steps of the ladder We falsified that you had to cross oceans to reach a 7-letter word. There are stepping stones. seanpit: If you want to model this type of evolution with English literature ... It's your model @36. seanpit: then selection based on entire sentences with additional selection based on entire paragraphs, then entire chapters, etc. You forgot phrases, per your original contention. seanpit: In such a situation, remember that the mutations within genomes can cut and paste anywhere within a sequence Been there, done that. You forgot to provide an operational definition of "beneficial function" per @36. That means being return a measure of "beneficial function" for a given sequence. Did you want to try some examples? Which has more "beneficial function"? "king" or "gdafmg" "king" or "the king" "the king" or "the king shall" "the king shall" or "the king shall drink" "to be or not to be" or "To thine own self be true" Zachriel

bill cole,

I do think you have an argument with a 7 letter word but in nature the words are much longer.

In nature there are very short proteins that are beneficially functional (equivalent to short 7-letter words). And, there is a spectrum between these very short functional proteins and much longer functional single-protein systems (going well over 1000aa in size). Then, beyond this, there are systems that are based on multiple specifically arranged proteins (like ATPsynthase or flagellar motility systems that collectively require a minimum of several thousand fairly specified amino acid residues). So, you see, there is an equivalent spectrum in real life. However, as one moves step-by-step up the ladder of functional complexity that exists within this spectrum, the average time needed for the Darwinian mechanism to achieve the next level increases exponentially... which is what Zachriel is having difficultly getting his mind around. He really is having trouble getting his mind around the reality of the truly "big numbers" involved with this situation for the Darwinian mechanism - and what they really mean. seanpit

Zachriel,

seanpit: Why then do you act like you’ve somehow disproved my original argument regarding a need for exponentially greater amounts of time with each step up the ladder of functional complexity?

What we directly falsified was “If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.”

Just the opposite is true. You didn’t falsify my position regarding an exponential increase in the average time required to make additional steps of the ladder; you actually supported it! Don’t you see that? Just because the “swim” appears to be relatively short at the 7-character level doesn’t mean that the swim doesn’t take exponentially greater amounts of time compared to 1 or 2-letter words. The entire point of this exercise was to get you to see the exponential nature of the problem with each step up the ladder. Again, why do you think that I drew the line at 1000? - instead of at 7?

seanpit: Your real issue, then, is in regard to your “Phrasentation” algorithm – an algorithm that does not show an exponential decline in evolutionary potential because the targets are pre-defined as being just a very short Levenshtein step away from the next target.

Shakespeare wrote Hamlet to minimize Levenshtein distances as well as in blank verse?! What a genius!

Yes, Shakespeare was a genius! Unfortunately, the Darwinian mechanism is not. Intelligent design can cross large Levenshtein distances (because of an ability to imagine the future) while a mindless Darwinian algorithm simply cannot do this in a reasonable amount of time.

seanpit: Of course, if any match, however small, to a template sequence is defined as “selectable”, there will in fact be a linear relationship between time and the length of the sequence!

No. They have to be whole word snips of Hamlet based on the idea that any consecutive words of Shakespeare are more “beneficial” than random words or letters. You expressed displeasure with the use of Hamlet as a phrase dictionary, even though your original claim concerned a word dictionary. That’s fine… Just provide an operational definition of “beneficial function” for letter sequences so we can test your claim by your proposed process @36.

Richard Dawkins’ “Methinks it is like a weasel” algorithm worked based on template matching where each additional letter match was defined as selectable. The only difference between his algorithm and yours is that you moved it up a bit too single words instead of single letters. However, the basic concept of template matching remains the same between his algorithm and yours. That is why both of your multi-word evolution algorithms work in a linear manner instead of an exponential manner – as is always the case for template matching algorithms for very simple statistical reasons. Again, when selection is defined based on a match of fairly consistent size to a pre-defined larger target sequence, regardless of the enhanced or non-enhanced function of the smaller evolving sequences, the amount of time required will increase in a linear manner as the size of the target sequence increases. That is why your Phrasenation algorithm worked in a linear manner while your word-evolution algorithm worked in an exponential manner. In real life, selection is not made based on template matching where a match of a small sequence of a particular size to a larger sequence is defined as selectable. That simply isn’t how natural selection works in real life. Natural selection works based on an improvement in beneficial function compared to what already exists in the genome – where a beneficial function is defined, in the context of Darwinian evolution, as enhanced survival/reproductive fitness among one’s peers within a given environmental context. There simply is no other definition for “beneficial function” when it comes to Darwinian-style evolution. If you want to model this type of evolution with English literature (without having to actually base selection on improved beneficial meaning/function according some kind of survival/reproductive advantage), as I mentioned earlier, you cannot use essentially the same size of sequence (i.e., an average of 5-letter words) to base your selection compared to some large meaningful target sequence (like Hamlet). A better model could start out with selection for English word matches as in your word-evolution algorithm (but with more reasonable reproductive/mutation rates and the allowance for random walks), then selection based on entire sentences with additional selection based on entire paragraphs, then entire chapters, etc. In such a situation, remember that the mutations within genomes can cut and paste anywhere within a sequence – not just between select “words” or intact “paragraphs”. Given such a situation, you might have all the words you need to produce an entire sentence, but getting them cut and pasted together in the proper order would take exponentially longer than it did to make individual words. The same would be true when it comes to making entire paragraphs, with a bit of a twist. Your genome size may not be large enough to store all the sentences that might be needed to produce any selectable paragraph at a particular point in time. This would only increase the exponential nature of the problem and would reflect a similar problem for evolution within living things. With living things, as the size of the needed translocation(s) increases in a linear manner, the odds that just the right sequence will already exist (preformed somewhere within the overall genome of sequence options) will decrease in an exponential manner. This is because the size of the gene pool is limited because of the finite environmental limitations on population sizes. After a certain point, the population size can no longer increase and this puts a finite limitation on the size of the population’s library of stored sequence options. Once this library size limitation is maxed out, it becomes dramatically more and more difficult to evolve functionally complex systems at higher and higher levels. seanpit

Zachriel This one is complex. I will think about beneficial function but at first blush as you add complexity the definition may move i.e. a stand alone enzyme vs a nuclear protein that binds with 14 different proteins. Honestly in my mind the sequential space enormity over 100 aa sequence makes the current mechanisms highly unlikely. I do think you have an argument with a 7 letter word but in nature the words are much longer. bill cole

bill cole: From reading the discussion I do not believe Zachriel or Nick understands the enormity of the problem. We're more than willing to explore the landscape in question to make that determination. For instance, we know that you don't have to cross an ocean of meaningless sequences to find 7-letter words. bill cole: Once you leave the island you are forever lost in an infinite ocean. That's the claim. We'd be happy to explore the wordscape to verify that prediction. We just need a valid measure of "beneficial function". Zachriel

seanpit Thanks again for siting your website discussions with Nick regarding sequential space and the bacterial flagellum potential evolution. I think that the difficulty of convincing someone of the problem you have surfaced is that the size if sequential space of sequences is not easy to comprehend. From reading the discussion I do not believe Zachriel or Nick understands the enormity of the problem. For all practical purposes we are dealing with exploration through infinity. Once you leave the island you are forever lost in an infinite ocean. There may be many functional possible proteins but because the space of the ocean in infinite you never land on one. In order to solve the problem through random change you need the target sequence as a guide. Any thing short of having a direct target to reference and you spent eternity in the ocean. Sequences are the largest mathematical spaces in the universe. They are great for creating almost infinite possibilities, but they are terrible for finding function through a random search. bill cole

seanpit: Of course, if any match, however small, to a template sequence is defined as “selectable”, there will in fact be a linear relationship between time and the length of the sequence! No. They have to be whole word snips of Hamlet based on the idea that any consecutive words of Shakespeare are more "beneficial" than random words or letters. You expressed displeasure with the use of Hamlet as a phrase dictionary, even though your original claim concerned a word dictionary. That's fine... Just provide an operational definition of “beneficial function” for letter sequences so we can test your claim by your proposed process @36. Zachriel

seanpit: Why then do you act like you’ve somehow disproved my original argument regarding a need for exponentially greater amounts of time with each step up the ladder of functional complexity? What we directly falsified was “If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” seanpit: Your real issue, then, is in regard to your “Phrasentation” algorithm – an algorithm that does not show an exponential decline in evolutionary potential because the targets are pre-defined as being just a very short Levenshtein step away from the next target. Shakespeare wrote Hamlet to minimize Levenshtein distances as well as in blank verse?! What a genius! In any case, you don't like the use of Hamlet as a phrase dictionary. That's fine. Please provide an operational definition of “beneficial function” for letter sequences so we can test your claim by your proposed process @36. Zachriel

Zachriel,

The time it takes to evolve words (based on the parameters repeatedly discussed, including use of the dictionary to determine fitness), increases rapidly with word length. Indeed, it will inevitably reach a limit — because words are only so long!

Hello! This was my original claim! I originally told you that, as an illustration of the problem, an evolutionary algorithm based on single words would show an exponential increase in the time required to achieve success. And, this is exactly what your word-evolution algorithm demonstrates. Why then do you act like you’ve somehow disproved my original argument regarding a need for exponentially greater amounts of time with each step up the ladder of functional complexity? – with a complete stalling out effect before the level of 1000 saars is reached? Why do you try to act like I claimed some kind of limit to evolutionary potential at just 7-letter words? That’s a ridiculous misrepresentation of my position – especially given that your own algorithm actually support my original claim! The exponential increase that your own algorithm shows is the result of the exponential increase in the “ocean” of non-target vs. target sequences where the target sequences are not pre-defined as being just one Levenshtein step away from the next target sequence (unlike how your “Phrasentation” algorithm works). Your real issue, then, is in regard to your “Phrasentation” algorithm – an algorithm that does not show an exponential decline in evolutionary potential because the targets are pre-defined as being just a very short Levenshtein step away from the next target. This is because your algorithm here isn’t based on the Darwinian mechanism. It’s based on Dawkins-like template matching – as I’ve explained to you many times. Of course, if any match, however small, to a template sequence is defined as “selectable”, there will in fact be a linear relationship between time and the length of the sequence! However, when you base your selection on changes in the actual beneficial function of the sequence, the average time required to find higher level systems increases exponentially. seanpit

seanpit: I’ve already given you the definition of beneficial function many times in this thread. We must have missed it. Please provide it again, or provide a reference. Thanks! seanpit: I’m sorry that you were unable to get your algorithms to actually work based on sequential changes in beneficial function. It's YOUR algorithm @36. Zachriel

Zachriel,

Cytochrome b evolves too fast in the primate lineage to return an accurate tree. To reconstruct a phylogeny, you have to look at the entirety of the evidence. When you do, there is a strong congruence between morphology and molecular nested hierarchies.

As already noted in my post above, this simply isn't true. The more one considers the "entirety of the evidence" the less congruence there is between trees based on morphology vs. genetics. And, this isn't just the position of IDists - this is the position of more and more modern evolutionists who study cladistics!

Now please provide an operational definition of “beneficial function” for letter sequences so we can test your claim by your proposed process @36.

Asked and Answered - I've already given you the definition of beneficial function many times in this thread. Clearly, for anyone candidly evaluating your algorithms, your selection method is not based on a sequential functional advantage. I'm sorry that you were unable to get your algorithms to actually work based on sequential changes in beneficial function. seanpit

seanpit: Why don’t you ever respond to my questions regarding the pattern of exponential decline in evolutionary potential with each step up the ladder of functional complexity? We have. Repeatedly. The time it takes to evolve words (based on the parameters repeatedly discussed, including use of the dictionary to determine fitness), increases rapidly with word length. Indeed, it will inevitably reach a limit — because words are only so long! seanpit: Again, even you have to admit that I never said that evolution was “impossible” or even unlikely at such low levels of functional complexity. You said, “If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” Based on your statement @1, that means a 7-letter word will never evolve. seanpit: What do you think would happen once you add in the limitation of selection based on sequentially beneficial changes? You explicitly defined "beneficial" as words found in the dictionary. We proposed using a phrase dictionary, but for some reason, you rejected this. That's fine. Please provide an operational definition of “beneficial function” for letter sequences so we can test your claim by your proposed process @36. Zachriel

Here is some relevant commentary on this discussion from Douglas Axe: __________________ The reported range [for functional vs. non-functional sequences is one in 10^77 (based on data from the more complex beta-lactamase fold; ? = 153) to one in 10^53 (based on the data from the simpler chorismate mutase fold, adjusted to the same length: ? = 153). As remarkable as these figures are, particularly when interpreted as probabilities, they were not without precedent when reported. Rather, they strengthened an existing case for thinking that even very simple protein folds can place very severe constraints on sequence... Rescaling the figures to reflect a more typical chain length of 300 residues gives a prevalence range of one in 10^151 to one in 10^104... On the one hand, this range confirms the very highly many-to-one mapping of sequences to functions. The corresponding range of m values is 10^239 (=20^300/10^151) to 10^286 (=20^300/10^104), meaning that vast numbers of viable sequence possibilities exist for each protein function. But on the other hand it appears that these functional sequences are nowhere near as common as they would have to be in order for the sampling problem to be dismissed. The shortfall is itself a staggering figure—some 80 to 127 orders of magnitude (comparing the above prevalence range to the cutoff value of 1 in 5×10^23). So it appears that even when m is taken into account, protein sequences that perform particular functions are far too rare to be found by random sampling... Two possibilities for mitigating the problem need to be considered. One of these has been mentioned already. It is the possibility that the multiplicity of sequences capable of performing the requisite functions, m, might be large enough for working sequences to be found by random searches. The second possibility is that functional protein sequences might bear a relationship to one another that greatly facilitates the search. In the desert metaphor, imagine all the different gems being together in close proximity or perhaps lined up along lines of longitude and latitude. In either of these situations, or in others like them, finding the first gem would greatly facilitate finding the others because of the relationship their positions bear to one another... When structurally unrelated protein domain sequences are aligned optimally, the resulting alignment scores are very similar to the expected scores for randomized sequences with the same amino acid composition. Since random sequences produced in this way are widely scattered through sequence space, this means that dissimilar natural sequences are as well. In fact, because amino acid composition correlates with structural class, we would expect random sequences with average compositions to align somewhat better than dissimilar natural sequences do. Indeed, such wide dispersion of natural domain sequences throughout sequence space is not surprising considering the great variety of domain structures that these sequences form (Figure 6)... It therefore seems inescapable that considerable distances must be traversed through sequence space in order for new protein folds to be found. Consequently, any shortcut to success, if it exists, must work by traversing those distances more effectively rather than by shortening them. The only obvious possibility here is that new folds might be assembled by recombining sections of existing folds. If modular assembly of this kind works, it would explain how just one or two gene fusion events might produce a new protein that differs substantially from its ‘parents’ in terms of overall sequence and structure. Of course, probabilistic limitations would need to be addressed before this could be deemed a likely explanation (because precise fusion events are much less likely than point mutations), but the first question to ask is whether the assumed modularity is itself plausible... Consequently, self-contained structural modules only become a reality at the domain level, which makes them unhelpful for explaining new folds at that level... Because structural reorganization requires elements of secondary structure to be grouped spatially in new ways, it necessarily involves new binding interfaces where the exteriors of helices and/or sheets must adhere to each other in new ways. But since these interfaces consist largely of side chains, they are necessarily sequence-dependent and therefore non-generic. This is important enough to be restated: The binding interfaces by which elements of secondary structure combine to become units of tertiary structure are predominantly sequence dependent, and therefore not generic. This presents a major challenge for the idea of modular assembly of new folds, at least as a general explanation... For those elements to work as robust modules, their structures would have to be effectively context-independent, allowing them to be combined in any number of ways to form new folds. A vast number of combinations was made by random ligation of the gene segments, but a search through 10^8 variants for properties that may be indicative of folded structure ultimately failed to identify any folded proteins. After a definitive demonstration that the most promising candidates were not properly folded, the authors concluded that “the selected clones should therefore not be viewed as ‘native-like’ proteins but rather ‘molten-globule-like’”, by which they mean that secondary structure is present only transiently, flickering in and out of existence along a compact but mobile chain. This contrasts with native-like structure, where secondary structure is locked-in to form a well defined and stable tertiary fold. Their finding accords well with what we should expect in view of the above considerations. Indeed, it would be very puzzling if secondary structure were modular. In fact, although whole structural domains may be self-contained in the sense of carrying complete information for their own folding, even they may fail to meet the second criterion for structural modularity given above, simply because they do not have generic exteriors. I describe here an experimental demonstration of this that was performed years ago but not previously reported. Again it uses beta lactamases, which are an attractive model system because of the abundance of published structures and the ease of measuring their activity in vivo. This test used the two natural beta lactamases shown in Figure 9, which have highly similar backbone structures despite the fact that their sequences match at only 26% of aligned positions. Both structures consist of two domains, the larger of which was referred to previously (Figure 5B). Sections of the two genes were recombined to encode a chimeric protein that combines the domains colored green and red in Figure 9. The overall structural and functional similarity of the parent enzymes suggests that this kind of domain recombination should work. But the non-generic nature of the interface between the two domains in combination with the substantial sequence dissimilarity indicates otherwise—a point confirmed by the lack of detectable function for the chimeric construct. Douglas Axe: http://biocomplexity.org/ojs/index.php/main/article/view/BIO-C.2010.1/BIO-C.2010.1 seanpit

seanpit: Take cytochrome b for example Cytochrome b evolves too fast in the primate lineage to return an accurate tree. To reconstruct a phylogeny, you have to look at the entirety of the evidence. When you do, there is a strong congruence between morphology and molecular nested hierarchies. Now please provide an operational definition of “beneficial function” for letter sequences so we can test your claim by your proposed process @36. Zachriel

Zachriel, Why don't you ever respond to my questions regarding the pattern of exponential decline in evolutionary potential with each step up the ladder of functional complexity? - present even within your own algorithm between 1 and 7+ letter sequences? - especially highlighted when you use smaller steady-state populations and more reasonable mutation and reproductive rates? What do you think would happen once you add in the limitation of selection based on sequentially beneficial changes? Wouldn't the exponential nature of the problem only become more dramatic? Why not at least address this fundamental problem for your theory? Again, even you have to admit that I never said that evolution was “impossible” or even unlikely at such low levels of functional complexity. It is very possible and very likely at such low levels. What I said is that there would be an exponential decline in evolvability with each step up the ladder of functional complexity – which is clearly true. It is this pattern that is important to recognize. And, I made it very clear that because of this exponential decline that there would be a complete stalling out of evolutionary progress, not at the level of 7-character sequences, but at the level of 1000. seanpit

Zachriel, It shows a general congruence between morphological and molecular phylogenies. Hardly. Depending upon which genetic sequence you pick you get very very different phylogenies that often disagree with standard morphologic phylogenies. Take cytochrome b for example:

"The mitochondrial cytochrome b gene implied... an absurd phylogeny of mammals, regardless of the method of tree construction. Cats and whales fell within primates, grouping with simians (monkeys and apes) and strepsirhines (lemurs, bush-babies and lorises) to the exclusion of tarsiers. Cytochrome b is probably the most commonly sequenced gene in vertebrates, making this surprising result even more disconcerting." Michael S. Y. Lee, “Molecular phylogenies become functional,” Trends in Ecology and Evolution, Vol. 14:177-178 (1999)

Even as far back as 1998 it was known that there were serious problems within the highest branches of the "Tree of Life". Amazingly enough, a 1998 article entitled, "Molecules remodel the mammalian tree", Je Jong (in Trends in Ecology and Evolution) concluded:

"The wealth of competing morphological, as well as molecular proposals [of] the prevailing phylogenies of the mammalian orders would reduce [the mammalian tree] to an unresolved bush, the only consistent clade probably being the grouping of elephants and sea cows."

And, this major problem doesn't seem to have gotten any better over time. In 2009, Syvanen compared two thousand genes that are common to humans, frogs, sea squirts, sea urchins, fruit flies and nematodes. In theory, he should have been able to use the gene sequences to construct an evolutionary tree showing the relationships between the six animals. He failed. The problem was that different genes told contradictory evolutionary stories. This was especially true of sea-squirt genes. Conventionally, sea squirts—also known as tunicates—are lumped together with frogs, humans and other vertebrates in the phylum Chordata, but the genes were sending mixed signals. Some genes did indeed cluster within the chordates, but others indicated that tunicates should be placed with sea urchins, which aren't chordates. “Roughly 50 per cent of its genes have one evolutionary history and 50 per cent another." This led Syvanen to conclude: "We’ve just annihilated the tree of life." Likewise, Carl Woese, a pioneer of evolutionary molecular systematics, observed that these problems extend well beyond the base of the tree of life:

"Phylogenetic incongruities [conflicts] can be seen everywhere in the universal tree, from its root to the major branchings within and among the various taxa to the makeup of the primary groupings themselves."

Likewise, in 2006, biologist Lynn Margulis wrote in her article, The Phylogenetic Tree Topples:

"Many biologists claim they know for sure that random mutation (purposeless chance) is the source of inherited variation that generates new species of life and that life evolved in a single-common-trunk, dichotomously branching-phylogenetic-tree pattern! Especially dogmatic are those molecular modelers of the ‘tree of life’ who, ignorant of alternative topologies (such as webs), don’t study ancestors."

Striking admissions of troubles in reconstructing the "Tree of Life" also came from a 2006 paper in the journal PLOS Biology entitled, Bushes in the Tree of Life. The authors acknowledge that, "A large fraction of single genes produce phylogenies of poor quality," observing that one study "omitted 35% of single genes from their data matrix, because those genes produced phylogenies at odds with conventional wisdom." The paper suggests that, "Certain critical parts of the [tree of life] may be difficult to resolve, regardless of the quantity of conventional data available." The paper even contends that, "The recurring discovery of persistently unresolved clades (bushes) should force a re-evaluation of several widely held assumptions of molecular systematics." Then, Elie Dolgin, in a June, 2012 article in Nature reported that short strands of RNA called microRNAs are, "tearing apart traditional ideas about the animal family tree." Dartmouth biologist Kevin Peterson who studies miRNAs lamented, "I've looked at thousands of microRNA genes, and I can't find a single example that would support the traditional tree." According to the article, miRNAs yielded "a radically different diagram for mammals: one that aligns humans more closely with elephants than with rodents." Peterson put it bluntly: "The microRNAs are totally unambiguous ... they give a totally different tree from what everyone else wants." A 2013 paper in Trends in Genetics reported that, "The more we learn about genomes the less tree-like we find their evolutionary history to be." What is also interesting is that this information isn't entirely new - yet it is still treated by many with a great deal of surprise. Even as far back as 2000 Trish Gura argued, also in the journal Nature, that there appeared to be no consistent agreement between genetic phylogenies and those based on more traditional morphological characteristics:

"On one side stand traditionalists who have built evolutionary trees from decades of work on species' morphological characteristics. On the other lie molecular systematists, who are convinced that comparisons of DNA and other biological molecules are the best way to unravel the secrets of evolutionary history. … So can the disparities between molecular and morphological trees ever be resolved? Some proponents of the molecular approach claim there is no need. The solution, they say, is to throw out morphology, and accept their version of the truth. 'Our method provides the final conclusion about phylogeny,' claims Okada. Shared ancestry means a genetic relationship, the molecular camp argues, so it must be better to analyse DNA and the proteins it encodes, rather than morphological characters that can end up looking similar as a result of convergent evolution in unrelated groups, rather than through common descent. But morphologists respond that convergence can also happen at the molecular level, and note there is a long history of systematists making large claims based on one new form of evidence, only to be proved wrong at a later date."

For further information on this topic see: http://www.detectingdesign.com/geneticphylogeny.html In short, your popular claim that, “The nested hierarchy, including the fossil succession, remains as strong evidence of common descent regardless of any explanatory mechanism” simply doesn’t hold water – beyond the fact that such nested hierarchical patters are found all over the place within human-designed systems – especially within computer operating systems. It simply doesn’t follow that just because a NHP is identified that this automatically indicates a mindless evolutionary origin – especially when there is no viable evolutionary mechanism to explain the phenomenon in question. seanpit

seanpit: However, what the evidence actually shows is some sequence homologies of subparts within high-level systems. It shows a general congruence between morphological and molecular phylogenies. seanpit: However, if you cannot demonstrate the consistent effectiveness of your naturalistic mechanism at various levels of functional complexity, your conclusion of evolutionary ancestry simply isn’t scientific or rational. The nested hierarchy, including the fossil succession, remains as strong evidence of common descent regardless of any explanatory mechanism. seanpit: I haven’t “retreated” from my original positions at all – not at all. Then you are still wrong that you have to cross oceans of meaningless sequences to evolve seven-letter words. Now please provide an operational definition of “beneficial function” for letter sequences so we can test your claim by your proposed process @36. Zachriel

Zachriel,

The stepping stones are in the lines of descent from the common ancestor.

That’s certainly one interpretation of the evidence. However, what the evidence actually shows is some sequence homologies of subparts within high-level systems. So, what does this evidence actually mean? The standard conclusion is that these homologies must represent a common evolutionary ancestry over and above the alternative interpretation of common design. However, if you cannot demonstrate the consistent effectiveness of your naturalistic mechanism at various levels of functional complexity, your conclusion of evolutionary ancestry simply isn't scientific or rational. Given that the evidence currently in hand strongly supports the conclusion of an exponential decline in effectiveness, your automatic conclusion that homology must mean common evolutionary ancestry simply doesn’t follow. The alternate conclusion of common intelligent ancestry becomes much more likely. A very similar situation exists for computer programs that also consistently show sequence homologies within subparts of larger and more complex codes that have the same programmer. Yet, no one argues that such homologies could only have been produced by a mindless mechanism acting over time. Why then, when the very same situation is discovered within living things, that the only possible hypothesis that is allowed to be considered is a mindless Darwinian mechanism? – despite any and all evidence that such a mechanism is very very limited?

In any case, while you have retreated from your position concerning having to cross oceans of meaningless sequences to evolve 7-letter words, you still claim that longer sequences can’t be similarly crossed. You proposed a process which entails determining the “beneficial function” of a letter sequence. Please provide an operational definition of “beneficial function” so that we can test your claim.

I haven’t “retreated” from my original positions at all – not at all. Rather, you’ve misstated my original argument and made claims that I never made. The fact remains that there is an exponential decline in evolvability even at the level of sequence spaces going from meaningful 1-character sequences to 7-character sequences! That’s what I was trying to get you to see way back in 2004! I never said that evolution was “impossible” or even unlikely at such low levels of functional complexity. It is very possible and very likely at such low levels. What I said is that there would be an exponential decline in evolvability with each step up the ladder of functional complexity – which is clearly true. It is this pattern that is important to recognize. And, I made it very clear that because of this exponential decline that there would be a complete stalling out of evolutionary progress, not at the level of 7-character sequences, but at the level of 1000. What do you not understand about this argument? It’s really a very simple and straightforward argument. I’m at a loss as to why you’re still so confused? It seems like you simply don’t want to present my argument as it really is. You seem to actually want to try to distort so that it appears to be something silly and nonsensical – like a strawman version of the real thing. seanpit

seanpit: The actual landscape for protein-based systems has been examined in fair detail and very clear patterns have emerged – as higher and higher levels of functional complexity are evaluated. Your claim concerned letter sequences. seanpit: These same features also exist within the English language system That's your claim. Now provide an operational definition of "beneficial function" for letter sequences so we can test your claim by your proposed process @36. Zachriel

seanpit: It is the argument regarding the “degree” of structure that is key here.

That’s right. And the only way to answer that question is to examine the actual landscape. You claim that there is no significant structure with regards to long English texts. You proposed a process to test that claim, but you haven’t been able to provide an operational definition of “beneficial function”.

The actual landscape for protein-based systems has been examined in fair detail and very clear patterns have emerged – as higher and higher levels of functional complexity are evaluated. 1) It is clear that there is an exponential decrease in the ratio of potentially beneficial vs. non-beneficial sequences. 2) The degree of structure or non-randomness or linearity to the arrangement of these targets does not increase as you claims would require. 3) The minimum Hamming distance between potential targets increases, in a linear manner. These same features also exist within the English language system or computer programs or any other system of meaningful information/function that is based on a specific arrangement of “characters” (i.e., letters, or amino acids, or 0s and 1s, etc). This means, of course, that any mechanism like the Darwinian mechanism will experience an exponential decline in effectiveness with each step up the ladder of functional complexity. You’re algorithms are no different. Even your word evolution algorithm, as non-Darwinian as it is, also experiences an exponential increase in the time required to find longer defined sequences – without regard to a sequential increase in beneficial function. Adding this additional qualification would only enhance the exponential nature of the pattern. I’m not sure, then, why you are so convinced that the Darwinian mechanism is so clearly responsible for the origin of no many qualitatively novel high levels systems of function? Upon what is your faith based? Where is your own evidence along these lines which trumps the evidence I’ve provided here? seanpit

seanpit: Beyond the demonstrated reality that the steppingstones in protein sequence space... The stepping stones are in the lines of descent from the common ancestor. In any case, while you have retreated from your position concerning having to cross oceans of meaningless sequences to evolve 7-letter words, you still claim that longer sequences can't be similarly crossed. You proposed a process which entails determining the "beneficial function" of a letter sequence. Please provide an operational definition of "beneficial function" for letter sequences so that we can test your claim. Zachriel

Zachriel: Pointing out that there are stepping stones across a creek does not imply that the stepping stones were put there by design. Beyond the demonstrated reality that the steppingstones in protein sequence space are no longer closely spaced (or significantly linear in their arrangement) beyond very low levels of functional complexity, it is mistaken to argue that if such a situation were ever observed that it would be more consistent with a mindless cause than with an intelligent cause. If the phenomenon in question goes significantly beyond what the known powers of mindless natural processes are likely to generate, but remain within what the known powers of intelligent design are able to generate, the most rational hypothesis is that of intelligent design to explain a given phenomenon. Here are a few examples along these lines: http://cdn.wonderfulengineering.com/wp-content/uploads/2014/11/Gravity-Glue-%E2%80%93-Michael-Grab-Rock-Balancing-Art9.jpg http://cdn.earthporm.com/wp-content/uploads/2014/11/gravity-stone-balancing-michael-grab-12.jpg http://www.thisiscolossal.com/wp-content/uploads/2015/01/cover-1.jpg https://katemckinnon.files.wordpress.com/2014/02/screen-shot-2014-02-23-at-8-23-44-pm.png While it would be obvious for most people coming upon such scenes that intelligent design had been at work (Michael Grab in these cases), the question is how is such a conclusion so obvious when it comes to objects like stacked rocks or a nice path of closely-spaced steppingstones across a lake or ocean where there are no other steppingstones for vast distances all around? http://www.educatetruth.com/wp-content/uploads/2014/01/stepping-stones.jpg http://www.picturesofengland.com/img/X/1031263.jpg Without a very compelling natural explanation, why would any reasonable person conclude that such extremely rare steppingstones are all lined up, as pictured in the links above, by random chance or some unknown mindless mechanism? Yet, this is what is often being done now by many evolutionists and methodological naturalists in general. During the latest debate between Krauss, Meyer, Lamoureux the now popular standby multiple-universe or multiverse argument was forwarded by Krauss to explain the extreme fine tuning of our universe – a situation for which there is simply no compelling naturalistic reason. What’s especially interesting about the multiverse argument, as pointed out by Meyer toward the end, is that it undermines the very basis of science itself. It can be used to explain anything and everything – and therefore nothing. It is essentially identical to the “God did it” hypothesis (which is different from the God-only hypothesis or that only an intelligent designer could have likely produced the phenomenon in question) since it can be used to explain, without invoking intelligent manipulation, the stacked rocks or steppingstones pictured above – or something like Arnold Schwarzenegger winning the California lottery 10 times in a row (just happened to be in the right universe at the time). This kind of desperation to avoid admitting that intelligent design, of any kind, could have been involved in the origin and/or the diversity of life on this planet, or the fine-tuned features of the universe, is not based on science, but upon a naturalistic philosophy that strongly resembles the blind-faith religion of various fundamentalist fanatics - fundamentalists who are bound to avoid any other conclusion regardless of the evidence presented. Sean Pitman seanpit

bill cole: Where this gets foggy is when you say there are structures that guide that starts to cross over to the design inference. Pointing out that there are stepping stones across a creek does not imply that the stepping stones were put there by design. Zachriel

Sean Thank you very much for the explanation. I think that progress in understanding is going on here which is great. Zachriel I see you understand the sequential space problem which is a good start. You are proposing solutions which is great. Where this gets foggy is when you say there are structures that guide that starts to cross over to the design inference. We know that intelligence evolved on earth the question is when and how. How was the incredible capability of cells inserted into or generated by cells that control the most precise nano manufacturing capability in the world:cell metabolism and the cell cycle. I am ok any way the data takes us. bill cole

seanpit: It is the argument regarding the “degree” of structure that is key here. That's right. And the only way to answer that question is to examine the actual landscape. You claim that there is no significant structure with regards to long English texts. You proposed a process to test that claim, but you haven't been able to provide an operational definition of "beneficial function". Zachriel

Bill Cole, Zachriel is arguing that because there is some structure to the location of beneficial sequences at very low levels of sequence space (i.e., their location isn't entirely random relative to each other), that such non-random patterns exist in all levels of sequence space - to the same degree. It is the argument regarding the “degree” of structure that is key here. As best as I've been able to figure, evolutionists, like Zachriel and many others, consistently appear to argue that the degree of non-randomness actually increases with increasing levels of sequence space – to compensate for the exponential decrease in beneficial vs. non-beneficial options. The overall effect, according to those like Zachriel, seems to be an essentially linear increase in the average time required to achieve success with each step up the ladder of functional complexity – not an exponential increase. What’s the evidence to support this notion? – according to them? Well, it has to do with sequence homologies… The fact is that there is a certain degree of non-randomness to functional sequences at all levels of functional complexity. Larger protein-based systems, like sentences in the English language system, are usually comprised of fairly common subsystems and subdomains. The argument, then, is that more complex systems, like the hemoglobin molecule for example, can easily be evolved in a reasonable amount of time by simply linking up these smaller subsystems or subsequences that already exist as parts of other systems of function. The similarities, or homologies, between these subsystems are used as evidence of common evolutionary ancestry. As an even more striking example, consider the multipart flagellar motility system. This system of around 40 different specifically arranged structural protein parts supposedly evolved by simply linking up the individual pre-existing proteins together to produce each steppingstone in the pathway toward full-blown flagellar motility. After all, every single protein in the flagellar motility system, except for one, shares non-random homologies with some other protein in the gene pool that is part of a different system of function. It seems intuitively obvious then that these homologies strongly suggest common evolutionary ancestry. Never mind that such homologies also exist in systems created by intelligent design – like computer codes or even the works of Shakespeare. So, how does one tell if a given homology is the result of deliberate design (as in conservation of design) or the result of non-designed evolutionary ancestry? Well, it all depends if the homologies are homologous enough to cross the gaps in sequence space in a reasonable amount of time without the need to invoke intelligent design. For lower levels of functional complexity, requiring fewer than a few hundred specifically arranged characters, the homologies are significant enough so that the hypothesis of intelligent design need not be invoked. However, the problem for the evolutionary perspective is that these homologies are not homologous enough beyond these very low levels. Beyond the level of 1000 specifically arranged characters, the needed homologies simply aren’t there for a successful “swap” to be realized with a single mutation – or even several dozen mutations of any kind. Why not? Contrary to the claims of Zachriel, evolution at these higher levels would require numerous additional modification mutations to produce a successful concatenation mutation that is functionally beneficial to a selectable degree – because the minimum Hamming gap distance are simply far too large at these levels to maintain all of the required subsystems within the genome that would be needed to realize a swap mutation that would be successful without any additional modifying mutations. That is why, in short, these evolutionary scenarios for how evolution must have produce these high-level systems are nothing more than just-so stories. They are statistically untenable and they never happen in real life – not beyond the level of 1000 specifically arranged amino acid residues. There’s not a single observable example of evolution in action at this level described in literature. Not a single one of the flagellar steppingstones have ever been crossed in real life. There’s just nothing supporting such notions beyond wishful thinking and a vivid imagination. In short, then, the science of detecting the activity of intelligent design (which is used all the time in mainstream sciences – such as forensic science, anthropology, and even SETI) is based on the idea that the phenomenon in question can only be reasonably explained by invoking intelligent design. For a detailed description of this problem, specifically dealing with the claims of Nick Matzke regarding flagellar evolution, see the following links: http://www.detectingdesign.com/flagellum.html http://www.detectingdesign.com/NickMatzke.html http://www.detectingdesign.com/JasonRosenhouse.html seanpit

bill cole: Are you saying there is a cellular mechanism that can narrow the sequential space of the genome? The distribution of words and phrases in sequence space is not random, but highly structured. By analogy, it shows how intuitive notions concerning high-dimension spaces are not necessarily accurate. bill cole: This would support James Shapiro’s theory of natural genetic engineering. It just shows that organic molecules exist in a highly structured universe. So, for instance, small changes in amino acid sequence often result in small changes in the three-dimensional structure of the protein, meaning that there are selectable pathways to increased specificity. Zachriel

Zachriel

He suggested wordspace as a proxy. From our experience, the space is highly structured.

I don't understand this point. Are you saying there is a cellular mechanism that can narrow the sequential space of the genome? If so can you support with evidence of this mechanism? This would support James Shapiro's theory of natural genetic engineering. bill cole

bill cole: I understand this is your opinion but I think that Sean is right here. Intuition is a valuable resource, but can often mislead us. bill cole: I understand in certain cases an adaptive change may take a few changes but at some point you need very different function and now Sean’s ocean awaits you and that ocean is longer then our universe. That's the claim. He suggested wordspace as a proxy. From our experience, the space is highly structured. He provided a process to test this proposition @36, but that entails selection for beneficial function, which is something for which hasn't been able to provide an operational definition. bill cole: Try to imagine designing the DNA sequences that produce hemoglobin. Wouldn't know how to design hemoglobin. However, hemoglobin is composed of a number of protein subunits called globins, which form a family congruent with the phylogenetics of organisms, having diverged from a common ancestor. https://www3.nd.edu/~aseriann/CHAP7B.html/img028.gif Hemoglobin is an example of how a large protein can evolve by concatenating smaller proteins. 4^umpteenth means very little when discussing such a process. http://antranik.org/wp-content/uploads/2011/12/hemoglobin-molecular-structure-alpha-beta-globin-chain-with-heme.jpg Zachriel

Zachriel

Sure. However, Sean Pitman’s claim about wordspace was that you had to cross oceans of meaningless sequences, which was not correct.

I understand this is your opinion but I think that Sean is right here. I understand in certain cases an adaptive change may take a few changes but at some point you need very different function and now Sean's ocean awaits you and that ocean is longer then our universe. Sequential space is essentially infinite when sequences get over 100 aa long. The mechanism you are trying to support says you can mutate your way through infinity to find advantage. The other point is that protein sequences are very sophisticated in what they do, Try to imagine designing the DNA sequences that produce hemoglobin. How would you go about this? Ok what are the baby steps to get to hemoglobin. Try to imagine this. bill cole

seanpit: The question is, why is a particular “word” or “phrase” more functionally beneficial compared to what came before? – not compared to a random sequence. Longer words were apparently more "beneficial" in your original statement. They are generally more complex, at the very least in terms of syllables, and often encapsulate multiple concepts. Longer snips of Shakespeare might reasonably be considered more "beneficial" then shorter snips of Shakespeare. But we're happy to defer to your own operational definition of "beneficial" with regards to phrases — once you provide one. seanpit: I’m sorry you misunderstood what I was saying. That's fine. As we have shown, you don't have to cross oceans of meaningless words to find seven-letter words, or ten-letter words, or twelve-letter words. Glad that's settled. seanpit: Not beyond very low levels of functional complexity. There are examples of increasing specificity, such as in Lenski's experiment which involve multiple moving parts. seanpit: Also, your algorithm does not allow for random walks – despite the fact that random walks are quite common in real life. We'd be happy to include random walks, but it wasn't material to your original proposed process. Now, provide an operational definition of “beneficial function” for letter sequences per your proposed process @36. seanpit: I explained in some detail why “nonhomologous” swapping of stable sequences would work for a while after individual point mutations no longer worked. Such swapping of larger sequences just moves the problem up a level is all YOU cited the paper which said “More generally, our simulations demonstrate that the efficient search of large regions of protein space requires a hierarchy of genetic events, each encoding higher-order structural substitutions. We show how the complex protein function landscape can be navigated with these moves.” In any case, in all that you posted, we didn’t see an operational definition of “beneficial function” for letter sequences. Zachriel

Zachriel,

As word evolution is based on a word dictionary, per his original proposal, it would seemingly be reasonable to use a phrase dictionary for multi-word sequences. Indeed, a word is reasonably considered more “functional” than a random sequence of letters, and a sequence of words from Shakespeare would seemingly qualify as more “functional” than a random sequence of words. Sean Pitman rejected this proposal, but has been unable to offer any other way to directly test his claim. He’s left waving in the general direction of big numbers. It’s obvious you see.

The question is, why is a particular “word” or “phrase” more functionally beneficial compared to what came before? – not compared to a random sequence. Remember, for the Darwinian mechanism to work, the steppingstone sequence must be made up of sequences that each show improved beneficial function compared to the previous steppingstone in the sequence. Neither your word or phrase evolution programs make selections in this way. Your Phrasenation program, in particular, shows selected partial sentences and phrases that are clearly nonsensical, much less sequentially beneficial. In short, I was trying to get you to think about the nature of sequence space and the exponential decay of potential targets within that space. While I’m sorry that your programs do not truly reflect the limitations of the Darwinian mechanism, the evidence regarding the exponential decline of evolutionary potential with each step up the ladder of functional complexity is overwhelming.

Sean Pitman’s claim about wordspace was that you had to cross oceans of meaningless sequences, which was not correct.

Again, I’m sorry you misunderstood what I was saying. Why on Earth do you think that I always cited the limit of 1000 saars if I thought that evolutionary progress would actually stall out within sequence spaces of less than 10 characters? How does make any sense to you given your description of my position? Of course evolution works within such relatively small spaces – despite the exponentially increasing amounts of time required even at such low levels. It is this pattern of exponentially increasing amounts of time that is of primary importance here. I’m still mystified as to why you don’t see this pattern as relevant?

An increase in specificity is something that can evolve, as can be easily shown.

Not beyond very low levels of functional complexity. When you’re talking about systems of function that require a minimum of more than 1000 specifically arranged characters, such systems are so far apart in sequence space that getting from one to any other would require trillions upon trillions of years. And, natural selection cannot come to the rescue here because natural selection cannot work at all until the next beneficial sequence is first discovered. You think that such large gap distances can be crossed by single recombination mutations. As I’ve explained, at the level of 1000 saars or above, the odds that anything already exists within a given gene pool which could be successfully recombined with another sequence to produce a qualitatively novel functional system are essentially nil. It just doesn’t happen and statistically it is extremely unlikely to happen.

The website doesn’t say that, but refers to phrases. However, based on your statement about crossing oceans, your claim is that selectable transitions do not exist even for 7-letter words. That means they will *never* evolve, per the algorithm. Selection is absolute, per your original statement “Start with a short 2 or 3-letter word and see how many words you can evolve that require greater and greater minimum sequence requirements.” If the sequence doesn’t form a valid word, it never enters the population.

Again, I never said that “transitions” or the possibility for successful random mutations of various kinds do not exist for 7-character sequences. I’ve consistently explained that they do exist at such low levels – which is somewhat unusual for Intelligent Design proponents to argue. Most IDist argue that there are no examples at all of functional evolution. This isn’t true. There are many examples of true evolution in action. It is just that all of these examples are at very low levels of functional complexity and show an exponential stalling effect with each step up the ladder of functional complexity. So, clearly, your description of my position is a mischaracterization of my true position – a strawman. By the way, you do realize what a translocation mutation is? – right? It is a large jump across “oceans” of sequence space. Sure, the “oceans” at such low levels are very small relatively speaking, but the idea is the same. At such very low levels the odds of taking a large successful leap across sequence spaces aren’t too bad – especially for a larger population undergoing extremely high reproductive and mutation rates. Also, your algorithm does not allow for random walks – despite the fact that random walks are quite common in real life. Such random walks would also have fairly good success at these low levels of functional complexity – as I’ve explained many times already in this thread.

seanpit: Your mechanism appears to be largely based on random sampling of sequence space that heavily favors locations very close to the starting point location.

Yes. It’s called evolution.

Indeed. And, despite your extremely generous mutation and reproductive rates, such “evolution” shows an exponential increase in required time with each step up the ladder of functional complexity. Even your own word-evolution algorithm, which isn’t based on sequentially increasing beneficial function, shows this non-linear increase in average time with each increase in the minimum size requirement.

Immediately followed by “nonhomologous DNA ‘swapping’ of low-energy structures is a key step in searching protein space”. Word evolution involves non-homologous swapping. In any case, in all that you posted, we didn’t see an operational definition of “beneficial function” for letter sequences.

Of course! As I previously explained in some detail, “nonhomologous” swapping of stable sequences will work for a while after individual point mutations no longer work. Such swapping of larger sequences just moves the problem up a level is all – like using whole words instead of individual letters for the “alphabet” of options for various positions within a sequence. What is interesting here is that point mutations work very well at very low levels of functional complexity – like evolving short individual words of less than 5 or 6 letters in length. However, very quickly the ability of point mutations to find new targets drops off, exponentially in fact, with additional steps up the ladder of functional complexity. At this point, as previously explained, the only way forward to cross the growing Hamming gap distances and more and more uniform distribution of functionally beneficial islands is to resort to recombination or “swap” mutations involving larger pre-existing sequences within the gene pool of options. While this does help for a little while, the success of these swap mutations also starts to drop off, exponentially, with each step up the ladder of functional complexity. Why? Because, in short, gene pools are limited. They can only store a limited number of sequences. As the minimum Hamming gap distance continues to increase, the odds that what is needed to undergo a successful “swap” decrease in an exponential manner. That is why the average time to achieve success, even with swap mutations, continues to increase in an exponential manner until, beyond the level of 1000 saars, trillions of years of time isn’t enough. seanpit

bill cole: It is the current evolutionary claim that life’s diversity is created by partial stochastic mechanisms. Sure. However, Sean Pitman's claim about wordspace was that you had to cross oceans of meaningless sequences, which was not correct. bill cole: Sean and I are skeptical of this because of exponential growth of the sequential space of proteins. Skepticism is fine, however, it doesn't substitute for evidence. In the case of the wordscape, given a operational definition of "beneficial", we should be able to test your intuition. bill cole: We also know that the chance of proteins folding to function varies depending of function and nuclear proteins that work together with other proteins are highly sequence specific. An increase in specificity is something that can evolve, as can be easily shown. seanpit: On your website you’re specifically claiming that I said that evolution was impossible, would take “zillions of years”, at the level of 7-character sequence space. The website doesn't say that, but refers to phrases. However, based on your statement about crossing oceans, your claim is that selectable transitions do not exist even for 7-letter words. That means they will *never* evolve, per the algorithm. seanpit: Beyond this, are you telling me that even at 7-character sequence space your random search algorithm never makes a wrong choice? Selection is absolute, per your original statement "Start with a short 2 or 3-letter word and see how many words you can evolve that require greater and greater minimum sequence requirements." If the sequence doesn't form a valid word, it never enters the population. seanpit: Your mechanism appears to be largely based on random sampling of sequence space that heavily favors locations very close to the starting point location. Yes. It's called evolution. seanpit: Very quickly your mechanism will stall out – at very low levels of functional complexity. That's your claim, but something you haven't been able to show. seanpit: "We demonstrate further that even the DNA shuffling approach is incapable of evolving substantially new protein folds. " Immediately followed by "nonhomologous DNA 'swapping' of low-energy structures is a key step in searching protein space". Word evolution involves non-homologous swapping. In any case, in all that you posted, we didn't see an operational definition of “beneficial function” for letter sequences. Zachriel

Zachriel,

seanpit: Why then lie about what I actually said?

What you said was “If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” This statement is demonstrably false.

First off, that’s not what you’re claiming on your website. On your website you’re specifically claiming that I said that evolution was impossible, would take “zillions of years”, at the level of 7-character sequence space. That’s simply not true and you know it. Why else do you suppose I’ve always drawn the line at 1000, not 7, specifically arranged characters? Why the need to lie about where I’ve always drawn the line for the limits of evolutionary potential? Beyond this, are you telling me that even at 7-character sequence space your random search algorithm never makes a wrong choice? If you set up the parameters like I described above, your own algorithm would indeed take exponentially greater amounts of time to find targets at this very low level. Even your own massive reproductive and mutation rates take greater amounts of time at the 7-character level as compared to lower levels that represents a non-linear increase in time...

seanpit: Isn’t your own algorithm is based on “random sampling”

Calling evolution random sampling is an equivocation. If you mean evolution, then use the term evolution, which we have defined a population undergoing random mutation and random recombination and selection.

I’m speaking specifically about the part of the Darwinian mechanism that takes place before selection takes place. Random mutations can come in the form of random sampling of sequence space or random walks. Your mechanism appears to be largely based on random sampling of sequence space that heavily favors locations very close to the starting point location. While this will indeed help your algorithms find closely-spaced steppingstones more readily, it will not help you as the minimum distance between the starting point and the next closest steppingstone increases in a linear manner with each step up the ladder of functional complexity. Very quickly your mechanism will stall out – at very low levels of functional complexity. Your “Phrasenation” algorithm isn’t based on sequentially increasing beneficial function, but on template matching to any portion of a pre-established template – without any regard to the actual function or meaning of the evolving sequence. Therefore, it fails to qualify as a valid example of “evolution” – at least in the Darwinian sense of the term. It is very much in line with Dawkins’ “Methinks it is like a weasel” algorithm – identical in fact.

seanpit: The fact is, there simply is no significant statistical difference for success between random sampling and random walks

A random walk or a random sampling of the entire space would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials.

And, I’ve already explained to you why this is: 1) Your “trials” are based on an extraordinarily high reproductive rate and mutation rate. 2) Your population size is very high considering the level of functional complexity under consideration. 3) Targets within sequence space are more clustered at these very low levels with more very short Hamming distances of 1 much much more likely at these very low levels. 4) Your own algorithm (as limited as it is) demonstrates a non-linear, even exponential, increase in the amount of time required to find targets at higher and higher levels. 5) Higher and higher level systems, beyond these extremely low levels of functional complexity, continue to show that exponentially more and more amounts of time are required until evolutionary progress completely stalls out, this side of trillions of years of time, shy of the level of 1000 saars. 6) Higher and higher levels of functional complexity continue to show a more and more randomly uniform distribution of functionally beneficial systems – or even stable proteins. 7) It’s an undeniable fact that these uniformly distributed targets (even when you’re only talking about stable proteins) are exponentially reduced in relative numbers with each step up the ladder of functional complexity.

seanpit: If you’re biasing your “random sampling” to positions located very short distances from your starting point, of course this would help find closely-spaced targets.

It’s called evolution.

Yes - based on random sampling with an emphasis on finding targets with a very small Hamming distance of 1 from the starting point – which only has any remote hope of success at very very low levels of functional complexity (even given extremely high reproductive and mutations rates that you use in your algorithm). While true Darwinian evolution of protein-based systems in real living things isn’t quite as heavily dependent on this particular type of random mutation (or the extremely high reproductive/mutation rates you use), the same basic problem is realized in both cases – an exponential increase in the average time required to take the next step up the ladder of functional complexity.

Your claim concerned words. (The study you cite concerns sequences that share the same fold, comparable to a study of word synonyms, not the distribution of enzymes in general, or the distribution of enzymes in nature.) In any case, we’re still waiting for a operational definition of “beneficial function” for letter sequences.

The same situation is true for all systems of meaningful information based on a specific sequence of characters. It doesn’t matter if the sequences produce the same qualitative function, the sequences that produce these functions are essentially randomly distributed in sequence space – without significant clustering. The same is true for all other types of functional sequences as well - with more and more prominence at higher and higher levels. Additional information along these lines include the finding that protein families are separated from each other by gaps of non-foldable / non-functional sequences. This is true even for very small sequence spaces comprised of only 16 binary characters in the sequence simulating certain features of protein folding: “This produces a frustration barrier, e.g., a region of frustrated sequences between each pair of minimally frustrated families. Any stepwise mutational path between one minimally frustrated sequence family and another must then visit a region of slow or nonfolding sequences… In the case of real proteins, the sequences in these high frustration regions are much less likely to meet physiological requirements on foldability (of course, real physiological requirements can be much more extensive than this). If the sequences in these regions do not meet the physiological criteria, then they cannot participate in biochemical processes, which means that they will be physiologically excluded. If the requirement is sufficient, the region between two families will be completely excluded, which cuts sequence space into separate fast-folding, stable parts. This provides a mechanism for partitioning protein sequence information into evolutionarily stable, biochemically useful (foldable) subsets… Thus, because p(x) is the best path, a gap will occur, completely separating the sequence families… A spontaneous double or triple exchange mutation is required to mutate across the gap.” http://www.pnas.org/content/95/18/10682.full Such non-beneficial “gaps” in sequence space become more and more prominent at higher and higher levels of functional complexity. With each step up the ladder, the minimum Hamming “gap distances” between potentially beneficial islands grow in a linear manner. This is why such gaps become problematic by the time the level of 100aa sequences are considered. At this point “point mutation alone is incapable of evolving systems with substantially new protein folds. We demonstrate further that even the DNA shuffling approach is incapable of evolving substantially new protein folds.” http://www.pnas.org/content/96/6/2591.full The authors in this particular article go on to argue that, because of the lack of usefulness of point mutations at the level of 100aa systems, evolution, at this point, became almost entirely dependent upon “nonhomologous DNA ‘swapping’ of low-energy structures.” In other words, if carefully selected stable protein folds are “swapped” at the 100aa level, novel stable proteins may be discovered with some rare beneficial proteins realized. That is why, usually, when the successful evolution of qualitatively novel protein-based systems is realized at the level of 100aa systems, it isn’t based on point mutations, but on multi-character indel mutations that consist of stable protein folds. It becomes quite clear, then, that by the time the level of 100-character sequence space is being searched, point mutations become pretty much pointless when it comes to evolutionary progress – because of linearly expanding Hamming gap distances within such higher level spaces. Of course, as one keeps moving up the ladder of functional complexity these minimum Hamming gap distances keep increasing in a linear manner. Fairly quickly these gap distances become so large that even the swapping of stable protein folds cannot cross the distance in a single bound. At this point, multiple specific swaps must be realized to achieve success. This means, of course, that the average time required increases exponentially so that only rarely do we see real-time examples of evolutionary progress at the level of 200 or 300 saars. By the time the level of 1000 saars is reached (usually involving multiple specifically arranged proteins within a system), finding a qualitatively novel system requires numerous specific “swaps” of stable protein sequences each consisting of dozens of specifically arranged amino acid residues. At this point, the statistical odds against success get so large that trillions upon trillions of years are required to overcome these odds, and “evolution” completely stalls out. And, that, in short, is why I’ve always drawn the line of the limit to evolutionary potential at 1000 saars (not 7). In any case, I grow tired of your dishonest and very repetitive strawmen misrepresentations. If you have nothing substantive to add to this discussion, beyond your very dishonest claim that I somehow said that evolution couldn't possibly work within the extremely low level of 7-character sequence space, I don't see the point of continuing to go round and round here... seanpit

Zachriel

As word evolution is based on a word dictionary, per his original proposal, it would seemingly be reasonable to use a phrase dictionary for multi-word sequences. Indeed, a word is reasonably considered more “functional” than a random sequence of letters, and a sequence of words from Shakespeare would seemingly qualify as more “functional” than a random sequence of words. Sean Pitman rejected this proposal, but has been unable to offer any other way to directly test his claim. He’s left waving in the general direction of big numbers. It’s obvious you see.

It is the current evolutionary claim that life's diversity is created by partial stochastic mechanisms. Sean and I are skeptical of this because of exponential growth of the sequential space of proteins. We also know that the chance of proteins folding to function varies depending of function and nuclear proteins that work together with other proteins are highly sequence specific. The bottom line is the evolutionary theory does not have a mechanism that passes the sniff test. How in the world can a stochastic process overcome the almost infinite statistical space of a long sequence? bill cole

seanpit: Beyond the fact that this isn’t true (a randomly located target within 7-character sequence space could quickly be discovered by the Darwinian mechanism in an evolving colony of organisms) It's your claim that you have to cross an ocean of meaningless sequences to find a seven-letter word that is false. seanpit: Why then lie about what I actually said? What you said was “If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” This statement is demonstrably false. seanpit: Isn’t your own algorithm is based on “random sampling” Calling evolution random sampling is an equivocation. If you mean evolution, then use the term evolution, which we have defined a population undergoing random mutation and random recombination and selection. seanpit: The fact is, there simply is no significant statistical difference for success between random sampling and random walks A random walk or a random sampling of the entire space would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials. seanpit: If you’re biasing your “random sampling” to positions located very short distances from your starting point, of course this would help find closely-spaced targets. It's called evolution. seanpit: For short RNA molecules ... For short proteins Your claim concerned words. (The study you cite concerns sequences that share the same fold, comparable to a study of word synonyms, not the distribution of enzymes in general, or the distribution of enzymes in nature.) In any case, we're still waiting for a operational definition of "beneficial function" for letter sequences. Zachriel

Zachriel,

seanpit: I never said that evolution at the level of 7-character sequences would take “zillions of years”

Actually, if you have to cross oceans of meaningless sequences, then you can *never* evolve 7-letter words stepwise.

Beyond the fact that this isn’t true (a randomly located target within 7-character sequence space could quickly be discovered by random mutations in a colony of organisms), I never said that evolution at the level of 7-letter sequence was impossible – just the opposite in fact. You've known this for a long time - before you created your website in fact. Why then lie about what I actually said? Why not say that I actually drew the limit to evolutionary progress at 1000 saars? – not at 7? Why not just present what I actually said? That evolution works at low levels, like 7-character sequences and the like, but then experiences and exponential decline until it completely stalls out at the level of functional/meaningful systems that require a minimum of at least 1000 specifically arranged characters? What advantage does it give you to lie and make it appear like I said something very different about the limit to evolutionary progress? You don't think it makes you look rather desperate?

seanpit: The odds of success are essentially the same

That’s demonstrably false. As noted above a random sampling would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials. That’s not “essentially the same”.

Isn't your own algorithm is based on “random sampling” Zachriel?! The difference is in the location of targets within sequence space - not the method of random sampling - right? The fact is, there simply is no significant statistical difference for success between random sampling and random walks, or any other random search algorithm, given a random location of a rare target.

Equivocation. Random sampling above referred to random sampling of the entire space, as contrasted with evolution and a random walk.

Huh? What do you mean by “evolution” here? How does "evolution" take place until either random walk or random sampling succeeds?! Then, and only then, will there be evolution in the sense that the seletable target has been discovered - right?! Also, how are your algorithms not undergoing “random sampling of the entire space”? If you’re biasing your “random sampling” to positions located very short distances from your starting point, of course this would help find closely-spaced targets. However, as the minimum target distance increases such a random search method will not help you. There will still be an exponential stalling effect with each step up the ladder of functional complexity – as your own algorithms demonstrate!

Which reference? [regarding random distributions of proteins etc.]

For short RNA molecules: “The sequences folding into a common structure are distributed randomly in sequence space. No clustering is visible.” http://www.sciencedirect.com/science/article/pii/S1359027897000370 For short proteins: “The data do not indicate a significant amount of clustering… We found essentially no homology between the inverse folded sequences and no discernible clustering. The distribution of amino acids in these sequences is essentially random.” http://www.sciencedirect.com/science/article/pii/S1359027897000370

If the sequences approach randomness, then they are not compressible. We know English is compressible, and we know that there are only a few tens-of-thousands of valid sequences that can be found between spaces.

You don’t get it. While protein sequences, like meaningful English sequences, are always compressible to some degree or another (because of the specific features I just described for you in my previous post), the degree of compressibility decreases with increasing minimum size and/or specificity requirements . . . quickly producing an fairly uniformly random appearance in the distribution (i.e., not really very predictable or significantly “compressible” beyond a certain point. And, the degree of compressibility is reduced with each increase in the minimum size and/or specificity requirement of the meaningful/beneficial sequence under consideration. After all, meaningful sequences, while not entirely random are also not entirely predictable either. In other words, they cannot be compressed as a truly predictable sequence can be compressed – like the infinite number Pi for example. This means that they are located between purely randomly generated sequences and those that are highly predictable or non-random. Meaningful/functional sequences therefore increase the appearance of their random location within sequence spaces with each step up the ladder of functional complexity. seanpit

seanpit: Zachriel fails to recognize the pattern of declining evolutionary potential – a pattern that is non-linear, exponential in fact, with each level up the ladder of functional complexity. We're willing to keep an open mind. What we have shown is that, contrary to your claim, you don't have to cross an ocean of meaningless sequences to evolve seven-letter or even longer words. We know that it is the structure of the landscape that determines whether evolutionary search will be effective. Having tested the wordscape extensively, your larger claim appears to false. However, it's up to you to provide a rigorous definition of "beneficial function" per your own proposed process @36, "Select based on changes in beneficial function". Zachriel

bill cole: As the sequence increases in length 10^5 becomes just as big of a problem as 10^10. Well, that's one of Sean Pitman's claims with regards to word evolution, but he can't provide a rigorous definition of "beneficial function" in order to test his claim. As word evolution is based on a word dictionary, per his original proposal, it would seemingly be reasonable to use a phrase dictionary for multi-word sequences. Indeed, a word is reasonably considered more "functional" than a random sequence of letters, and a sequence of words from Shakespeare would seemingly qualify as more "functional" than a random sequence of words. Sean Pitman rejected this proposal, but has been unable to offer any other way to directly test his claim. He's left waving in the general direction of big numbers. It's obvious you see. Now, you are making the claim, but have as little evidence as he does — other than waving in the general direction of big numbers. It's obvious you see. seanpit: I never said that evolution at the level of 7-character sequences would take “zillions of years" Actually, if you have to cross oceans of meaningless sequences, then you can *never* evolve 7-letter words stepwise. seanpit: The odds of success are essentially the same That's demonstrably false. As noted above a random sampling would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials. That's not "essentially the same". seanpit: How are your algorithms not based on random sampling of the surrounding search space? Equivocation. Random sampling above referred to random sampling of the entire space, as contrasted with evolution and a random walk. seanpit: I’ve already given you the relevant references regarding the more and more randomly uniform nature of potential targets within sequence space numerous times. Which reference? seanpit: It’s the overall pattern that’s important here – not the demonstration of some comprehensibility by itself. If the sequences approach randomness, then they are not compressible. We know English is compressible, and we know that there are only a few tens-of-thousands of valid sequences that can be found between spaces. Zachriel

Bill Cole,

I think you are doing interesting work here but unfortunately not really validating current evolutionary mechanisms as viable including RMNS and neutral theory. As the sequence increases in length 10^5 becomes just as big of a problem as 10^10. If you jump out of the 50th floor of a building the results are generally the same as jumping out of the 100th floor.

Exactly! Zachriel fails to recognize the pattern of declining evolutionary potential - a pattern that is non-linear, exponential in fact, with each level up the ladder of functional complexity. He thinks that because there is a bit of non-randomness to the distribution of lower-level target sequences that this remains true for higher and higher levels of functional complexity - to the same degree. This clearly isn't true. The degree of "structure" to the location of potentially beneficial sequences at high levels of functional complexity would have to be truly amazing indeed, even designed, if evolutionary progress were to be tenable at these high levels. The problem, of course, is that the needed degree of structure simply isn't there - not remotely. It's not even there at the relatively low level of just 1000 specifically arranged characters... seanpit

Zachriel,

What you said was that If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” But you don’t have to swim through meaningless sequences to evolve 7-letter words. Your claim is false.

Yet again, I never said that evolution at the level of 7-character sequences would take “zillions of years” – which is what you claim on your website. That's a bold-faced lie - and you know it. As you know, I’ve always said that evolution completely stalls out, this side of a practical eternity of time, at the level of 1000 specifically arranged characters – not 7. You clearly know this - since 2004. So why do you claim something I never said on your website? Why misrepresent me like this? Why the need to lie and paint a false picture of my actual position? Too hard to deal with a limit of 1000 specifically arranged characters? Why not just be honest about what I'm really saying? Beyond this, yet again, even in your own algorithm (with the use of a sizable population and an enormous reproductive rate and mutation rate) searches numerous 7-character sequences before it finds defined “words” – and we’re not even talking about beneficial function here. Yet again, just because a potential target is just 1 mutation away from the starting position in hyperdimensional sequence space does not mean that a random walker (especially a single random walker) doesn’t have to swim through quite a few non-target sequences. Your argument that “random sampling” somehow avoids this problem is nonsense. The odds of success are essentially the same and get exponentially worse with each step up the ladder of functional complexity.

That is false. There are many selectable pathways in the region of shorter (ten or fewer letter) words. There are stepping stones!

Again, while it’s true that at very low levels of functional complexity (requiring fewer than 10 specifically arranged characters) that the odds are very good that the distance between a given starting point within a large population and the next closest potential target will be just one mutation, that doesn’t mean that these closely-spaced steppingstone are easy to find in higher dimensional space as compared with 2 or 3-character sequences – resulting in a non-linear increase in the average number of random walk mutations or random sampling mutations from a given starting point. And, with each step up the ladder of functional complexity these closely-spaced steppingstones become exponentially less and less common and therefore much more difficult to find within a given span of time for a steady-state population. And, as you move farther and farther up the ladder of functional complexity, the odds that a single mutation of any kind finding a higher level beneficial system drop off exponentially as well until it is essentially impossible that such a situation exists – well before you reach the level of 1000 saars.

seanpit: Random walks or even random sampling, if you prefer, will be successful, very quickly in fact at such low levels (given a reasonably-sized population).

Random walks and random sampling are much slower than evolution in the region of shorter (ten or fewer letter) words. A random sampling would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials.

What? How are your algorithms not based on random sampling of the surrounding search space? What is your definition of “evolution”? The evolutionary mechanism in particular? After all, according to most evolutionists, “evolution” must be based on a random search algorithm. That’s how the Darwinian mechanism works – via random sampling or random walks into the surrounding search space. It isn’t until a target beneficial sequence is actually discovered that natural selection comes into play. Beyond this, even your own algorithms “accomplish the task” in far less than 1e5 trials when you’re talking 2 or 3- character sequences. Even as they currently stand, your algorithms take exponentially longer to evolve longer words. And, your 10-letter words would take significantly longer, exponentially so, if you reduced your population size to 2 or 3 and your reproductive rate to 2 or 3 per individual per generation and your mutation rate to one mutation per individual per generation. Why the non-linear increase in the success of your own algorithms? – even starting with a fairly large population, very high reproductive rates, and very high mutation rates?

seanpit: Beyond this, your assumption that meaningful/beneficial sequences will remain significantly clustered beyond the lowest levels of functional complexity is demonstrably false.

Then demonstrate it.

I’ve already given you the relevant references regarding the more and more randomly uniform nature of potential targets within sequence space numerous times. It’s not like this is some kind of secret.

Even simple statistical tests show that texts of English are not evenly distributed in sequence space. If so, English texts would not be compressible, when, in fact, they are highly compressible.

As I’ve explained before, protein-based systems are also compressible – but become less and less so with each step up the ladder of functional complexity. Again, the observation that beneficial protein-based systems (or even stable systems in general) take on a more and more randomly uniform distribution within higher and higher levels of sequence space has been published many times. You see, while words are often based on similar smaller clusters of “letters”, and while phrases are often based on similar underlying “words”, and while sentences are often based on similar “phrases”, such similarities start to break down, more and more with each additional increase in the minimum size and/or specifically of a beneficial higher-level sequence. It's the overall pattern that's important here - not the demonstration of some comprehensibility by itself. What is the pattern of comprehensibility at various levels of functional complexity? The very same thing happens to protein-based systems. This is why non-beneficial gaps start to grow, in a linear manner, between potentially beneficial sequences with each step up the ladder of functional complexity. And, of course, with each linear increase in the minimum non-beneficial gap distance, the average time for a random search algorithm to achieve success increases exponentially. seanpit

Z

. As noted above a random sampling would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials.

interesting, and I think this is a good simulation of certain proteins. I think you are doing interesting work here but unfortunately not really validating current evolutionary mechanisms as viable including RMNS and neutral theory. As the sequence increases in length 10^5 becomes just as big of a problem as 10^10. If you jump out of the 50th floor of a building the results are generally the same as jumping out of the 100th floor. bill cole

bill cole: Can you explain how you came up with this hypothesis? The contrary hypothesis was by Sean Pitman, that evolving seven-letter words would require crossing oceans of meaningless sequences. The test of the hypothesis is empirical. Short words can evolve (a population subject to random point mutations and recombination) from short words to long words, with each intermediate being a word as found in the dictionary. As noted above a random sampling would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials. http://www.zachriel.com/mutagenation/ Zachriel

Zachriel

That is false. There are many selectable pathways in the region of shorter (ten or fewer letter) words. There are stepping stones!

Can you explain how you came up with this hypothesis? bill cole

seanpit: As already explained, you do have to “swim” through a small ocean of meaningless/non-beneficial sequences to find relatively rare targets – even at the level of 7-characters sequences. That is false. There are many selectable pathways in the region of shorter (ten or fewer letter) words. There are stepping stones! seanpit: Random walks or even random sampling, if you prefer, will be successful, very quickly in fact at such low levels (given a reasonably-sized population). Random walks and random sampling are much slower than evolution in the region of shorter (ten or fewer letter) words. A random sampling would take about 10^10 trials to find a ten-letter word, while evolution accomplishes the task in about 10^5 trials. seanpit: Beyond this, your assumption that meaningful/beneficial sequences will remain significantly clustered beyond the lowest levels of functional complexity is demonstrably false. Then demonstrate it. It's your claim, after all. You proposed a process to test your proposition, which hinges on "beneficial function". Your original claim seemed to take a dictionary as a test of "beneficial function" for words, so not sure why a dictionary of phrases isn't sufficient. Please provide an unambiguous measure of "beneficial function" for letter sequences. seanpit: With each step up the ladder of functional complexity the distribution of potential targets takes on an more and more randomly uniform appearance. Even simple statistical tests show that texts of English are not evenly distributed in sequence space. If so, English texts would not be compressible, when, in fact, they are highly compressible. In addition, we can simply inspect English texts and note that between spaces there are only a few tens-of-thousands of possible sequences. Zachriel

Zachriel,

seanpit: You falsely claim, here and on your website, that I drew the limit for evolutionary progress at the level of single words.

What you said was that If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” But you don’t have to swim through meaningless sequences to evolve 7-letter words. Your claim is false.

As already explained, you do have to “swim” through a small ocean of meaningless/non-beneficial sequences to find relatively rare targets – even at the level of 7-characters sequences. Does this therefore mean that evolutionary progress is statistically impossible at the level of 7-character sequences? Of course not! Random walks or even random sampling, if you prefer, will be successful, very quickly in fact at such low levels (given a reasonably-sized population). Why? Because the potential targets are relatively close and because there is still some reasonable clustering at this very low level. Yet, you falsely claim, on your website, that I said it would take “zillions of years” for success at such low levels to be realized. That’s a deliberate lie since you know full well that I've always said that evolution at such low levels happens all the time. You also know that I’ve always drawn the line for a complete stalling of evolutionary progress at the level of 1000 specifically arranged characters (amino acid residues or letters in the English language system or characters in other systems such as computer codes – etc) – which is a far cry from the very short sequences in your word evolution algorithm. Clearly then, you’ve built a strawman misrepresentation that you know isn’t true. Why? Where’s the advantage for you for such an obvious misrepresentation of someone else’s position? – beyond an effort to make it look worse than it really is? – in an effort to downplay something you think might be true? Beyond this, your assumption that meaningful/beneficial sequences will remain significantly clustered beyond the lowest levels of functional complexity is demonstrably false. With each step up the ladder of functional complexity the distribution of potential targets takes on an more and more randomly uniform appearance. This particular feature has been published numerous times in mainstream literature. So, given the reality of this situation, what on Earth makes you think that the decline in evolutionary progress, with each step up the ladder of functional complexity, will follow anything other than an exponential decay pattern? How can you possibly believe that there will only be a linear decline in evolutionary potential? – when your own algorithms suggest otherwise? – given the parameters that I’ve suggested for you above? Sean Pitman DetectingDesign.com seanpit

Me_Think,

First, points which lay on edge in smaller dimensions will come nearer in higher dimension – it will no longer be on edge.

Again, you seem to be confusing the fact that because the shortest possible linear distance (“as a crow flies”) between starting point and target does in fact significantly decrease in higher dimensions that therefore the average number of random walk steps required to find the target will also decrease in the same manner. This simply isn’t true. You’d see why if you actually sat down and did the math for the average number of random walk steps required to reach a particular target within various dimensions of sequence space (given the same ratio of very rare targets vs. non-targets). You say that the relevant math for random walks “is not relevant”, but I fail so see how? It seems to me like you simply don’t want to actually calculate the odds of a random walk hitting a particular target within a given number of steps…

Second, you don’t search the space- you search the ‘search space’. If there are totally 5,000 metabolic pathways, it is the ‘search space’ in both small and higher dimensions. IOW 5,000 metabolic pathways are spread across volume in higher dimension. So, in a 1 unit circle, those 5,000 metabolisms are spread across area of 3.14159 (Pix1^2). In a 10 dimension sphere, those 5,000 pathways are spread across(2Pi^(10/2)/Gamma[10/2])/10 volume of 2.55016. In 20 dimensions, the 5,000 metabolic pathways are spread across a volume of just 0.0258069; in a 30 dimension, the metabolic pathways are spread across a volume of just 0.0000219154 ! It takes more than 4 steps to reach edge of a unit circle but in higher dimension, it is not even a single step. In fact, in every dimensions after 5, the volume will keep decreasing and hence the random walk step to reach edge or anywhere inside will decrease too.

It is very difficult for me to follow your argument here. First off, given a particular radius, the space or “volume” of potential options within various dimensions obviously increases - exponentially. Of course, given a set number of potential options within various dimensions, the “radius” would in fact decrease exponentially with each increase in dimension. However, the volume or number of potential options or “positions” within sequence space would not decrease, but would remain the same. Likewise, the number of potential targets vs. non-targets would also remain the same. This also means that even though the shortest possible distance between the starting point and the target would in fact decrease dramatically with each increase in dimensional space, the average number of random walk steps needed for a random walker to find the target would not decrease at all – not even a little bit. https://en.wikipedia.org/wiki/N-sphere#Other_relations Why might this be? – given such a dramatic decrease in the linear distance with each increase in dimensional space? Because, 1) the number of options doesn’t change, 2) the ratio of target vs. non-target options doesn’t change, and 3) each option or location within sequence space maintains equal odds of being hit by a random walk step. That means, of course, that our rare target will not get hit by a random walk step any faster at higher vs. lower dimensions. It’s simple math. Remember now, what counts is the average number of steps required to hit a target – not the odds of getting from one side of sequence space to the other. That’s irrelevant. You can bounce all over sequence space, from one side to the other, but that doesn’t increase your odds of a random walker hitting the target. Now, please, sit down and do the math for the average number of random walk steps it takes to reach a specific target at higher and higher dimensions. It simply doesn’t change – as my previous illustrations highlighted for you. But, you need to do the math for yourself, and show your work as to how higher dimensions could possibly reduce the average number of random walk steps to find a very rare target… seanpit

seanpit: You falsely claim, here and on your website, that I drew the limit for evolutionary progress at the level of single words. What you said was that If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” But you don’t have to swim through meaningless sequences to evolve 7-letter words. Your claim is false. Zachriel

seanpit: Your version of “selective evolution” appears to be nothing more than a random sampling of the sequence space surrounding a starting position until a target is found. It's your version — evolving words one letter at a time. seanpit: “When it comes to locating small targets in large spaces, random sampling and random walks are equally ineffective.” Given two frisbees separated on a vast landscape, both random sampling and random walks will eventually find the other frisbee. Selective evolution will not. seanpit: And, in real life, each step up the ladder of functional complexity (i.e., minimum structural threshold requirements) will continue to require exponentially greater and greater amounts of time to realize via the Darwinian mechanism of random mutation and function-based selection. Repeatedly handwaving in the general direction of big numbers doesn't constitute an argument. seanpit: You’re only using these “statistical tests” at very low levels of functional complexity within the English language system. Statistical tests of an entire library show that English texts are not scattered randomly in sequence space. Otherwise, the text would not be compressible, which it is. For instance, there are only a few tens-of-thousands of possible sequences than can appear between spaces. (We call them "words".) seanpit: As far as an “unambiguous measure” of a beneficial function, I’m not sure it is possible to be clearer than the definition used for biological systems – i.e., a system that produces a survival/reproductive advantage for the organism in a given environment vs. the rest of its peers. You're just rewording the question. How do we test the "survival" of a sequence of letters? It's your claim, after all. Let's try this: A sequence of words from Shakespeare have obviously survived. So we might use the works of Shakespeare as a dictionary, or even just a single play, say Hamlet. If a word sequence matches something in Shakespeare, then it has met the test of survival. seanpit: If we’re not talking about biological evolution, then our conversation is over. YOU made the claim about words. Sure, it is meant to be representative of biological evolution, but it was specifically a claim about words. As such, either you need to support it or abandon it. At this point, we're still trying to apply some rigor to your notion of "2) Select based on changes in beneficial function". Zachriel

seanpit @ 118

Here’s a relevant math question for you:

It is not relevant. Please see comment @ 119 Me_Think

seanpit @ 113

Again, sit down and actually do the math… Here, I’ll help you: For 2D space: The number of possible positions at a Distance of D = circumference of a circle (2*?*R) or 2*?*D. Volume: ?*R^2 Odds of hitting a particular point at Distance D = 1/(2*?*D). A rough calculation of the average number of steps (N) to find a particular point at Distance D = D^2 / 1/(2*?*D) If D = 5 the answer would be: 52 / 1/ (2 * 3.14 * 5) = 25 / (1 / 31.42) = 25 / 0.0318 = ~785.5 steps For 3D space: The number of possible positions at Distance D = surface area of a sphere (4*?*R^2) or 4*?*D^2 Volume: 4/3 ?*R^3

First, points which lay on edge in smaller dimensions will come nearer in higher dimension - it will no longer be on edge. Second, you don't search the space- you search the 'search space'. If there are totally 5,000 metabolic pathways, it is the 'search space' in both small and higher dimensions. IOW 5,000 metabolic pathways are spread across volume in higher dimension. So, in a 1 unit circle, those 5,000 metabolisms are spread across area of 3.14159 (Pix1^2). In a 10 dimension sphere, those 5,000 pathways are spread across (2Pi^(10/2)/Gamma[10/2])/10 volume of 2.55016. In 20 dimensions, the 5,000 metabolic pathways are spread across a volume of just 0.0258069; in a 30 dimension, the metabolic pathways are spread across a volume of just 0.0000219154 ! It takes more than 4 steps to reach edge of a unit circle but in higher dimension, it is not even a single step. In fact, in every dimensions after 5, the volume will keep decreasing and hence the random walk step to reach edge or anywhere inside will decrease too. P.S: I have taken unit circle as the template to avoid any ambiguity in calculating random walk steps. Me_Think

Me_Think, Here's a relevant math question for you: 2D space: Draw a circle around a random walk starting point at a given radius within a space that has a radius at least twice as large as the circle. How many random walk steps are expected to reach the edge of the circle? Now, how many random walk steps are expected to reach a particular point on the edge of the circle? Please show your math... 3D space: Draw a sphere around a random walk starting point at a given radius within an overall space that has a radius at least twice as large as the sphere. Make sure the volume of the sphere contains the same number of potential locations as contained with the 2D circle. How many random walk steps are expected to reach the edge of the sphere? Now, how many random walk steps are expected to reach a particular point on the edge of the sphere? Please show your math... seanpit

Zachriel,

seanpit: Not, it’s not because natural selection cannot select, in a positive manner, until a new beneficial sequence is found by purely random chance.

Which is why selective evolution does not work the same. The question is whether a random walk and selective evolution work the same on a vast landscape with just two viable points, depending on dimension. They do not. A random walk works whether the landscape is two dimensions or with vast dimensions. Selective evolution only works with the latter case. Random walks and selective evolution do not work the same.

Your version of “selective evolution” appears to be nothing more than a random sampling of the sequence space surrounding a starting position until a target is found. Such random sampling of sequence space is not affected by the dimensions of that space given the same ratio of potential targets vs. non-targets within that space. In other words, random sampling would be just as effective in higher dimensional space at it would be in 2D space. The same is true for random walks - on a bounded k-dimensional lattice (or hypercube). You see, “When it comes to locating small targets in large spaces, random sampling and random walks are equally ineffective.” There simply is no advantage for using one vs. the other. http://billdembski.com/documents/2005.03.Searching_Large_Spaces.pdf

In our previous conversations, we published voluminous data concerning the behavior of selective evolution, including the increasing time to reach longer sequences. It took a few billion mutations to reach sequences of several hundred bits (5 bits per character).

And, in real life, each step up the ladder of functional complexity (i.e., minimum structural threshold requirements) will continue to require exponentially greater and greater amounts of time to realize via the Darwinian mechanism of random mutation and function-based selection. Again, this is why evolutionary progress stalls out, this side of a practical eternity of time before the level of systems of function that require a minimum of 1000 specifically arranged characters can be reached.

Even simple statistical tests show that texts of English are not randomly distributed in sequence space. In any case, you seemingly can’t provide an unambiguous measure “beneficial function” of letter sequences.

You’re only using these “statistical tests” at very low levels of functional complexity within the English language system. Of course short words and even phrases in English do not have an entirely uniform random distribution within sequence space! I’m the one who pointed this out to you back in 2004! However, if you analyze the pattern at higher and higher levels of functional complexity, you will notice that functionally meaningful/beneficial sequences start to take on a more and more randomly uniform distribution. The very same thing happens with functional DNA and proteins in living things. Very quickly these sequences take on a uniformly random appearance in their locations within sequence space. As far as an “unambiguous measure” of a beneficial function, I’m not sure it is possible to be clearer than the definition used for biological systems – i.e., a system that produces a survival/reproductive advantage for the organism in a given environment vs. the rest of its peers. If you wish to simulate this situation with computer code or the English language, you will have to set up an environment with organisms that function based on coded sequences where certain types of sequences produce functions that allow for better survival/reproduction in that environment relative to the other “organisms” in the population. In other words, you have to create a competitive environment where different sequences provide some kind of functional advantage. Then, you have to see if organisms that have these sequences will evolve additional beneficial sequences that have even greater minimum size/specificity requirements.

seanpit: It is just that at such a high level of functional complexity, the potential beneficial targets in this space are so extremely rare that the minimum Levenshtein distances between what exists within the current gene pool and the next closest potentially beneficial sequence within 5000aa sequence space is simply too far to cross this side of a practical eternity of time.

That would imply that such sequences can’t sustain any mutations, which is simply not the case.

You’re mistaken. Such 5000 character systems and sequences can and do sustain many mutations in real life - and even undergo random walks in real life. However, none of these mutations or random walks ever comes across another qualitatively novel beneficial system at such a high level of functional complexity nor is this likely to happen this side of a practical eternity of time.

We’re not talking about biological evolution, but your claim about words. “Beware a war of words …”

If we’re not talking about biological evolution, then our conversation is over. For me, the conversation has always been about the potential and limits of biological evolution via the Darwinian mechanism of random mutations and function-based selection. You falsely claim, here and on your website, that I drew the limit for evolutionary progress at the level of single words. You know that’s not true, but you keep saying it anyway. You knew full well that I’ve always drawn the line for a limit to evolutionary progress at the level of 1000 specifically arranged characters. Of course, the same limitations with the Darwinian mechanism would also exist for any other system of meaningful/functional information – to include the English language system. This is why the Darwinian mechanism is never used to create masterpieces of literature – like the works of Shakespeare for instance. Such masterpieces are always produced by intelligent design and could only be produced by intelligent design this side of a practical eternity of time. seanpit

seanpit: Not, it’s not because natural selection cannot select, in a positive manner, until a new beneficial sequence is found by purely random chance. Which is why selective evolution does not work the same. The question is whether a random walk and selective evolution work the same on a vast landscape with just two viable points, depending on dimension. They do not. A random walk works whether the landscape is two dimensions or with vast dimensions. Selective evolution only works with the latter case. Random walks and selective evolution do not work the same. seanpit: A mutation that provides a survival/reproductive advantage in a given environment – as you already know. Rewording the question isn't an answer. seanpit: Select based on changes in beneficial function – not template-matching which doesn’t happen in real life. Your basic claim concerns words, so words in the dictionary are a reasonable way to determine whether a sequence of letters is a valid word. If we were to use a similar dictionary of phrases, then it would be the equivalent. A play of Shakespeare comes to mind. Any consecutive words in Shakespeare are certainly "more beneficial" than random words strung together. However, you didn't like this method. That's fine; however, as you made the claim, it is up to you to provide an unambiguous way to determine the "beneficial function" of strings of characters. Or you could simply abandon your claim as an ill-chosen analogy. seanpit: Now, you know that longer beneficial sequences take longer to find, but you have yet to admit that the relationship isn’t a linear relationship, but an exponential relationship. In our previous conversations, we published voluminous data concerning the behavior of selective evolution, including the increasing time to reach longer sequences. It took a few billion mutations to reach sequences of several hundred bits (5 bits per character). seanpit: The credibility of this claim is backed up by multiple lines of evidence – to include the observed nature of sequence space where known beneficial sequences have an essentially uniformly random distribution ... Even simple statistical tests show that texts of English are not randomly distributed in sequence space. In any case, you seemingly can't provide an unambiguous measure "beneficial function" of letter sequences. seanpit: It is just that at such a high level of functional complexity, the potential beneficial targets in this space are so extremely rare that the minimum Levenshtein distances between what exists within the current gene pool and the next closest potentially beneficial sequence within 5000aa sequence space is simply too far to cross this side of a practical eternity of time. That would imply that such sequences can't sustain any mutations, which is simply not the case. seanpit: I’ve never seen any other evolutionist even attempt to argue that random walks (neutral evolution over time) weren’t a significant part of Darwinian evolution. We're not talking about biological evolution, but your claim about words. "Beware a war of words ..." Zachriel

Zachriel,

They’re available, they’re just not viable. Consider again the two-dimensional case. There are two frisbees separated on a vast plain. We can’t leave the first frisbee because every point in the neighborhood is unviable. That doesn’t mean there aren’t two dimensions. The availability of viable points is contingent on the particular landscape. If there happens to be a frisbee in the neighborhood, then we can move to that point. If not, we can’t.

While it is correct to argue that highly beneficial sequences will be maintained by nature selection, thereby limiting the random walk potential from this starting point, it is completely wrong to argue that therefore everything within a genome will be so constrained by natural selection. I'm really amazed that you keep presenting this argument as it is completely off base. In fact, it is so off base that I've never heard it presented by any other evolutionist before. The fact is that there are many sequences that provide no significant functional advantage within genomes as far as their specific sequence structure is concerned (i.e. they are truly "neutral sequences" with respect to function). Such neutral sequences can and do in fact undergo true random walks without any significant influence or restraint from natural selection. And, that means, of course, that all of sequence space is open to potential discovery by these random walks. seanpit

Origenes,

Sure, I agree that from a viable position an organism can do a search. However, my self-evident point is that it cannot do so from an unviable position. IOWs unviable positions cannot serve as pathways/connections in Wagner’s hypercube. From a viable position 5000 ‘dimensions’ are searchable, from an unviable position zero dimensions are searchable.

Not quite true since a genome, like the human genome, can and does maintain essentially random sequences of DNA that can undergo random walks and “search” within all 5000 dimensions. seanpit

Me_Think

I think you are confusing Root Mean Square distance of random walk with the mean distance of random walk. The formula for root mean square is square root of the steps, but in higher dimensions, the mean distance is represented by the formula Sqrt[(2 N)/d]*([CapitalGamma]*((d + 1)/2))/([CapitalGamma]*(d/2))where d is the number of dimension and N is the number of steps. so for 10,000 steps, the mean random walk distance in 4,5,6,7,8,9 and 10 dimensions will be 88.3883,75.8947,67.3575,61.0883,56.25,52.3783 and 49.1935 respectively. (Reference:http://math.stackexchange.com/.....andom-walk)

Again, your logic is mistaken here. You see, I’m not confusing the root mean distance with the mean distance of a random walk. These are calculations as to how far, in absolute terms, a random walker is likely to be from the starting point after N number of steps. However, this isn’t the right formula for calculating the average number of steps to find a particular target location in sequence space at various dimensions. In other words, this isn’t the formula you need to calculate the average time to find our Frisbee within various dimensions where the number of potential locations remains the same. Again, this your formula here is for the expected absolute distance that a random walker would be from the original starting point after a given number of steps – without regard to finding a specific location within sequence space. In other words, this is not the formula for determining the average time needed for a random walker to hit upon a particular point in sequence space that is a particular distance or radius from the starting point. So, while it is quite clear, even intuitive, that the absolute linear distance a random walker will likely travel away from a particular starting point will be shorter in higher dimensions of sequence space (given the same number of random walk steps), it is not true that this therefore means that a rare target will be found faster in higher dimensions of sequence space (given the same ratio of targets vs. non-targets). In other words, this by no means helps you find the Frisbee any faster in higher dimensional space. Yet again, I advise you to actually sit down and do the math with regard to the average number of steps our random walker will take to find our rare Frisbee within sequence spaces at different dimensions given the same ratio of targets vs. non-targets.

What is interesting is that the Polya Constants decreases as the dimensions increase(0.19,0.13,0.10,0.08 and 0.07 in 4,5,6,7 and 8 dimensions respectively ) , which indicates that the random walk’s probability of returning to same point (or even origin) decreases. This means the chance of finding new genotype/metabolism increases

Not true. This does not mean that the chance of finding our rare Frisbee, or any other rare target, increases with increasing dimensions of sequence space. The “Polya Constant” simply means that for sequence spaces of dimension 1 or 2, the odds of returning to the starting point for an infinite random walk are 100%. However, for higher dimensions than 2, the odds of returning to the starting point, even for an infinite random walk distance, are less than 100%. In other words, the random walker can get “lost” in higher dimensional sequence space and never return home. This reality, unfortunately, does not improve the odds of success when it comes to finding a rare target in higher dimensions of sequence space (given the same ratio of targets vs. non-targets). Again, sit down and actually do the math… Here, I’ll help you: For 2D space: The number of possible positions at a Distance of D = circumference of a circle (2*?*R) or 2*?*D. Volume: ?*R^2 Odds of hitting a particular point at Distance D = 1/(2*?*D). A rough calculation of the average number of steps (N) to find a particular point at Distance D = D^2 / 1/(2*?*D) If D = 5 the answer would be: 52 / 1/ (2 * 3.14 * 5) = 25 / (1 / 31.42) = 25 / 0.0318 = ~785.5 steps For 3D space: The number of possible positions at Distance D = surface area of a sphere (4*?*R^2) or 4*?*D^2 Volume: 4/3 ?*R^3 A rough calculation of the average number of steps (N) to find a particular point at Distance D = D^2 / 1/(4* ?*D^2) If D = 5 the answer would be: 25/ (1/314) = 25 / 0.0032 = ~7850 steps Notice how expected number of random walk steps required to find a specific target in higher dimensional space increases exponentially if the radius remains the same. But, what happens if the absolute volume of sequence space remains the same? – if the number of possible positions within sequence space remains the same between lower and higher dimensions? The radius between starting point and target would be reduced from 5 to 2.924. So, how many random walk steps should be expect now to reach the target now? If D = 2.924 the answer would be: (2.924)^2 / 1 / (4*?*(2.924)^2) = 8.549776 / (1 / 107.44) = 918 steps The difference here between the 2D and 3D numbers is due to the rough nature of the calculation. Using larger distances would produce more similar values. But, you get the idea. Searching for rare targets within higher dimensions simply offers no advantage compared to lower dimensions.

This will help conceptualize why sparse landscape doesn’t matter: Draw a line of 100 cm on a piece of paper. At the end of the line draw a green circle to represent a new genotype. Now, in one dimension, the number of random walk steps to reach that circle will be more than 100 (since random walk can also be backward in 1 dimension). Crush the paper into a ball to represent higher dimension. The circle is now just a step or at most few step away from the start. It doesn’t matter that the landscape is sparse in higher dimension. Look at how dense a 7 and 10 dimension Hypercube is here : http://imgur.com/a/eZzZA Now imagine how much area will be covered in just 1 step as the dimensions grow.

Again, it doesn’t matter the linear distance that is traveled from your starting point. Your random walker is likely to travel the entire distance of higher dimensional sequence space many times over as compared to lower dimensional sequence space (given the same number of steps in both situations). However, this reality will not help the random walker find a rare target within higher dimensional sequence space any more efficiently. Again, do the math yourself if you don’t believe me. Actually sit down and calculate the odds of finding a particular location within sequence spaces within various dimensions of sequence space where the total number of possible positions remains the same. seanpit

Origenes: How can one coherently claim that there are so many dimensions if the vast majority of those dimensions are unavailable? They're available, they're just not viable. Consider again the two-dimensional case. There are two frisbees separated on a vast plain. We can't leave the first frisbee because every point in the neighborhood is unviable. That doesn't mean there aren't two dimensions. The availability of viable points is contingent on the particular landscape. If there happens to be a frisbee in the neighborhood, then we can move to that point. If not, we can't. Generally, we define the dimensions by the number of independent variables, such as genetic bases. Origenes: How can dimensions, where an organism cannot go simply because those dimensions are not viable, still function as pathways/connections to other viable points? If they're not viable, then they can't connect us to other viable points. Again, in two-dimensions, just because we can't go north, doesn't mean we can't go east or west, if there happens to be a frisbee on that spot. Phinehas: So, was there any intelligence involved in the construction of the above or not? Of course. However, it answers the objection that there are no connections through phase-space. Zachriel

Me_Think,

MT: The network is made of ‘viable’ and ‘unviable’ pathways.

Indeed. And the vast majority of 'pathways' is unviable, which renders them not-functional. IOWs the vast majority of "pathways" are actually no pathways at all.

MT: Why do you think all pathways should be viable?

Well, obviously because unviable pathways imply sickness and death of an organism. What do you think that unviable means?

MT: A random walk doesn’t need viable pathway to traverse across the network.

Why not? Do you hold that unviable mutations can function as stepping stones? In Wagner's hypercube a series of unviable metabolic sequences may provide a shortcut to evolutionary success for an organism. However its obvious that an organism cannot tolerate those steps. Even Zachriel understands this:

What happens is that the organism takes a step, and if that step isn’t viable, the organism *remains where it is*. It continues to check each neighboring step until or if it finds one where it is viable.

IOWs "unviable pathways" are no pathways at all, do not provide shortcuts and do not connect anything, they function as stop signs instead. Origenes

Origenes @ 109

In order to connect the circle with the start, in order for all those dimensions to serve as pathways and to be interconnected, one needs unviable positions to function as pathways, and this they cannot do.

The network is made of 'viable' and 'unviable' pathways. Why do you think all pathways should be viable? A random walk doesn't need viable pathway to traverse across the network. Me_Think

Seanpit: A dimension space of 5000 isn’t really that difficult to achieve. For example, a protein-based system that uses a minimum of 5001aa is contained within a sequence space of 5000 dimensions. And, all of those dimensions are in fact searchable by random walk mutations.

Sure, I agree that from a viable position an organism can do a search. However, my self-evident point is that it cannot do so from an unviable position. IOWs unviable positions cannot serve as pathways/connections in Wagner’s hypercube. From a viable position 5000 ‘dimensions’ are searchable, from an unviable position zero dimensions are searchable.

Me_Think: Draw a line of 100 cm on a piece of paper. At the end of the line draw a green circle to represent a new genotype. Now, in one dimension, the number of random walk steps to reach that circle will be more than 100 (since random walk can also be backward in 1 dimension). Crush the paper into a ball to represent higher dimension. The circle is now just a step or at most few step away from the start.

In order to connect the circle with the start, in order for all those dimensions to serve as pathways and to be interconnected, one needs unviable positions to function as pathways, and this they cannot do. Origenes

This will help conceptualize why sparse landscape doesn't matter: Draw a line of 100 cm on a piece of paper. At the end of the line draw a green circle to represent a new genotype. Now, in one dimension, the number of random walk steps to reach that circle will be more than 100 (since random walk can also be backward in 1 dimension). Crush the paper into a ball to represent higher dimension. The circle is now just a step or at most few step away from the start. It doesn't matter that the landscape is sparse in higher dimension. Look at how dense a 7 and 10 dimension Hypercube is here : http://imgur.com/a/eZzZA Now imagine how much area will be covered in just 1 step as the dimensions grow. Me_Think

Origenes @ 100

It seems to me a matter of logic that all those countless unviable dimensions cannot function as pathways/connections. IOWs also for this reason Wagner’s concept of a 5000-dimensional cube doesn’t make sense.

The 5000 dimensions is based on 5000 (now more) actual metabolic pathways. You can explore the pathways at http://biocyc.org/biocyc-pgdb-list.shtml#tier1 If you really want to understand his concept, you have to download the free Math Lab package and explore. the MathLab package - Hyperspace is at : http://www.ieu.uzh.ch/wagner/publications-software.html Me_Think

seanpit @ 93

You’re mistaken. Again, even though the linear distance is much much shorter within a higher dimensional room, the average random walk distance needed to find a Frisbee is not shorter at all because, in our illustration, there are the same number of non-target options regardless of the dimensions of the search space being considered.....Why don’t you just sit down and actually do the math?

I think you are confusing Root Mean Square distance of random walk with the mean distance of random walk. The formula for root mean square is square root of the steps, but in higher dimensions, the mean distance is represented by the formula Sqrt[(2 N)/d]*([CapitalGamma]*((d + 1)/2))/([CapitalGamma]*(d/2)) where d is the number of dimension and N is the number of steps. so for 10,000 steps, the mean random walk distance in 4,5,6,7,8,9 and 10 dimensions will be 88.3883,75.8947,67.3575,61.0883,56.25,52.3783 and 49.1935 respectively. (Reference:http://math.stackexchange.com/questions/103142/expected-value-of-random-walk) What is interesting is that the Polya Constants decreases as the dimensions increase (0.19,0.13,0.10,0.08 and 0.07 in 4,5,6,7 and 8 dimensions respectively ) , which indicates that the random walk's probability of returning to same point (or even origin) decreases. This means the chance of finding new genotype/metabolism increases Me_Think

Zachriel,

We have seen the explanation. We have also shown why evolution does not work like a random walk. Consider a very simple case, with two viable points separated on a vast two-dimensional landscape. A random walk will eventually lead from one viable point to another viable point. However, evolution will never leave the first viable point because every neighbor to it is unviable. Consider now two viable points on a landscape with a vast number of dimensions such that any two points are now a single step away from another. Yes, you may have to test large numbers of neighbors, but you remain viable the entire time during and until you find the other viable point.

I’ve never seen any other evolutionist even attempt to argue that random walks (neutral evolution over time) weren’t a significant part of Darwinian evolution. Beyond the fact that you’re simply wrong here (random walks do indeed happen in all gene pools over time) it doesn’t matter when it comes to solving the problem at hand. You’re claiming that when evolution works it always works by single steps off a starting island into the surrounding ocean – sampling to see what might be found from a safe vantage point. Even this view were in fact the only way that evolution could proceed, it simply wouldn’t proceed fast enough beyond very low levels of functional complexity. At the level of 1000 saars, your method would be very very unlikely to find anything this side of trillions upon trillions of years of time. There simply is no statistical advantage to this method of yours when it comes to solving the problem at hand. It just wouldn’t work remotely fast enough when it comes to finding higher level systems. In short, you’re in the same boat as all the other evolutionists. Your mechanism just can’t do the job you claim it did. seanpit

Origenes,

How can one coherently claim that there are so many dimensions if the vast majority of those dimensions are unavailable? How can dimensions, where an organism cannot go simply because those dimensions are not viable, still function as pathways/connections to other viable points? How can a dimension, where no organism can go, still be of assistance to an organism? It seems to me a matter of logic that all those countless unviable dimensions cannot function as pathways/connections. IOWs also for this reason Wagner’s concept of a 5000-dimensional cube doesn’t make sense.

A dimension space of 5000 isn’t really that difficult to achieve. For example, a protein-based system that uses a minimum of 5001aa is contained within a sequence space of 5000 dimensions. And, all of those dimensions are in fact searchable by random walk mutations. It is just that at such a high level of functional complexity, the potential beneficial targets in this space are so extremely rare that the minimum Levenshtein distances between what exists within the current gene pool and the next closest potentially beneficial sequence within 5000aa sequence space is simply too far to cross this side of a practical eternity of time. It just doesn’t happen… seanpit

Zachriel,

The behavior is different with evolution by selection.

Not, it’s not because natural selection cannot select, in a positive manner, until a new beneficial sequence is found by purely random chance. As the minimum Levenshtein distance to the next closest potentially beneficial sequences increases in a linear manner, with each step up the ladder of functional complexity, the average time to success increases exponentially. Natural selection cannot solve this problem…

Great! Then you should have no problem providing an unambiguous measure of “beneficial function”.

A mutation that provides a survival/reproductive advantage in a given environment – as you already know.

No need to reinvent the wheel. It does take longer to find longer words. However, you claim this results in an effective wall, but have been unable to even define a valid selection criterion, meaning your claim is not testable.

Yet again, I’ve been quite clear over the years discussing this topic with you and many others that the “effective wall” for the Darwinian mechanism is at the level of systems of function that require a minimum of 1000 specifically arranged characters (amino acids residues when discussing living things). I’ve also been quite clear that evolution works just fine when you’re talking about systems of function that require only a handful of specifically arranged characters – or even a couple hundred! My claim has always been that you will see an exponential stalling effect with each step up the ladder of functional complexity. By the time the level of 1000 saars is reached, evolutionary progress completely stalls out this side of a practical eternity of time. Now, you know that longer beneficial sequences take longer to find, but you have yet to admit that the relationship isn’t a linear relationship, but an exponential relationship. Systems of function that require a greater minimum size (given a certain degree of specificity) take exponentially longer to find in sequence space by random search algorithms. It is this pattern that you can extrapolate quite clearly to higher levels of functional complexity and see that the Darwinian mechanism simply isn’t up to the job that you claim it is. It just doesn’t work, it cannot work, beyond the lowest rungs of the ladder. The credibility of this claim is backed up by multiple lines of evidence – to include the observed nature of sequence space where known beneficial sequences have an essentially uniformly random distribution which becomes more and more prominent at higher levels of functional complexity. It is also known that examples of evolution in action become exponentially rarer with each step up the ladder of functional complexity. And, there are absolutely no observable examples of evolution in action beyond the level of 1000 saars. It just doesn’t happen for very good empirical and statistical reasons. Your vision that high level sequence space is somehow “stacked” just right so that the Darwinian mechanism can work – despite the extreme odds against this notion of yours, is just wishful thinking. There is no evidence to back up this truly fantastic notion. In any case, such an extreme example of stacking the deck, if ever identified, would be very very good evidence of deliberate design – not mindless chance. Unless, of course, you subscribe to the multiverse nonsense where our universe just happened to turn our right in so many extremely unlikely ways “By Sheer Chance”. Please… seanpit

Z:

Now, please provide an unambiguous way to determine changes in “beneficial function” so we can try the test what you proposed.

Step one: Publish on Amazon the first book your Phrasenation generates. Step two: Note the average number of stars in the book's reviews. Step three: Keep generating and publishing books until you get one with a higher rating. Step four: Once you hit three stars on a book, charge money. Step five: Profit! I hope there isn't anything ambiguous about the above. Phinehas

O Sean Pitman Beware a war of words ere you err. A man wins the crown, but lowers his helm. A kiss Is a kiss, and a war can be just, but a war of words Just irks the crowd and leads you far astray. Words, you know, can lead to a clash of swords. Why do you think that you alone have it Legit when sages aver another idea? Could it be that you could see the light But choose instead to close your eyes and block The sight? The origin of the life we know Just like this poem rose from simple forms, In meaning, and in kind, step-by-step.

So, was there any intelligence involved in the construction of the above or not? Phinehas

How can one coherently claim that there are so many dimensions if the vast majority of those dimensions are unavailable? How can dimensions, where an organism cannot go simply because those dimensions are not viable, still function as pathways/connections to other viable points? How can a dimension, where no organism can go, still be of assistance to an organism? It seems to me a matter of logic that all those countless unviable dimensions cannot function as pathways/connections. IOWs also for this reason Wagner's concept of a 5000-dimensional cube doesn't make sense. Origenes

Origenes: Only if hyper-dimensions would improve things, which it doesn’t; see the explanations by Sean Pitman. We have seen the explanation. We have also shown why evolution does not work like a random walk. Consider a very simple case, with two viable points separated on a vast two-dimensional landscape. A random walk will eventually lead from one viable point to another viable point. However, evolution will never leave the first viable point because every neighbor to it is unviable. Consider now two viable points on a landscape with a vast number of dimensions such that any two points are now a single step away from another. Yes, you may have to test large numbers of neighbors, but you remain viable the entire time during and until you find the other viable point. Zachriel

Zachriel: It’s schematic, but shows how complex systems can evolve from less integrated structures.

Only if hyper-dimensions would improve things, which it doesn't; see the explanations by Sean Pitman. Origenes

Origenes: Unfortunately that doesn’t solve the problem It's schematic, but shows how complex systems can evolve from less integrated structures. Zachriel

Zachriel: What you are doing is quoting a rhetorical question, posed in the prologue of a book wherein the author answers the question.

Correction: the author attempts to answer the "rhetorical question" by proposing the existence of a metabolic library in a 5000-dimensional hypercube. Unfortunately that doesn't solve the problem; see the explanations by Sean Pitman. Origenes

Origenes: "And not nearly enough time." Evolution isn't random. What you are doing is quoting a rhetorical question, posed in the prologue of a book wherein the author answers the question. seanpit: That is why each linear increase in the Levenshtein distance will result in an exponential increase in the average random walk distance – regardless of the dimensions of the search spaces you’re considering. The behavior is different with evolution by selection. seanpit: There are many different ways to do this – as I’ve already explained. Great! Then you should have no problem providing an unambiguous measure of "beneficial function". seanpit: In other words, given these parameters, you will see a non-linear increase in the average number of generations needed for your small population to find longer and longer defined words No need to reinvent the wheel. It does take longer to find longer words. However, you claim this results in an effective wall, but have been unable to even define a valid selection criterion, meaning your claim is not testable. Zachriel

Zachriel

Sure, there are random walks in biological evolution, but never such that the organism is no longer viable. In any case, your original claim concerned selectable stepwise evolution, not random walks.

The genomes of highly complex organisms, like humans for instance, consists of functionally beneficial as well as functional neutral genetic sequences. And, even functionally beneficial genetic sequences are not so rigid that they cannot experience random walks within their “neutral nets” as well. So, any reasonable simulation of the evolutionary algorithm will need to model random walks. Beyond this, your notion that evolution must always proceed by individual mutations leaping into the surrounding sequence space from a starting point island until something beneficial is found is complete nonsense. Such a notion would only be successful at very very low levels of functional complexity. It is just extraordinary to me that you actually believe that it is statistically likely, within what anyone would call a reasonable amount of time, to find something beneficial at a level beyond 1000 specifically arranged amino acid residues – with just a single mutation of any kind! That’s complete nonsense! By all means, what makes you think that this situation is remotely likely beyond the lowest levels of functional complexity? What makes you think that sequence spaces at such high levels are so extraordinarily structured as you imagine them to be? Oh, I know, because they would have to be for your evolutionary philosophy to be tenable. However, where's the "science" to support that notion? Where is your actual evidence that sequence space at high levels is set up like this?

Great! According to your test, we have to “Select based on changes in beneficial function.” Now, please provide an unambiguous way to determine changes in “beneficial function” so we can try the test what you proposed.

There are many different ways to do this – as I’ve already explained. It’s just that no one has ever observed, in computer simulations or biological systems, evolution of qualitatively novel beneficial systems beyond the level of 1000 specifically arranged characters. It’s not even close. Evolutionary progress clearly stalls out, in all of these simulations and in real life, on very low rungs of the ladder of functional complexity. Even in your own word-evolution simulations, with selection only based on a match to the dictionary, you will see early features of this stalling effect - if you: 1) reduce the steady state population size to or two or three, 2) limit the reproduction rate to 2 offspring per individual per generation, and 4) limit the mutation rate to one mutation per individual per generation, and 5) allow for some actual random walk (by starting with a longer randomly-generated sequence for the “genome” of say 20 characters for each individual in your starting population divided into, say, 5 separate sequences, divided by spaces (an extra character to give a total of 27 character options) with a 6) maximum genome size of, say, 50 characters per individual genome – where one of these original sequences is your starting single character “word” – like “O” or “I” or “A”). Positive selection will be based on the discovery, by any part of the genome in any individual within the overall population, of a new longer word. At this point, this particular individual genome will be used to populate the next generation (i.e., the other two genomes will be killed off). Also, remember, “recombination mutations” can only occur within the genomes of a single individual in this situation – not between different individuals in the population. In other words, given these parameters, you will see a non-linear increase in the average number of generations needed for your small population to find longer and longer defined words – even without basing selection on an increase in beneficial function. If you want to move on from this point to modify your “Phrasentation” program to be more reflective of the problems the Darwian mechanism faces in real life, set it up, along the same parameters described above, where selection, in this case, is based, not on an increase in beneficial function, but upon sequence matching to entire sentences, then entire paragraphs, then entire chapters – etc. You can increase the maximum genome size to 50,000 characters per individual as additional levels of success are achieved. In such a situation, you would see a non-linear increase in the number of generations required to achieve the next “level” of evolutionary progress. Using real function-based selection would only add, obviously, to the difficulties for the Darwinian mechanism. Also, it would be helpful to modify your program to also work with 64-bit Excel spreadsheets. seanpit

Me_Think

Sean Pitman: You see, you are confusing a linear distance with a random walk distance. They aren’t the same thing. While the shortest linear distance is indeed significantly decreased within higher spacial dimensions, the random walk distance remains the same.

Certainly not. If you lock up a drunkard each in a large and small room, the drunkard in the smaller room will hit the wall ( or, if you like, step on Frisbee thrown on floor) first. This is because the unit size of a random walk step doesn’t decrease with decease in linear distance in any dimensions.

You’re mistaken. Again, even though the linear distance is much much shorter within a higher dimensional room, the average random walk distance needed to find a Frisbee is not shorter at all because, in our illustration, there are the same number of non-target options regardless of the dimensions of the search space being considered. That is why it takes just as many steps for a random walker to find the Frisbee regardless of the dimensions of a room with the same number of potential locations. This is also why, at higher dimension, the same linear distance to a target that existed within a lower-dimensional space will result in an exponential increase in the random walk distance within a higher-dimensional space. In this line, the Levenshtein distance is a measure of distance within hyper-dimensional space. That is why each linear increase in the Levenshtein distance will result in an exponential increase in the average random walk distance – regardless of the dimensions of the search spaces you’re considering. Why don’t you just sit down and actually do the math? If you do, you’ll soon recognize your mistake. seanpit

Zachriel #91, Andreas Wagner makes it sound even worse:

The first vertebrates to use crystallins in lenses did so more than five hundred million years ago, and the opsins that enable the falcon’s vision are some seven hundred million years old.10 They originated some three billion years after life first appeared on earth. That sounds like a helpfully long amount of time to come up with these molecular innovations. But each one of those opsin and crystallin proteins is a chain of hundreds of amino acids, highly specific sequences of molecules written in an alphabet of twenty amino acid letters. If only one such sequence could sense light or help form a transparent cameralike lens, how many different hundred-amino-acid-long protein strings would we have to sift through? The first amino acid of such a string could be any one of the twenty kinds of amino acids, and the same holds for the second amino acid. Because 20 × 20 = 400, there are there are 400 possible strings of two amino acids. Consider also the third amino acid, and you have arrived at 20 × 20 × 20, or 8,000, possibilities. At four amino acids we already have 160,000 possibilities. For a protein with a hundred amino acids (crystallins and opsins are much longer), the numbers multiply to a 1 with more than 130 trailing zeroes, or more than 10^130 possible amino acid strings. To get a sense of this number’s magnitude, consider that most atoms in the universe are hydrogen atoms, and physicists have estimated the number of these atoms as 10^90, or 1,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000, 000,000,000,000,000,000,000,000,000,000,000,000,000,000. This is “only” a 1 with 90 zeroes. The number of potential proteins is not merely astronomical, it is hyperastronomical, much greater than the number of hydrogen atoms in the universe.11 To find a specific sequence like that is not just less likely than winning the jackpot in the lottery, it is less likely than winning a jackpot every year since the Big Bang.12 In fact, it’s countless billions of times less likely. If a trillion different organisms had tried an amino acid string every second since life began, they might have tried a tiny fraction of the 10^130 potential ones. They would never have found the one opsin string. There are a lot of different ways to arrange molecules. And not nearly enough time.

Origenes

Origenes: The more specified a function/organism is, the more specified parts are involved, the more parts are pointed in the same direction, the less evolvable/flexible a function/organism becomes. Gee whiz! It almost sounds like it might take millions of years for many evolutionary changes to occur! Zachriel

Origenes @ 88 ..Which is why he wrote the book to explain how robustness and hyper-dimensional search helps in increasing the probability of locating sparse genotypes / metabolism Me_Think

You see, you are confusing a linear distance with a random walk distance. They aren’t the same thing. While the shortest linear distance is indeed significantly decreased within higher spacial dimensions, the random walk distance remains the same.

Certainly not. If you lock up a drunkard each in a large and small room, the drunkard in the smaller room will hit the wall ( or, if you like, step on Frisbee thrown on floor) first. This is because the unit size of a random walk step doesn't decrease with decease in linear distance in any dimensions. Me_Think

Seanpit: If the mechanism involves random mutations and function-based selection, that mechanism, beyond very low levels of functional complexity, will stall out, in an exponential manner, with each step up the ladder of functional complexity.

Intuitively this makes a lot of sense. The more specified a function/organism is, the more specified parts are involved, the more parts are pointed in the same direction, the less evolvable/flexible a function/organism becomes. BTW Andreas Wagner made a similar observation against interest:

Some of nature’s ways to find new metabolic texts are familiar, because they dominate in large multicellular animals like us. They include the changes accompanying sexual reproduction, which shuffles chromosomes like decks of cards, so that each of our children starts with a new deal. Then there are the spontaneous mutations in a DNA’s letter sequence, arising through chance events such as when photons of ultraviolet radiation smash into the genome, or through highly reactive oxygen radicals that are by-products of chemical reactions and burst the chemical bonds of nearby DNA. Neither way to explore the metabolic library is very effective. Since the shuffling of sexual reproduction occurs between highly similar genomes—two human genomes share 99.9 percent of their DNA letter sequence—it is not the most effective way to create new metabolisms.15 It is like trying to write a new play by changing thirty words in Hamlet. And while mutations can create new proteins, including new enzyme catalysts, they are rare, which means the process is rather slow.

Origenes

seanpit: functionally neutral genetic sequences are often maintained within various gene pools (to include the human gene pool) and these can experience a random walk through multidimensional sequence space without the influence of natural selection. Sure, there are random walks in biological evolution, but never such that the organism is no longer viable. In any case, your original claim concerned selectable stepwise evolution, not random walks. seanpit: I also do not withdraw my claims about the limited evolutionary potential of any functionally-beneficial system via a Darwinian-like mechansim – be that system comprised of characters in the English-language system or computer codes or a biological system. Great! According to your test, we have to “Select based on changes in beneficial function.” Now, please provide an unambiguous way to determine changes in "beneficial function" so we can try the test what you proposed. Zachriel

Zachriel,

A random walk is not the algorithm under discussion. You should either withdraw your claims about word-space or explain how to “Select based on changes in beneficial function.” Otherwise, your claim is just, well, words.

Why isn’t random walk under discussion? – since random walk is quite real in biological systems and is supposed to be a significant part of biological evolution? I also do not withdraw my claims about the limited evolutionary potential of any functionally-beneficial system via a Darwinian-like mechansim – be that system comprised of characters in the English-language system or computer codes or a biological system. If the mechanism involves random mutations and function-based selection, that mechanism, beyond very low levels of functional complexity, will stall out, in an exponential manner, with each step up the ladder of functional complexity. That is why you base the selection process, within your own evolution algorithms, not on the recognition of the appearance of some kind of new beneficial function, but on template matching to some pre-established sequence irrespective of any change in the underlying function or meaning of the evolving sequence in question. You know as well as I do how to select based on changes in beneficial function. And, you know as well as I do that your evolution algorithms simply don’t do this. seanpit

Zachriel,

Consider a very simple case, with two viable points separated on a vast two-dimensional landscape. A random walk will eventually lead from one viable point to another viable point. However, evolution will never leave the first viable point because every neighbor to it is unviable. Consider now two viable points on a landscape with a vast number of dimensions such that any two points are now a single step away from another. Yes, you may have to test large numbers of neighbors, but you remain viable the entire time during and until you find the other viable point.

This is a mistaken view of biological evolution - or the evolution of any kind of information-based system for that matter. First off, functionally neutral genetic sequences are often maintained within various gene pools (to include the human gene pool) and these can experience a random walk through multidimensional sequence space without the influence of natural selection. This forms the basis of the neutral theory of evolution. Beyond this, it is quite clear that at higher levels of functional complexity it is effectively impossible to evolve between any one island and the next closest in sequence space with a single mutation of any kind – i.e., a single recombination or point mutation. Multiple rather specific mutations would be required to cross the gap distances that exist beyond very low levels of functional complexity. For example, try going from any one of the proposed subsystems in flagellar evolution to the next with just one mutation of any kind. That notion is extremely unlikely, even given the pre-established existence of homologous structures within the gene pool (which do exist as parts of other systems of function), because of the problem that these subsystems would require significant modifications (numerous mutations) before they would actually work, advantageously, as part of the newly evolving system. This assumption of yours is a key to your misunderstanding of the potential of the Darwinian mechanism. The single-step steppingstones that were so common at very low levels of functional complexity become exponentially less and less common with each step up the ladder of functional complexity until they simply disappear, altogether, beyond the level of 1000 specifically arranged amino acid residues. And, of course, at this point evolutionary progress stalls out complexity this side of a practical eternity of time. seanpit

bill cole, Thanks Bill. I appreciate your note. As far as what will happen to the theory of evolution? I'm not sure what will happen because, for now, it is being treated more like a religious philosophy than an empirical science. It is very difficult for people, even scientists, to let go of their closely held religious or philosophical positions. I just don't see it happening very soon... seanpit

Sean Thanks for the post. The sequential space problem is why I became interested in learning about evolution. It originally came up in an origin of life discussion but I soon realized that it left the theory of evolution without a viable mechanism. I have read through the post and watched you patiently work people through it. As the public begins to comprehend the magnitude of this obsticle what do you think will happen to the theory as it is being taught in schools? Will we teach intelligent design or just modify the theory of evolution to an untested hypothesis or stop teaching it all together? bill cole

Mung: Origines asked if you had any proof of the reality Which is why we referred to genes, rather than a schematic metabolism. Genes are connected in sequence space in as many dimensions as there are bases. Zachriel

seanpit: It doesn’t matter if the target is only one step away if there are a million non-target options that are also one step away. They randomly walk off a cliff and survive to reproduce at the bottom of the cliff. It's all uphill from there. Mung

Zachriel: Metabolic genotypes, which consists of on-off switches for chemical reactions. Wagner’s basic concept is the same. Origines asked if you had any proof of the reality, not for a restatement in different words. Mung

Me_Think

I think it will be easier for you to imagine the search space as lattice. if you consider the search space as lattice, then you can easily see that in higher dimensions, the density of lattice increases, so when you do a random walk, you cover more space in less steps. Some thing which was at the edge of the field will now be only a few steps away.

It would be only a "few steps away" (or far fewer steps away in a higher dimensional space) if your random walker wasn't blind - if he could walk directly toward the target. However, that's not how random walks work. Random walks are, well, random. It doesn't matter if the target is only one step away if there are a million non-target options that are also one step away. You see, you are confusing a linear distance with a random walk distance. They aren't the same thing. While the shortest linear distance is indeed significantly decreased within higher spacial dimensions, the random walk distance remains the same. seanpit

Mung: Except Wagner’s not talking about genes, he’s talking about metabolisms. Metabolic genotypes, which consists of on-off switches for chemical reactions. Wagner's basic concept is the same. "Any neighbor would differ in exactly one of these letters, one chemical reaction that may be either present or absent" Zachriel

Zachriel: A gene has neighbors along as many dimensions as it has bases. Except Wagner's not talking about genes, he's talking about metabolisms. Mung

Zachriel:

What happens is that the organism takes a step, and if that step isn’t viable, the organism *remains where it is*. It continues to check each neighboring step until or if it finds one where it is viable.

LoL. This is worthy of laughter and not much else. Mung

Origenes: Assuming arguendo that in hyperspace things are somehow only a few steps away, it follows that that there are an enormous amount of things only a few steps away. And this means that there is an enormous number of ways to not find the right “door”. The answer depends on the structure of the fitness landscape. We've been discussing word-evolution, and with words at least, viable words are much more likely to have viable neighbors than do random letter sequences. Origenes: Finally, is there any proof of the reality of the hyperspace as proposed by Wagner? The dimensions are defined by neighborhoods. A gene has neighbors along as many dimensions as it has bases. Zachriel

Origenes: If 1000 mutations (reactions/letters) is the divide between organism A and organism B, isn’t it so that one needs ± 1000 steps in hyperspace also — given that each step (“visiting a neighbor”) is a change in one reaction/letter? Not quite with regards to evolution. If we were to take a step then another step then another step, a random walk, then yes. However, evolution by natural selection is not a random walk. What happens is that the organism takes a step, and if that step isn't viable, the organism *remains where it is*. It continues to check each neighboring step until or if it finds one where it is viable. Consider a very simple case, with two viable points separated on a vast two-dimensional landscape. A random walk will eventually lead from one viable point to another viable point. However, evolution will never leave the first viable point because every neighbor to it is unviable. Consider now two viable points on a landscape with a vast number of dimensions such that any two points are now a single step away from another. Yes, you may have to test large numbers of neighbors, but you remain viable the entire time during and until you find the other viable point. Zachriel

Me_Think: Some thing which was at the edge of the field will now be only a few steps away.

I'm trying to understand this concept. Wagner writes that neighbors differ in one reaction/sequence/letter.

Likewise, in a 5,000-dimensional cube, each and every metabolism has as many neighbors as there are dimensions, five thousand in all. You can walk from each metabolic text in five thousand different directions, to find one of its five thousand neighbors in a single step. Each of these neighbors differs from the text in exactly one reaction. Either the neighbor has an additional reaction—in this case one entry of the string changes from 0 to 1—or it has one fewer reaction—one entry changes from 1 to 0. (...) Any neighbor would differ in exactly one of these letters, one chemical reaction that may be either present or absent. (It cannot possibly differ in less than that, and if it differed in more, it would no longer be a neighbor.) There is one neighbor that differs in the first letter of this string, another that differs in the second letter, one that differs in the third letter, and so on, until the very last of these letters. [Andreas Wagner]

If 1000 mutations (reactions/letters) is the divide between organism A and organism B, isn't it so that one needs ± 1000 steps in hyperspace also — given that each step ("visiting a neighbor") is a change in one reaction/letter? Put simply, hyperspace or not, the elephant and the mouse are not neighbors. BTW how many of the inhabitants of hyperspace are supposed to be viable organisms? Is the vast majority dead? If not, why not? "There is one neighbor that differs in the first letter of this string, another that differs in the second letter, one that differs in the third letter, and so on, until the very last of these letters.", (Wagner), obviously such a process cannot produce function and a majority of viable organisms. Assuming arguendo that in hyperspace things are somehow only a few steps away, it follows that that there are an enormous amount of things only a few steps away. And this means that there is an enormous number of ways to not find the right "door". Finally, is there any proof of the reality of the hyperspace as proposed by Wagner? For instance, is there anything in the evolution experiment by Richard Lenski that is suggestive of the existence of such a hyper-dimensional space? Origenes

seanpit @ 65

The problem with your argument, you see, is that the ratio of potential targets vs. non-targets doesn’t change with an increase in the dimension of the search space. Therefore, the average number of random walk steps required to find a target remains the same.

I think it will be easier for you to imagine the search space as lattice. if you consider the search space as lattice, then you can easily see that in higher dimensions, the density of lattice increases, so when you do a random walk, you cover more space in less steps. Some thing which was at the edge of the field will now be only a few steps away. Me_Think

seanpit: As I’ve already mentioned, you don’t start on a Frisbee. A random walk is not the algorithm under discussion. You should either withdraw your claims about word-space or explain how to “Select based on changes in beneficial function.” Otherwise, your claim is just, well, words. Zachriel

Zachriel: The landscape has three points of high fitness, and millions of points of zero fitness on a two-dimensional landscape. If we start on a frisbee, evolution by selectable point-mutation can never leave the first frisbee. You will never find another frisbee.

As I've already mentioned, you don't start on a Frisbee. You start in the middle of the field and try to find the first randomly-placed Frisbee by unguided random walk. You will eventually find the first Frisbee using random walk - and the average number of steps for success is not dependent upon the dimensions of the search space nor the size of the steps being taken. I know that your programs don't model random walk, but random walks can and do take place in biological systems. Functionally neutral DNA sequences, for example, can undergo random walk within sequence spaces without being restricted to a particular starting point by natural selection. Also, random walk can occur within neutral nets where various sequences within the name net or "island" cluster produce the same type of function to essentially the same selectable level of functionality.

seanpit: However, as I’ve explained to you in some detail, this effect is not maintained beyond very low levels of functional complexity.

You mean as you repeatedly claimed.

And as I've repeatedly shown you evidence of the appearance of real protein sequence spaces that show an essentially uniform distribution of target islands throughout... more and more so at higher and higher levels of functional complexity.

The random ones and zeros referred to the landscape.

The "ruggedness" in the Cui paper is in reference to changes in function of a 16-character binary sequence in response to various point mutations. A landscape of random binary sequences would be "flat" with respect to function - i.e., functionally neutral rather than functionally rugged. seanpit

seanpit: However, as I’ve explained to you in some detail, this effect is not maintained beyond very low levels of functional complexity. You mean as you repeatedly claimed. seanpit: Come on now, a random sequence of zeros and ones is unlikely to have a functional advantage over another random sequence of zeros and ones. Huh? The random ones and zeros referred to the landscape. seanpit: Again, my claim is and has always been that any model that is based on random mutations and function-based selection will stall out, in an exponential manner, beyond very low levels of functional complexity. You claim it applied to word-space, but you can't or won't unambiguously define "functional complexity" for word-space. You should either withdraw your comments about word-space or explain how to "Select based on changes in beneficial function." Otherwise, your claim is just, well, words. seanpit: It would take a long time, but not forever. The landscape has three points of high fitness, and millions of points of zero fitness on a two-dimensional landscape. If we start on a frisbee, evolution by selectable point-mutation can never leave the first frisbee. You will never find another frisbee. Zachriel

Zachriel,

But if we use a stepwise selection algorithm, how many steps would it take? Forever.

It would take a long time, but not forever. The point is that changing the dimensions doesn't change the average time to success.

Now think about what happens if we increase the number of dimensions to, say, 27,878,400,000,000. The results are not the same, and we haven’t even considered possible structuring, rather than a random distribution.

Yes, they are the same. Regardless of the number of dimensions you use, given the same ratio of potential targets vs. non-targets, the average random walk time will not change. The only way to change the average random walk time is with the addition of some non-random "structure" to the location of our targets within the search space. That's the only way to significantly reduce the average random walk time for a given ratio of targets vs. non-targets. Your only problem is, of course, that the structure necessary to prevent an exponential increase in the average random walk time, with each step up the ladder of functional complexity, just isn't there. Your necessary structure starts to break up, early on, so that by the time you're at a level of say, 1000 saars, your structure is so fragmented that the average random walk time is pretty close to how it would be given a truly random distribution of targets within this higher level search space. seanpit

Zachriel,

Even simple statistical tests demonstrate that English is not evenly distributed in character-space.

Indeed! And the same is true for proteins as well – at very low levels of functional complexity. However, as I’ve explained to you in some detail, this effect is not maintained beyond very low levels of functional complexity. The “distribution” of beneficial sequences does in fact take on a fairly uniform appearance within sequence space. You just refuse to acknowledge the evidence is all…

A random landscape of ones and zeros is not neutral to function, but is characterized by a multitude of fitness peaks.

This is like saying that a landscape of Zs and Ts is “not a neutral function”. Come on now, a random sequence of zeros and ones is unlikely to have a functional advantage over another random sequence of zeros and ones. That means, of course, that these sequences are “neutral” with respect to their functional advantage.

It’s YOUR citation, and concerned whether recombination was effective in evolutionary search. The paper directly contradicted your position. So now you say it is irrelevant.

I didn’t say it was “irrelevant”. It is quite relevant in that it shows how recombination mutations vs. point mutations work in different situations. If the fitness landscape has sharp increases or decreases in sequence function, recombination mutations don’t provide a significant advantage. However, if the gradient of functionality is fairly smooth, obviously recombination mutations would be able to scale the smooth slopes more efficiently compared to point mutations acting alone. Also, this paper, along with the others I cited for you, show that as the ratio of potential targets vs. non-targets gets very small, that the beneficial islands (including their neutral nets) become more and more isolated from each other and become more and more uniformly distributed within sequence space – just like I’ve been trying to explain to you. Of course, you yourself agree that given a situation where targets are extremely rare within sequence space (like 1 in 1e600 or so), and given that they are fairly evenly distributed within that space, that evolutionary progress would clearly stall out. Yet, you cling to this idea that the structure that exists at very very low levels of functional complexity also exists at very high levels of functional complexity. The problem, of course, is that the available evidence does not support this claim of yours.

You didn’t like our model. You said to “Select based on changes in beneficial function,” but you haven’t been able to provide an unambiguous measure of “beneficial function” with regards to the English language.You made a claim. You point in the general direction, but haven’t been able to support it.

Again, my claim is and has always been that any model that is based on random mutations and function-based selection will stall out, in an exponential manner, beyond very low levels of functional complexity. That’s my claim and that claim is very well supported by the data I’ve already provided you here and on my website. To get you to at least start to think about this concept I told you to think about the English language system and how the ratio of potentially beneficial vs. non-beneficial significantly changes as you move up the ladder of functional complexity. You took this to mean that I saw no structure at low levels of functional complexity within the English language system. That’s nonsense. Of course there is some structure at the lowest levels of functional complexity – both in the English language system as well as in protein-based systems of function. I’ve explained this over and over again – even back in 2004 when we started this conversation. Where we disagree is in your fantastic notion that this structure remains at higher and higher levels so that no exponentially stalling effect is ever realized – regardless of the level of functional complexity under consideration. Your argument is that somehow, somewhat, the underlying structure of sequence space will, without fail, arrange the extremely rare steppingstones in nice little closely spaced rows that are easily found by random large steps into sequence space. Now, I’m sorry that your programs fail to use the Darwinian mechanism. But, they clearly don’t use it. That’s not my problem. I didn’t create these programs of yours. If I knew how to create a computer program that could use the Darwinian mechanism to truly create something like the works of Shakespeare, I’d be a very wealthy man indeed! My argument, of course, is that such a computer program cannot be produced – that intelligent design will always be required to produce meaningful sequences beyond very low levels of meaningful/functional complexity. It just doesn’t happen without intelligent design – as your own website and algorithms demonstrate quite nicely. If anything, your efforts only serve to highlight the truth of my position. If you want to actually demonstrate the creative potential of the Darwinian mechanism, you have to think of a scenario where sequences compete for resources over time and that this competition leads to higher and higher levels of functional complexity. That’s how the Darwinian mechanism is supposed to work. Numerous efforts have been published along these lines. The problem, of course, is that none of them actually generate qualitatively novel functional complexity beyond very very low levels. It just doesn’t happen – not in computer simulations and certainly not in biological systems. There simply are no such examples anywhere where some new system of function is produced that requires a minimum of more than 1000 specifically arranged characters to perform its novel function. And, statistically, it is extremely unlikely to ever happen this side of a practical eternity of time. But, go ahead, prove me wrong. In the mean time, I wouldn’t keep forwarding your current algorithms as having anything to do with the Darwinian mechanism. It makes you look like you don’t know what you’re talking about . . . seanpit

seanpit: How many steps, on average would it take a random walker to find any one of the Frisbees? – starting from the center of the field? But if we use a stepwise selection algorithm, how many steps would it take? Forever. Now think about what happens if we increase the number of dimensions to, say, 27,878,400,000,000. The results are not the same, and we haven't even considered possible structuring, rather than a random distribution. Zachriel

Me_Think You don’t seem to understand how increasing the dimensions of a search space really works when it comes to random walks and the odds of successfully finding a rare target within a certain number of options. Consider our large field example again as an illustration. Our field measures 1000 miles on each side for a total of 1,000,000 square miles, which equals 27,878,400,000,000 square feet. Say that we have three Frisbees in the field that each measure 1 square foot scattered randomly within this space. How many steps, on average would it take a random walker to find any one of the Frisbees? – starting from the center of the field? Now, turn these 2.78e13 squares into cubes. Originally each side of our field measured 5,280,000 feet. Now, each side of our cube measures 30,321 feet (just 5.74 miles). So, if I understand you correctly, your argument is that this reduced distance per axis in a higher dimension would significantly reduce the average random walk time to find one of the target positions in that space. Let me ask you, how many potential locations are there in the cube vs. the number of potential locations in the field? It’s the same number, right? What about the ratio of targets to non-targets in the cube vs. the field? It’s also the same. The problem with your argument, you see, is that the ratio of potential targets vs. non-targets doesn’t change with an increase in the dimension of the search space. Therefore, the average number of random walk steps required to find a target remains the same. Nothing changes by moving up to a higher dimension as far as average random walk “distances” or “steps” are concerned. seanpit

seanpit: And, that is precisely why the Darwinian mechanism stalls out, in an exponential manner, as this lower-level structure starts to break down and the potential target islands become significantly more and more isolated and randomly uniform in their apparent distribution within higher and higher levels of sequence spaces. That's your claim, however, even simple statistical tests demonstrate that English is not evenly distributed in character-space. seanpit: The concept of “ruggedness” is based, in this case, on the degree of increase or decrease in sequence function – described as “fitness” in the paper (as an increase or decrease in the “mortality” or “internal free energy” of the sequence). Yes, it's based on fitness in relation to point-mutations; however, ruggedness is contrasted with smoothness: "exploration is more efficient when the evolutionary landscape is smooth (small ?), but as ruggedness or the average selection gradient increases (larger ?), exploration becomes sluggish. seanpit: A random landscape of “zeros and ones” would be neutral with respect to function and therefore perfectly flat – not “rugged” with respect of function. No. A random landscape of ones and zeros is not neutral to function, but is characterized by a multitude of fitness peaks. seanpit: Consider also that the Cui paper was only talking about short sequences with a chain length of 18 with a total sequence space size of 2^18 = 262,144. It's YOUR citation, and concerned whether recombination was effective in evolutionary search. The paper directly contradicted your position. So now you say it is irrelevant. seanpit: Wait a minute. You’re the one who hasn’t modeled function-based selection with your own algorithms – and you’re trying to blame me for that? You didn't like our model. You said to “Select based on changes in beneficial function," but you haven't been able to provide an unambiguous measure of "beneficial function" with regards to the English language. You made a claim. You point in the general direction, but haven't been able to support it. Zachriel

There's a serious mistake @ 62. I am leaving it there so you can find out (Just another way of saying, I didn't edit in time :-) ! Me_Think

seanpit @ 56

Why not? Because, it doesn’t matter that the starting point in hyperdimenstional sequence space is surrounded by large numbers of neighbors – or that these next-door-neighbors are in turn surrounded by large numbers of neighbors (so that within a very short Levenshtein distance the total number of neighbors is absolutely enormous. None of that helps find a target sequence any faster via a random search algorithm given a particular ratio of uniformly distributed targets among non-target options.

That's not true for hyperdimensions. Take the frisbee example @ 57- the search area for the 1000 miles field is about 1.609344X10^9 square meters but in hyperdimension of 4, the length of side of the hypercube will be just 200 meter ! (1.609E+9 = a^4, Solving for a you get about 200.). In 10 dimension, it's just 8.33 meter! So any random walk will find the frisbee quickly. Me_Think

Zachriel,

If words were random letters, then word evolution wouldn’t work… Apparently one in 10^20 is not sparse enough for you.

Yes, it would work at very low levels – just not as well. And, that is precisely why the Darwinian mechanism stalls out, in an exponential manner, as this lower-level structure starts to break down and the potential target islands become significantly more and more isolated and randomly uniform in their apparent distribution within higher and higher levels of sequence spaces. And, no, a ratio of 1 in 1e20 isn’t remotely “sparse enough” to stall evolutionary progress for a reasonably sized population with a reasonable reproductive rate and mutation rate. When you get to the level of 1000 saars, the ratio is more like 1 in 1e600. At this point, the structure needed to support evolutionary progress simply isn’t there.

A vast flat zero landscape with a small gentle hill in the middle is sparse, but not rugged. A random landscape of zeros and ones is rugged, but not sparse.

No. The concept of “ruggedness” is based, in this case, on the degree of increase or decrease in sequence function - described as “fitness” in the paper (as an increase or decrease in the “mortality” or “internal free energy” of the sequence). A random landscape of “zeros and ones” would be neutral with respect to function and therefore perfectly flat – not “rugged” with respect of function. In a “rugged” landscape, “a few random mutations randomize structure ensembles” – and disrupt function (Fontana et al., 1993; Schuster et al., 1994; Bomberg-Bauer, 1996; Tacker et al., 1996; Renner and Bomberg-Bauer, 1997). So, as either the ruggedness of the fitness landscape increases or the structure of the fitness landscape decreases, the helpfulness of recombination mutations vs. point mutations also decreases. And, with each step up the ladder of functional complexity, this is exactly what happens. The structure or location of potentially beneficial island targets becomes more and more uniformly random in appearance. Consider also that the Cui paper was only talking about short sequences with a chain length of 18 with a total sequence space size of 2^18 = 262,144. That’s a tiny sequence space compared to my proposed limiting sequence space size of 20^1000 sequences = 1e1301 sequences. Yet, there the features of this limited model that are interesting - despite the very small size of it's sequence space. “In analogy to protein families, nets [in the HP model] are dense and well separated in sequence space... [Also], although only a small fraction of sequence space yields uniquely folding sequences, sequence space is occupied nearly uniformly. No ‘higher order’ clustering (i.e., except the trivial case of the homologous sequences) is visible… Studies on shape space covering show that it is very difficult to convert one structure into another by a few point mutations.” http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1181141/pdf/biophysj00028-0151.pdf The same thing is true for 100aa proteins when it comes to sequences that can at least produce “stable folds” (which are required for functional proteins). Such sequences are rare and show an essentially random distribution within sequence space. http://www.sciencedirect.com/science/article/pii/S1359027897000370 And, when you’re talking about 1000aa sequence space, the number of stable/viable 1000aa sequences in sequence space is around 1e707. Given the size of sequence space at this level is 20^1000, the ratio of viable vs. non-viable is ~1e-594. And, this isn't the worst of it. This number is "further reduced by the dual requirements of stability and kinetic accessibility and the number of sequences that are biologically competent.” In short, the ratio of 1e-594 potential targets vs. non-targets is being generous for 1000aa sequence space. http://www.detectingdesign.com/flagellum.html#Few

seanpit: in biology, as in the English language system, sequences can be functionally neutral, detrimental, or beneficial relative to what came before. Surely you understand that – right?

Sure. And that’s what we’re asking. In order to test your claim about word evolution, you need to “Select based on changes in beneficial function,” that is, you need an unambiguous function that returns the difference in what you call beneficial function. You haven’t been able to do that, so your claim is essentially undefined.

Wait a minute. You’re the one who hasn’t modeled function-based selection with your own algorithms – and you’re trying to blame me for that? The concepts of beneficial, detrimental, and neutral have been defined, in literature, with respect to evolving sequences. You just haven’t used them in your algorithms is all. Rather, you base your selection on template matching – not any kind of change in beneficial function. I’m sorry, but that’s not my fault nor is it a problem that the relevant concepts haven’t been adequately defined for you. They have been very clearly defined. seanpit

seanpit: You admit, now, that if the targets are in fact randomly distributed, uniformly, within sequence space that there would be no advantage for recombination vs. point mutations? Zachriel @25: Sure (density descreases). However, the space is not random, but highly structured. If words were random letters, then word evolution wouldn't work. seanpit: the only argument you really have left is your argument that the targets are not randomly distributed, but are lined up and clustered in related groups – – at all levels of functional complexity! Of course they're not randomly distributed. Even a simple statistical test of the English language shows letter sequences are not evenly distributed. seanpit: while this is true at low levels of functional complexity, it becomes less and less true with each step up the ladder of functional complexity. That's your claim. Apparently one in 10^20 is not sparse enough for you. seanpit: At this point, the gaps between one island and the next closest island start to grow, in a linear manner, with each additional step up the ladder. That's your claim. We can use intelligence to test this claim, because your claim isn't that we can't find the link, but that such a link doesn't exist. seanpit: The only argument you must present, at this point, is to continue to argue, in face of the overwhelming evidence ... You haven't provided any evidence. That requires looking at the actual landscape in question, something you haven't been able to unambiguously define. seanpit: Ruggedness is produced by sparseness. That is incorrect. A vast flat zero landscape with a small gentle hill in the middle is sparse, but not rugged. A random landscape of zeros and ones is rugged, but not sparse. seanpit: There is no contradiction to my position beyond very low levels of functional complexity – as I’ve already explained. Of course it directly contradicts your claim that recombination is irrelevant. “Efficient structural exploration requires intermediate nonextreme ratios between point-mutation and crossover rates." seanpit: in biology, as in the English language system, sequences can be functionally neutral, detrimental, or beneficial relative to what came before. Surely you understand that – right? Sure. And that's what we're asking. In order to test your claim about word evolution, you need to “Select based on changes in beneficial function," that is, you need an unambiguous function that returns the difference in what you call beneficial function. You haven't been able to do that, so your claim is essentially undefined. Zachriel

Zachriel,

seanpit: There simply is no advantage for one random search algorithm over any other when it comes to finding rare targets within unknown locations within a sequence space of options.

That’s assuming the targets are randomly distributed, which they clearly are not.

It seems like we’re finally getting somewhere. You admit, now, that if the targets are in fact randomly distributed, uniformly, within sequence space that there would be no advantage for recombination vs. point mutations? That’s good. Now, the only argument you really have left is your argument that the targets are not randomly distributed, but are lined up and clustered in related groups - - at all levels of functional complexity! As I’ve already pointed out in fair detail in this thread, while this is true at low levels of functional complexity, it becomes less and less true with each step up the ladder of functional complexity. With each step up the ladder, the bridges of closely-spaced steppingstones within higher and higher level sequence spaces become more and more narrowed until, not too far up the ladder, they start to snap and break down completely. At this point, the gaps between one island and the next closest island start to grow, in a linear manner, with each additional step up the ladder. By the time you get to the level of 1000 saars, the minimum likely non-beneficial gap distance is over 50 mutational changes wide. At this point there is in fact a fairly uniform distribution of potential target sequences throughout sequence space. While island clusters of sequences with a given function still exist within high-level sequence space, the distribution of these island clusters relative to each other takes on the appearance of a random uniform distribution. That is why, at this point, there really is no significant advantage to recombination vs. point mutations – as you yourself would agree given such a situation. The only argument you must present, at this point, is to continue to argue, in face of the overwhelming evidence, that regardless of the level of functional complexity, even beyond the level of 1000 saars, that potentially beneficial sequences within these higher level sequence spaces are still just as lined-up and clustered as they were at very low levels of sequence space. That’s really the only argument you have left in order for the odds to work out for continued evolutionary progress, at the same rate, without an exponential stalling effect while moving up the ladder. The problem is, this notion of yours simply doesn’t hold true in real life. There simply is no web of narrow bridges of nice lined-up steppingstones at these higher levels of functional complexity. Potentially beneficial targets really do have a uniform distribution within higher levels of sequence space. That’s why your algorithms cannot work beyond very low levels of functional complexity without resorting, as you have consistently done on your website, to either intelligent manipulation or template matching to some pre-determined target sequence where selections are made in each generation without respect to changes in function. However, if you base your selection on improvements in beneficial function, your Darwinian mechanism will in fact stall out at the lowest levels of functional complexity - because sequence space does in fact take on the appearance that I describe and does not resemble what you describe.

The quoted section referred to ruggedness, not sparseness.

Ruggedness is produced by sparseness. A nice tight sequence of steppingstones where the next steppingstone in the sequence is just a bit more beneficial than the one that came before would produce a nice smooth slope to the landscape. However, a landscape that is largely flat and scattered with numerous sinkholes, comprised primarily of neutral or detrimental sequences, with only occasional clusters of beneficial sequences with sharp peaks, would produce a very “rugged” landscape. You yourself have already agreed that such a situation would in fact make point mutations and recombination mutations essentially equivalent. So, I don’t understand why you’re even arguing this point?

“Efficient structural exploration requires intermediate nonextreme ratios between point-mutation and crossover rates,” directly contradicting your position, and directly supporting ours.

There is no contradiction to my position beyond very low levels of functional complexity – as I’ve already explained. The bridges break down very quickly as you move up the ladder of functional complexity until the appearance of the islands in sequence space does in fact take on a random uniform appearance without the otherwise flatness of the rest of the vastness of the ocean of non-beneficial options within these higher-level spaces.

Given two sequences, you claim that we have to “Select based on changes in beneficial function.” If you can’t do that, then your claim is essentially undefined.

I’m sorry, but in biology, as in the English language system, sequences can be functionally neutral, detrimental, or beneficial relative to what came before. Surely you understand that – right? I’m not sure what you don’t understand about this? seanpit

seanpit: Not true. That's your claim. seanpit: What are the odds that a large random leap into sequence space will happen to land on a rare target vs. the odds that a small step into sequence space will happen to land on a rare target? It depends on the fitness landscape, something you can't seem to be able to provide. seanpit: What are the odds of winning the lottery by changing one number in your starting sequence of numbers vs. changing many of them at the same time? Nothing! A sixteen-letter word is rarer than a lottery ticket (one in 10^20), yet evolution works quite efficiently at finding such words. seanpit: There simply is no advantage for one random search algorithm over any other when it comes to finding rare targets within unknown locations within a sequence space of options. That's assuming the targets are randomly distributed, which they clearly are not. seanpit: As the ratio of targets vs. non-targets gets is significantly reduced No. The quoted section referred to ruggedness, not sparseness. seanpit: Yan Cui, Wing Hung Wong, Erich Bornberg-Bauer, Hue Sun Chan "Efficient structural exploration requires intermediate nonextreme ratios between point-mutation and crossover rates," directly contradicting your position, and directly supporting ours. seanpit: Two randomly generated sequences of characters (English letters or amino acid residues) are most likely to be equally meaningless with respect to function – and therefore produce a neutral selective advantage relative to each other. Selection between two such random sequences would therefore be random as well. Probably true, but not an unambiguous measure. Given two sequences, you claim that we have to "Select based on changes in beneficial function.” If you can't do that, then your claim is essentially undefined. Zachriel

Zachriel,

We’re talking about the wordscape. Recombination makes a significant difference in the behavior of evolutionary search of the wordscape.

Not true. Recombination makes no significant difference in finding target sequences more effectively (compared to point mutations) when targets are very rare relative to non-targets and there is a uniform distribution of targets within sequence space (i.e., a ratio of targets vs. non-targets of 1 in 1e500 sequences or so).

That’s your claim. Now how do you intend to support it?

Do the math yourself if you don’t believe me. What are the odds that a large random leap into sequence space will happen to land on a rare target vs. the odds that a small step into sequence space will happen to land on a rare target? It's essentially the same odds of success. What are the odds of winning the lottery by changing one number in your starting sequence of numbers vs. changing many of them at the same time? Nothing! There simply is no advantage for one random search algorithm over any other when it comes to finding rare targets within unknown locations within a sequence space of options. Imagine, if you will, a large flat square field measuring 1000 miles on each side. Now, say that there are 3 Frisbees measuring 1 foot in diameter randomly distributed out there somewhere in this large field. Now, you get 10 blindfolded men to search for these Frisbees. They have the option of taking small steps of 6 inches per step, larger steps of 3 feet per step, even larger jumps of 10 feet per jump, or even huge jumps of one mile per random jump. Which type of random walk, using small or larger steps, will be most effective in finding the Frisbees in this field? My conclusions here are also backed up by a paper published by Cui et. al. published in PNAS in 2002. Toward the end, Cui argues that:

“The benefit provided by nonhomologous recombinations [compared to point mutations] decreases as the ruggedness of the fitness landscape increases; and a very rugged landscape provides only marginal benefit compared to a less rugged landscape… When the landscape is rugged, the number of sequence explored by point mutations alone is comparable to that explored by point mutations plus [non-homologous] crossovers. This is because point mutations are more effective in finding a low-mortality area from an already well populated spot nearby, whereas when the landscape is rugged many crossover offspring are likely to end up at high-mortality spots.” Yan Cui, Wing Hung Wong, Erich Bornberg-Bauer, Hue Sun Chan, Recombinatoric exploration of novel folded structures: A heteropolymer-based model of protein evolutionary landscapes., PNAS, Vol. 99, Issue 2, 809-814, January 22, 2002.

So, there you have it. As the ratio of targets vs. non-targets gets is significantly reduced, the potential benefits of recombination mutations are also significantly reduced – to the point where there simply is no significant advantage between recombination vs. point mutations. Again, sit down and do the math for yourself. If you don’t agree, actually show me why the math favors larger steps in sequence space vs. smaller steps… beyond very low levels of functional complexity.

That’s not an unambiguous measure of fitness [reproductive fitness in a given environment]. Given two arbitrary sequences, you have to provide their relative fitness as you made explicit in your statement to “Select based on changes in beneficial function”.

Two randomly generated sequences of characters (English letters or amino acid residues) are most likely to be equally meaningless with respect to function – and therefore produce a neutral selective advantage relative to each other. Selection between two such random sequences would therefore be random as well. As I explain on my website, what’s the meaningful difference between: “quiziligook” vs. “quiziliguck”? Nothing – right? Therefore, selection between two such sequences would not show any preference, but would be entirely random. seanpit

Me_Think,

This particular RNA enzyme happens to have 129 neighbors, and because we can compute their shapes, we can determine that there are forty-six new shapes in this neighborhood. That’s the number of shapes evolution can explore without genotype networks. And with them? If we only step to the text’s neutral neighbors—those with the same hammerhead shape—and determine the shape of all their neighbors, we already find 962 new shapes. And if we just walk one step further, to those neighbors’ neutral neighbors, we find 1,752 new shapes. Just two steps along this ribozyme’s genotype network, we can access almost forty times more shapes than in its immediate vicinity. The genotype network of the hammerhead shape of course extends much further than just two steps, and it has more than 1e19 members

Unfortunately, this doesn’t help solve the problem. Why not? Because, it doesn’t matter that the starting point in hyperdimenstional sequence space is surrounded by large numbers of neighbors – or that these next-door-neighbors are in turn surrounded by large numbers of neighbors (so that within a very short Levenshtein distance the total number of neighbors is absolutely enormous. None of that helps find a target sequence any faster via a random search algorithm given a particular ratio of uniformly distributed targets among non-target options. It doesn’t change the fact that with each linear increase in the Levenshtein distance that the target is located from the starting point, the average time to success increases exponentially. Do the math. Set up a program as see that I’m correct here. The basic problem, you understand, is that as the minimum Levenshtein distance increases linearly, the total number of possibilities increases exponentially. That means, of course that the total number of non-targets that could be searched increases at an exponentially greater rate as compared to the potential targets sequences. And, quite clearly, that means that the average time to successfully finding the target sequences, via any kind of random walk, increases exponentially. I’m sorry, but you haven’t solved the problem here – not by a long shot. seanpit

seanpit: If the landscape where set up so that the potential targets where set up in nice little rows of closely-space steppingstones, you’d be right. We're talking about the wordscape. Recombination makes a significant difference in the behavior of evolutionary search of the wordscape. seanpit: So, at this point, at higher levels of functional complexity, do recombination mutations provide of substantive advantage over point mutations? – so that there is no exponential decline in the success rate over a given span of time? The answer to that question is no. There simply is no significant statistical advantage regardless of the type of random mutations employed. That's your claim. Now how do you intend to support it? seanpit: As I’ve already mentioned, for Darwinian evolution the answer is simple – an increase in relative reproductive fitness compared to one’s peers. That's not an unambiguous measure of fitness. Given two arbitrary sequences, you have to provide their relative fitness as you made explicit in your statement to "Select based on changes in beneficial function". It doesn't seem you can do this, so your claim is unsupported. Zachriel

seanpit @ 41

This makes absolutely no difference for a random search algorithm as compared to a two or three dimensional search. The odds of success still decrease exponentially as the minimum Levenshtein distance increases linearly. Beyond this, none of the papers you reference explain how random search algorithms within hyperdimensional space can cover a linearly expanding Levenshtein distance between strings or character sequences without an exponential increase in required time…

This has been discussed long ago. Reposting from old thread: Imagine a solution circle (the circle within which solution exists) of 10 cm inside a 100 cm square search space. The area which needs to be searched for solution is pi x 10 ^2 = 314.15 The total Search area is 100 x 100 = 10000. The % area to be searched is (314.15/10000) x 100 = 3.14% In 3 dimensions,the search area will be 4/3 x pi x 10^3 Area to search is now cube (because of 3 dimensions) = 100^3. Thus the % of area to be searched falls to just 4188.79/100^3 = 0.41 % only. Hypervolume of sphere with dimension d and radius r: [Pi]^(d/2))/[CapitalGamma](d/2+1) HyperVolume of Cube = r^d At 10 dimensions, the volume to search reduces to just: 0.000015608 % But in nature, the actual search area is incredibly small. As wagner points out in Chapter six of his book (Arrival of the Fittest) :

In the number of dimensions where our circuit library exists—get ready for this—the sphere contains neither 0.1 percent, 0.01 percent, nor 0.001 percent. It contains less than one 10^ -100th of the library

The library that wagner talks about is based on actual metabolic pathways. There are 5,500 metabolic pathways. You can explore all the pathways at biocyc.org These are represented by the hypothetical library (Just like landscapes in Dembski, Axe papers). The library is an analogy – See Chapter Three Notes: This analogy is inspired by a famous short story of the Argentine author Jorge Luis Borges entitled “The Library of Babel” (Spanish original: “La biblioteca de Babel”), published in English translation in Borges (1962). The idea behind this short story, however, predates Borges. It has been used by many other authors, including Umberto Eco and Daniel Dennett Here’s another example of how RNA enzyme called hammerhead ribozyme search is made easy:

This particular RNA enzyme happens to have 129 neighbors, and because we can compute their shapes, we can determine that there are forty-six new shapes in this neighborhood. That’s the number of shapes evolution can explore without genotype networks. And with them? If we only step to the text’s neutral neighbors—those with the same hammerhead shape—and determine the shape of all their neighbors, we already find 962 new shapes. And if we just walk one step further, to those neighbors’ neutral neighbors, we find 1,752 new shapes. Just two steps along this ribozyme’s genotype network, we can access almost forty times more shapes than in its immediate vicinity. The genotype network of the hammerhead shape of course extends much further than just two steps, and it has more than 1019 members

P.S: Dimensions are mathematical representation of the structure/process features under study. It has got nothing to do with spatial dimension. I can represent the search hills in search landscape in “height and coordinate dimensions” too. Note that polytope naturally forms network. (Hyper cube is family of polytope). Me_Think

Zachriel:

Then let’s avoid getting sidetracked on that issue again. Evolution, for our purposes, is defined as a population that undergoes random point-mutations and random recombination, with selection for properly spelled words.

That is artificial selection which is very different from natural selection. So thank you for admitting that your model doesn't apply to what Sean is talking about. Virgil Cain

Zachriel,

seanpit: In short, please do explain to me how completely random recombination mutations can somehow maintain the odds of success, without a significant decline in the success rate, for a population maintained at a constant size – all while the ratio of targets vs. non-targets decreases at an exponential rate?

It depends on the landscape. We know that for words evolution with recombination works much differently than evolution without recombination.

If the landscape where set up so that the potential targets where set up in nice little rows of closely-space steppingstones, you’d be right. If the Levenshtein distances were consistently small, it would make a big difference. However, if the landscape were one where the targets had an apparently random essentially uniform distribution, you’d be wrong. And, this is what the landscape of functional sequence space looks like, more and more, with each step up the ladder of functional complexity. So, at this point, at higher levels of functional complexity, do recombination mutations provide of substantive advantage over point mutations? – so that there is no exponential decline in the success rate over a given span of time? The answer to that question is no. There simply is no significant statistical advantage regardless of the type of random mutations employed. Statistically, the only difference is that point mutations cover regions closer to the starting point compared to recombination or indel mutations that take larger steps into the surrounding sequence space. That’s the only real statistical difference. What this means is that if the targets are close to home, then they will be found more often by point mutations as compared to recombination mutations involving longer sequences. However, once you start talking about longer and longer Levenshtein distances to the next closest target, the odds of success start to become more and more similar between random walks based on point mutations and recombination mutations. Pretty soon, the ratio between targets and non-targets gets so low that there really is no statistical advantage between various kinds of random walks or search algorithms when it comes to finding such rare targets that are randomly distributed in an essentially uniform manner within a very large search space. It simply doesn’t matter anymore, statistically, if you take large steps or small steps. The odds of success remain essentially the same.

seanpit: 2) Select based on changes in beneficial function

How do you intend to unambiguously define “beneficial function” for longer sequences of letters so as to test your claim?

As I've already mentioned, for Darwinian evolution the answer is simple – an increase in relative reproductive fitness compared to one’s peers. The same could be true for any other system of functional information where a beneficial goal might be defined where any improvement in achieving this goal would give an advantage and therefore be preferentially selected to populate the next generation. If you aren’t modeling some kind of functional/meaningful advantage in your algorithms, you’re not modeling the Darwinian mechanism. It’s really as simple as that. And, so far, your algorithms are actually based on intelligent design or template matching – not the Darwinian mechanism where selection is only based on improved beneficial function. seanpit

seanpit: In short, please do explain to me how completely random recombination mutations can somehow maintain the odds of success, without a significant decline in the success rate, for a population maintained at a constant size – all while the ratio of targets vs. non-targets decreases at an exponential rate? It depends on the landscape. We know that for words evolution with recombination works much differently than evolution without recombination. seanpit: 2) Select based on changes in beneficial function How do you intend to unambiguously define "beneficial function" for longer sequences of letters so as to test your claim? Zachriel

Zachriel, The Odds of Success for Recombination Mutations: In short, please do explain to me how completely random recombination mutations can somehow maintain the odds of success, without a significant decline in the success rate, for a population maintained at a constant size - all while the ratio of targets vs. non-targets decreases at an exponential rate? Please explain the math behind this notion of yours and how it can actually work? - given that the minimum Levenshtein distance between your starting point and the next closes target is getting longer and longer. Really, please explain the basis for your assumption of steady odds here and the lack of any significant corresponding increase in non-beneficial options. I'd be most interested. (Hint: To understand the math more easily, try reducing the steady state population size to two or three and the reproductive rate to two per individual per generation. While increasing the population size and/or reproductive rate helps for a while, there is only so much that this can achieve before even a very large populations with very high reproductive rates can no longer keep up with an additional step up the ladder of functional complexity. The pattern of an exponential increase in the time required will set in at this point. Not only that, but populations that have low reproductive rates and higher mutation rates will actually devolve from their starting point fitness level. They will not even stay neutral much less "evolve" higher level systems of function over time). seanpit

Zachriel,

That is demonstrably incorrect. It [recombination mutations] makes a significant difference with word-evolution. You are simply wrong.

Where's your "demonstration" beyond very low levels of functional complexity? I'm telling you that the odds are not significantly improved by recombination mutations when it comes to finding potential targets at higher levels of functional complexity (i.e., when it comes to finding functional sequences that require at least 1000 specifically arranged characters to work in a selectably beneficial manner). The odds of success for recombination mutations at such higher levels is essentially the same as it is for point mutations alone. In other words, recombination mutation don't solve the problem of an exponential increase in the time required for success as one moves up the ladder of functional complexity. The very same problem remains regardless of the types of random mutations you're using. You simply have not "demonstrably solved" this problem! - not even close! If anything, your Phrasenation program proves my point here! Now, what you have very clearly demonstrated on your website is that evolution works just fine if you throw in a little intelligent design or a bit of template matching to help out your recombination mutations - beyond very low levels of functional complexity. That makes it very easy to get across very large gaps (Levenshtein distances) of non-beneficial function in short order. Of course, if you just stick with the Darwinian mechanism things don't work so well beyond very low levels of functional complexity - regardless of the types of random mutations you decide to use in your search algorithms. As far as your other repeated questions, I've already responded in some detail. Why not go back and read what I wrote when you asked these questions the first time? Now, why not substantively address the questions I've presented to you? - instead of simply ignoring the main points I've presented regarding why your arguments and algorithms don't truly reflect the limitations of the Darwinian mechanism? seanpit

seanpit: As I’ve already explained to you in fair detail (and even greater detail on my website), while genetic recombination is real ... Great! Then let's avoid getting sidetracked on that issue again. Evolution, for our purposes, is defined as a population that undergoes random point-mutations and random recombination, with selection for properly spelled words. seanpit: it doesn’t help you solve the problem if you are using truly mindless random mutations and function-based selection. The odds of success are essentially unchanged regardless of the type of mutations you use. That is demonstrably incorrect. It makes a significant difference with word-evolution. You are simply wrong. --------------------- In any case, let's return to this: seanpit: 1) Generate truly random mutations (point, indel, recombination, etc) that aren’t limited to determining and clipping out intact words or select “phrases” (something that doesn’t happen in real life). Been there, done that. seanpit: 2) Select based on changes in beneficial function – not template-matching which doesn’t happen in real life. How do you intend to do that? It’s your claim, after all, you are trying to prove. seanpit: 3) Have a reasonable maximum steady state population size with a reasonable reproductive rate and mutation rate. In other words, old sequences must “die off” as fast as new ones are “born” so that the overall population size remains the same. Been there, done that. seanpit: If you actually model how the Darwinian mechanism really works, you will quickly discover that your neat little pathways of shortly-spaced steppingstones break apart and become widely separated very quickly as you move up the ladder of functional complexity beyond your short little sequences. How do you intend to do that? It’s your claim, after all, you are trying to prove. Zachriel

Virgil,

Sean, I am of the type that says organisms were designed to evolve and evolve by design. Meaning most genetic changes are directed by the organisms’ programming in response to some cue(s), environmental or internal.

The problem here is that there are very clear limitations to how much organisms can change respond to environmental changes via Mendelian variation or other forms of low-level evolution. However, Darwin argued along the lines of Zachriel that there are no such limitations to what the Darwinian mechanism can achieve. The problem, of course, is that the Darwinian mechanism is very clearly limited to very very low levels of functional complexity. So, whatever design there may be that allows for variation, such design was very limited and does not allow for the Darwinian story of origins or the development of qualitatively novel functional systems, beyond very low levels of functional complexity, within gene pools that were not already there - pre-created within the original parental gene pool. seanpit

Sean, I am of the type that says organisms were designed to evolve and evolve by design. Meaning most genetic changes are directed by the organisms' programming in response to some cue(s), environmental or internal. Virgil Cain

Virgil Cain,

And as I have explained to you no one can say if recombination is a happenstance occurrence. Most likely it is an intelligently designed feature that allows for genetic diversity in a short time.

This depends upon what type of genetic "recombination" you're talking about. If you're talking about meiotic recombination, it's true that this particular type of genetic recombination is highly constrained and controlled and only cuts and pastes in very specific pre-defined locations - and doesn't produce novel functionality that wasn't already there within the gene pool of options. Rather, this form of Mendelian variation simply allows for expression of various forms of pre-existing functionality that were already there within the gene pool of functional options for a particular location within the genome. However, there are in fact other far less common ways that genetic sequences can be "cut and pasted" together that are not so constrained, but are truly random in both the cutting and the pasting. Of course, the reason that I'm not making a big deal about Zachriel's use of such recombination methods is because, statistically, it really doesn't matter which method of "recombination" you're talking about when it comes to the problem of crossing larger and larger Levenshtein distances within sequence space. The odds of success remain essentially the same regardless of what types of truly random mutations are being considered. seanpit

In this line, an interesting and relevant paper was once published by Lenski et. al., entitled, "The Evolutionary Origin of Complex Features" in the 2003 May issue of Nature. In this particular experiment the researchers studied 50 different populations, or “genomes”, collectively comprised of 3,600 individuals. Each individual began with 50 lines of code and no ability to perform "logic operations". Those that evolved the ability to perform logic operations were rewarded, and the rewards were larger for operations that were "more complex". After 15,873 generations, 23 of the genomes yielded descendants capable of carrying out the most complex logic operation: taking two inputs and determining if they are equivalent (the "EQU" function). The lines of code that made up these individuals ranged from 49 to 356 instructions long. The ultimately dominant type of individual contained 83 instructions and the ability to perform all nine logic functions that allowed it to gain more computer time. In principle, 16 mutations (recombinations) coupled with the three instructions that were present in the original digital ancestor could have combined to produce an organism that was able to perform the complex equivalence operation. In this particular experiment, the “recombinations” of code where limited to include particular portions or lines of code, without random cuts anywhere within the lines of code or random pasting of lines of code just anywhere within the evolving sequences – but only in particular locations. Still, even at this relatively low level of functional complexity (requiring no more than 16 mutations to achieve success) evolution of novel function didn’t happen.

"At the other extreme, 50 populations evolved in an environment where only EQU was rewarded, and no simpler function yielded energy. We expected that EQU would evolve much less often because selection would not preserve the simpler functions that provide foundations to build more complex features. Indeed, none of these populations evolved EQU, a highly significant difference from the fraction that did so in the reward-all environment (P = 4.3 x 10e-9, Fisher's exact test). However, these populations tested more genotypes, on average, than did those in the reward-all environment… because they tended to have smaller genomes, faster generations, and thus turn over more quickly. However, all populations explored only a tiny fraction of the total genotypic space. Given the ancestral genome of length 50 and 26 possible instructions at each site, there are ~5.6 x 10e70 genotypes; and even this number underestimates the genotypic space because length evolves."

Isn't that just fascinating? Even within a relatively small sequence space of just 10e70 sequences (equivalent to no more than a 50 character sentence in English), evolution stalled out with a gap distance of just 16 “neutral” mutations wide. When the intermediate steppingstones were no longer defined as “beneficial”, the relatively small neutral gap that was created successfully blocked the evolution of the EQU function (despite the hyperdimentionality of the sequence space here). Now, isn't this consistent with my predictions? This experiment was only successful when the intelligent designers were capable to defining what intermediate closely-space sequences or functions were "beneficial" for their evolving "organisms" (as in Zachriel’s “Phrasenation” algorithm) and exactly how the random mutations would be able to cut and paste lines of code. Obviously, if enough sequences or functions are defined as beneficial, producing a short average Levenshtein distance between beneficial islands or steppingstones within sequence space, then certainly this situation will result in rapid evolution - as we saw here in Lenski's 2003 demonstration. However, when neutral gaps grow in a linear manner with each step up the ladder of functional complexity, this quickly becomes a real problem for evolutionary progress - as many of Lenski's other evolution "demonstrations" have shown over the years since (when it comes to both the evolution of functionally novel computer code and novel functionality in real organisms in real life). seanpit

Zachriel:

Recombination is an observed mechanism of biological evolution.

And as I have explained to you no one can say if recombination is a happenstance occurrence. Most likely it is an intelligently designed feature that allows for genetic diversity in a short time. If you ignore that then you are arguing from ignorance, which all observations and experiences have shown, is your favorite place to argue from. Virgil Cain

Zachriel,

Recombination is an observed mechanism of biological evolution. If you ignore recombination, then your claim is irrelevant as an argument against evolution. However, you are correct that point-mutation alone cannot explain the evolution of complex structures.

As I’ve already explained to you in fair detail (and even greater detail on my website), while genetic recombination is real, it doesn’t help you solve the problem if you are using truly mindless random mutations and function-based selection. The odds of success are essentially unchanged regardless of the type of mutations you use. The use of recombination mutations, or whatever other type of truly random mutation you wish to consider, simply don't solve the exponential problem at hand. How can this be true? Why don't recombination mutations help to solve the problem? Again, the ability to produce functionally-beneficial recombinations at higher and higher levels of functional complexity is based on the odds of several things to include 1) the proper sequence existing pre-formed within your “pool” of options and 2) the proper cutting of just the right sequence and then pasting it into just the right location within a large pool of non-beneficial options. The problem for the Darwinian mechanism is that these problems become exponentially more and more problematic with each step up the ladder of functional complexity. You don’t recognize these as problems on your website because you are easily able to circumvent them by intelligent design. You intelligently set up your “pool” of options and then intelligently pick just the right sequences to undergo recombination at just the right time and location. Of course it’s great that you can demonstrate the effectiveness of intelligent design here, but what does this have to do with the Darwinian mechanism of truly random unguided mutations/recombinations and function-based selection where each step is sequentially beneficial compared to what came before? Your “Phrasentation” program, in particular, demonstrates my point here quite nicely. The steppingstones it selects as are clearly not sequentially beneficial when it comes to their meaning/functionality. You see, without the input of your own intelligent design, a mindless search algorithm simply can’t do the job beyond very very low levels of functional complexity this side of a practical eternity of time. Why not? Because, the number of non-beneficial options expands at an exponentially greater rate compared to those options that would actually be functionally beneficial – with each step up the ladder of functional complexity. While there might be the potential for finding a successful path, this potential becomes exponentially less and less likely with each linear increase in the minimum likely Levenshtein distance between the starting point and the next closest potentially beneficial target in sequence space. And, there is nothing in literature, absolutely nothing, that substantively addresses this problem for your Darwinian mechanism. seanpit

Me_Think,

Representing biological landscape search in terms of Levenshtein distances is pretty archaic. Searches are multidimensional, which reduces the search space drastically.

While you’re correct in pointing out that the “sequence spaces” in question here are hyperdimensional, you’re mistaken to think that this reduces the distances or increases the odds of successfully finding novel beneficial target sequences over linearly expanding Levenshtein distances. You see, a linear expansion of a Levenshtein distance between a starting point and a potential target sequence will in fact decrease the odds of successfully finding this target sequence in an exponential manner – regardless of the fact that the space is hyperdimensional. This makes absolutely no difference for a random search algorithm as compared to a two or three dimensional search. The odds of success still decrease exponentially as the minimum Levenshtein distance increases linearly. Beyond this, none of the papers you reference explain how random search algorithms within hyperdimensional space can cover a linearly expanding Levenshtein distance between strings or character sequences without an exponential increase in required time… However, if you do have a real solution to this problem, by all means let me know. I’d be most interested indeed! seanpit

seanpit: Just because the Levenshtein distance between one word and the next closest might be 1, that doesn’t mean that there is no random walk involved to get across even this short Levenshtein distance – i.e., that you’re guaranteed to “keep your feet dry”. You draw diagrams showing oceans, you use the word ocean. The question you raised was whether there are evolutionary steps connecting words. It's right up there @1. seanpit: Beyond this, there is no 16-letter word (or longer), that I know of, that can be built upon a sequence of words that are each separated by a Levenshtein distance of 1. Recombination is an observed mechanism of biological evolution. If you ignore recombination, then your claim is irrelevant as an argument against evolution. However, you are correct that point-mutation alone cannot explain the evolution of complex structures. Zachriel

Representing biological landscape search in terms of Levenshtein distances is pretty archaic. Searches are multidimensional, which reduces the search space drastically. You can get more information at http://www.ieu.uzh.ch/wagner/publications.html and MathLab package for calculating hyper-dimension search can be found here - http://www.ieu.uzh.ch/wagner/publications-software.html Me_Think

Zachriel:

There is no swimming with words. You can keep your feet dry from one-letter words up to 16-letter words and more. (The odds of randomly stumbling across a 16-letter word is about 1 in 10^20.)

Again, there is a “swim” or “random walk” even with spaces dealing with defined English words as “targets” – and this is without even considering the concept of sequentially improved beneficial function. Just because the Levenshtein distance between one word and the next closest might be 1, that doesn’t mean that there is no random walk involved to get across even this short Levenshtein distance – i.e., that you’re guaranteed to “keep your feet dry”. That notion is simply mistaken – as you very well know. Beyond this, there is no 16-letter word (or longer), that I know of, that can be built upon a sequence of words that are each separated by a Levenshtein distance of 1. In order to cross these Levenshtein distance gaps that are greater than 1, you resort to “recombination” or “concatenation” of pre-established sequences in your “pool” of options. http://www.zachriel.com/mutagenation/Pudding.htm This creates a problem as you move up the ladder of functional complexity where the ratio between targets and non-targets gets exponentially smaller and smaller – as I’ve already explained above and in significant detail on my own website: http://www.detectingdesign.com/flagellum.html#Calculation You don’t recognize this problem because your algorithms are not based on sequentially beneficial function (and neither are your manually-generated sequences). There is also an additional problem with your manually-generated sequences that is common for evolutionists in general. That is, you assume that a sequence that would help you cross a larger Levenshtein distance already exists within your “gene pool’ of options without actually considering the odds that such a sequence would actually exist in a particular pool just when it might be needed. You also don’t consider the odds of precisely cutting out this sequence from its original location and then precisely pasting into its new position to create the larger more functionally complex sequence. This is another fundamental problem with your thought experiments along these lines as detailed on your own website. You argue:

We know that a path exists between the single-letter word "O" and "Beware a war of words ere you err", so it is only a matter of determining how many mutations are required to discover that path.

It’s not that simple. You see, it would be quite simple if you had an infinitely large “gene pool”. Unfortunately, however, your gene pool is not only finite, but quite small relatively speaking (as is the case in real Darwinian style evolution among living organisms). There are only so many sequences your small “gene pool” can store at any given point in time. How does it know which sequences to store? – sequences which might result in longer and more and more functionally complex sequences in the future? Outside of intelligent design, there is no way to know this. That is why you resort to “template matching” algorithms, rather than function-based selection, to keep your gene pool of options on the right path toward your pre-determined goals or “templates”.

Hold it now. Your claim was that there were no single-step pathways. If that is correct, intelligence won’t help.

I never made this claim. I never said that there were no possible single-step pathways at very high levels of functional complexity. What I said is that there are no series of closely-spaced steppingstones within sequence spaces beyond very low levels of functional complexity (i.e., steppingstone pathways separated by very short Levenshtein distances of just 1 or 2). The Levenshtein distances between potential “targets” within sequence spaces at higher and higher levels of functional complexity get longer and longer with each step up the ladder. At this point one is required to make large leaps across sequence space – leaps that cover large Levenshtein distances. Of course, it is quite possible to make large leaps into sequence space, covering large Levenshtein distances, with a single bound, and be successful in landing on a target sequence (as you have demonstrated quite nicely on your website). This is always possible. However, such leaps become exponentially less and less likely to be successful in one single step using a mindless search algorithm - with each step up the ladder of functional complexity. Intelligent design, on the other hand, can find such sequential single-step pathways were each step is indeed sequentially beneficial. You see, intelligent design can plan for the future – natural selection cannot. That is why intelligent design is able to set up systems in order to cross gaps in sequence space (as you have done with your illustrations on your own website) that natural selection cannot cross within a reasonable amount of time. By intelligent design I can pre-form two sequences that I know will, when merged together in just the right way, produce a larger much more functionally complex system. Natural selection cannot plan ahead like this. Natural selection does not know what might work, ahead of time, if various sequences were concatenated this way or that way. This is the advantage that intelligent design has over natural selection… and it’s a key advantage. What you have to demonstrate, now, is that a mindless search algorithm, like the Darwinian mechanism of random mutations and function-based selection, is actually up to the job of crossing large Levenshtein distances at these higher levels of functional complexity in a reasonable amount of time – like intelligent design can do. This you have yet to achieve because your algorithms don’t use function-based selection.

How do you intend to do that? It’s your claim, after all, you are trying to prove.

I don’t know how many times I have to refer you to my website before you’ll actually read what I wrote and at least try to substantively address the problems for the Darwinian mechanism that I’ve listed: http://www.detectingdesign.com/flagellum.html#Calculation In short, it is quite clear that, beyond extremely low levels of functional complexity, there are no “networks” of potentially beneficial target steppingstones like you imagine where each steppingstone has a very close Levenshtein distance compared to the next in the pathway (your own arguments on your website illustrate this particular point quite nicely). So, by the time you get beyond the level of 1000 saars, the uniformly distributed potential target islands are extremely isolated from each other, completely surrounded, on all sides, by truly vast oceans of non-beneficial sequences. In order to cross these huge gaps between these island targets, you have to resort to multi-character mutations or the cutting and pasting of just the right large sequences in just the right positions – over and over again (as you have done on your own website – right in line with what I’m saying on my website). The odds of such successful recombinations drop off, exponentially, with each step up the ladder (if you’re actually using the Darwinian mechanism). The multiple reasons for this exponentially problem are listed in detail on my website – if you care to actually look. Suffice it to say that your own efforts to falsify my position have actually ended up supporting my own claims. You haven’t even begun to address the problem because your algorithms and your assumptions are based on either intelligent design or template matching – not the actual Darwinian mechanism of random mutations and function-based selection. . . just like I’ve always predicted since I first ran into you on Talk.Origins back in 2004. seanpit

seanpit: The random walk or “swim” within such small spaces is of course relatively short indeed! There is no swimming with words. You can keep your feet dry from one-letter words up to 16-letter words and more. (The odds of randomly stumbling across a 16-letter word is about 1 in 10^20.) seanpit: However, this is not the case at higher and higher levels of functional complexity. That's your claim. Now show it. seanpit: I’ve already explained to you why point mutations and/or recombination/concatenations will not help solve the problem in real life – without the use of intelligent design (which forms the basis of the illustrations on your own website). Hold it now. Your claim was that there were no single-step pathways. If that is correct, intelligence won't help. seanpit: 1) Generate truly random mutations (point, indel, recombination, etc) that aren’t limited to determining and clipping out intact words or select “phrases” (something that doesn’t happen in real life). Been there, done that. seanpit: 2) Select based on changes in beneficial function – not template-matching which doesn’t happen in real life. How do you intend to do that? It's your claim, after all, you are trying to prove. seanpit: 3) Have a reasonable maximum steady state population size with a reasonable reproductive rate and mutation rate. In other words, old sequences must “die off” as fast as new ones are “born” so that the overall population size remains the same. Been there, done that. seanpit: If you actually model how the Darwinian mechanism really works, you will quickly discover that your neat little pathways of shortly-spaced steppingstones break apart and become widely separated very quickly as you move up the ladder of functional complexity beyond your short little sequences. How do you intend to do that? It's your claim, after all, you are trying to prove. Zachriel

Zachriel,

But we’re glad you agree that word-space is generally structured so as to be navigable by an evolutionary (selectable stepwise) search.

Again, as I originally explained to you back in 2004, this is only true at very very low levels of functional complexity within very small sequence spaces. The random walk or "swim" within such small spaces is of course relatively short indeed! However, this is not the case at higher and higher levels of functional complexity. This situation changes quite dramatically beyond these very low levels. Why then do you continually misrepresent my position here and on your own website when you know that your arguing against a strawman misrepresentation of my actual position? Why not at least be honest enough to present and candidly deal with my true position? - unless that's just too hard for you? ;-)

seanpit: You can’t just keep adding single characters to your sequence where each single character addition will be functionally beneficial.

Don’t forget point mutation and recombination.

I’m not. I’ve already explained to you why point mutations and/or recombination/concatenations will not help solve the problem in real life – without the use of intelligent design (which forms the basis of the illustrations on your own website). Such necessarily precise recombinations could not be realized, without the additional input of intelligent design, by any algorithm based on a sequentially beneficial selection process that models how natural selection works in real life. Again, your manually-generated illustrations work because they are based on intelligent selection, not mindless random mutations and function-based selection. And, your "Phrasenation" algorithm works because it is based on template matching - not function-based selection. Again, you're simply not modeling natural selection. If anything, your website illustrations highlight the truth of my position - not yours.

seanpit: Try it and see. Very quickly you will come to breaks in your pathway where the distance that needs to be crossed is more than a Levenshtein distance of 1.

Try it how?

By actually modeling what the Darwinian mechanism does in real life: 1) Generate truly random mutations (point, indel, recombination, etc) that aren’t limited to determining and clipping out intact words or select “phrases” (something that doesn’t happen in real life). 2) Select based on changes in beneficial function – not template-matching which doesn’t happen in real life. 3) Have a reasonable maximum steady state population size with a reasonable reproductive rate and mutation rate. In other words, old sequences must “die off” as fast as new ones are “born” so that the overall population size remains the same. If you actually model how the Darwinian mechanism really works, you will quickly discover that your neat little pathways of shortly-spaced steppingstones break apart and become widely separated very quickly as you move up the ladder of functional complexity beyond your short little sequences. The option for the recombination of sequences or “cutting and pasting” sequences together that already exist in your established “gene pool” of options won’t help you at higher levels because of the statistical problems I’ve already explained above. In short, at higher and higher levels the odds that just the right sequences will exist, pre-formed, within the established pool of options so that only one or two or three mutations would be needed to cross the gap decrease, exponentially, with each step up the ladder. seanpit

Mung: Yet more evidence for intelligent design. Thanks Obama! Zachriel

Zachriel:

But we’re glad you agree that word-space is generally structured so as to be navigable by an evolutionary (selectable stepwise) search.

Yet more evidence for intelligent design. Mung

seanpit: There is a “random walk” even when the next steppingstone is just 1 Levenshtein step away from the starting point (i.e., a random search algorithm is not successful every single time at this distance). There's no reasonable way to read "swim through this ocean of meaningless words," as meaning the same as there are stepping stones the entire distance. You emphasize that with your diagram. http://www.educatetruth.com/wp-content/uploads/2014/01/Sequence-Space.png But we're glad you agree that word-space is generally structured so as to be navigable by an evolutionary (selectable stepwise) search. seanpit: You can’t just keep adding single characters to your sequence where each single character addition will be functionally beneficial. Don't forget point mutation and recombination. seanpit: Try it and see. Very quickly you will come to breaks in your pathway where the distance that needs to be crossed is more than a Levenshtein distance of 1. Try it how? Zachriel

One more thing you don't seem to realize when it comes to your argument on your website: http://www.zachriel.com/mutagenation/Beware.htm When you "concatenate" your words and phrases in your "evolving" sequences, you do so with the use of intelligent design. It's not a random concatenation process as happens in real life. In real life sequences are cut out at random and pasted at random within the middle of other previously functional sequences. The result is almost always a loss of function, not a gain in novel beneficial function. And, this becomes exponentially more and more true with each step up the ladder of functional complexity... You see the problem here? Just because you have all the right words, preformed, in your pool of options to produce all the works of Shakespeare does not mean that they will assemble themselves, without intelligent design, to form anything functionally meaningful/beneficial beyond very very low levels of functional complexity. All you've done here is moved the problem up a scale. Instead of concatenating individual letters to make longer and longer sequences, you've resorted to concatenating entire words and phrases. You assume that, once an individual word or phrase is formed that your problems are over - that these words and short phrases can easily concatenating themselves, randomly, in the proper order to produce longer and longer Shakespearean phrases and poems, etc... ad infinitum, without any significant gaps in sequence space to slow this process down. That's just not how it works in real life. In real life it becomes harder and harder to get even pre-existing short meaningful/functional sequences of DNA (or English words) to concatenate property together to form novel systems of function at higher and higher levels of functional complexity - exponentially more and more difficult with each step up the ladder. By the time you're talking about systems that require a minimum of more than 1000 saars, there simply are no existing two or three subsystems that are already pre-formed that can be concatenated together without requiring numerous significant modifications that are not sequentially selectable as "beneficial". Suddenly you're left with a very large non-beneficial gap problem. And, if you argue that there are several dozen or so smaller systems that could, theoretically, be concatenated properly to form the larger system. Well, you run into the statistical problem of getting them all to arrange themselves properly by random chance alone until the larger system is complete and novel beneficial function can be realized. seanpit

Zachriel,

Your statement clearly states it requires a random walk. It does not. There are selectable stepwise evolutionary pathways.

There is a "random walk" even when the next steppingstone is just 1 Levenshtein step away from the starting point (i.e., a random search algorithm is not successful every single time at this distance). Of course, this random walk is a very small walk with pretty high success rates at these low levels. I never said otherwise regarding 7-character sequence space. As I explained to you way back in 2004, the "swimming distance" isn't very far at all when you're talking about finding beneficial targets within sequence spaces that are less than a dozen characters in length! I specifically explained to you, way back then, that there are in fact "bridges" or closely-spaced "pathways" of "steppingstones" within such low levels of sequence space. Again, I went on to explain to you, in some detail, that such pathways rapidly break down as you move into higher and higher levels of sequence space/functional complexity. You can't just keep adding single characters to your sequence where each single character addition will be functionally beneficial. Try it and see. Very quickly you will come to breaks in your pathway where the distance that needs to be crossed is more than a Levenshtein distance of 1. Pretty soon you're talking about minimum Levenshtein distances of 2 or 3. And, as you move up the ladder of functional complexity, the minimum Levenshtein distance that must be crossed grows, in a linear manner. And, with each linear increase in this gap distance, the average time required to cross this distance increases exponentially. I explained this all to you way back in 2004... if you will recall:

Sean Pitman: It is my position that all language systems, to include English as well as genetic and protein language systems of living cells are not lined up nice and pretty like at all and that the clustering that does indeed exist at lower levels of complexity get smaller and smaller and more and more widely spaced, in and exponential manner, as one moves up the ladder of functional complexity.

What is so confusing about this concept? And, how does your "Phrasenation" algorithm solve the problem? Hmmmm? Why did you even create your Phrasenation algorithm if you had no idea what I was talking about? Please do explain your position beyond the lowest levels of functional complexity - because your Phrasentation algorithm simply isn't helpful in modeling the actual Darwinian mechanism. I've been waiting a very long time for you to come up with something to support your primary argument that the Darwinian mechanism is actually up to the task as you claim it is... and I'm still waiting. Oh, but what about your argument:

The doggerel “O Sean Pitman” shows that at least some long sequences are not disconnected in phrase-space.

There are several problems here. First off, and most importantly, your intermediate sequences are not sequentially beneficial in meaning/function compared to what came before. That's a fundamental problem for your position. Next, your steppingstones are not separated from each other by a Levenshtein distances of just 1. And, you don't get remotely close to a 1000 character sequence where each step is not only closely spaced, but is also sequentially beneficial in function/meaning compared to what came before - which has been my long-stated limitation for evolutionary progress via any kind of Darwinian algorithm (which yours is not). Your "Phrasenation" algorithm proves my point here. Look at the phrases that it generates as they "evolve". They are not sequentially meaningful/beneficial as they evolve... not even close. Most of the time they are completely meaningless. They aren't selected based on functionally beneficial meaning at all. They are selected based on template matching. Now, you tell me, how is this remotely comparable to the Darwinian mechanism of random mutation and function-based selection? - beyond the lowest levels of functional complexity? You still really believe that your Phrasentation program/argument explains anything along these lines? Really? How so? seanpit

Your statement clearly states it requires a random walk. It does not. There are selectable stepwise evolutionary pathways.

Darwinian mechanisms are clearly a random walk Virgil Cain

seanpit: And you took this statement as me arguing that the Darwinian mechanism would stall out at 7-character sequences? Your statement clearly states it requires a random walk. It does not. There are selectable stepwise evolutionary pathways. seanpit: http://www.educatetruth.com/wp-content/uploads/2014/01/Sequence-Space.png The doggerel "O Sean Pitman" shows that at least some long sequences are not disconnected in phrase-space. Zachriel

Here’s a portion of our conversation from many years ago (2004), where you first argued for your idea that there are always nice little pathways of closely-spaced beneficial sequences throughout sequence spaces regardless of the level of functional complexity under consideration:

Sean Pitman: Since you are quoting me on your website Zach, it might be good to note that I never said that a 7-letter sequences would take "zillions" of generations much less years to evolve. In fact, I have said just the opposite many times… It seems very likely to me that the next higher levels (i.e: 8, 9, 10, etc) will take only one or two generations for your population to evolve 1,000 uniquely meaningful sequences at each level. However, by the time you get to level 25, I am thinking that your population is going to start noticeably stalling in its ability to evolve the 1,000 uniquely meaningful English sequences. By level 50 I'm not sure that your population of even 100 trillion will succeed in less than a million generations...

Zachriel: You have not shown that. Indeed, there is no way to know that from mere mathematical analysis. You have to know their distribution. For all we know, words are all lined up nice and pretty in "permutation space". It turns out that many, perhaps most, of them are!

- I say that the odds are very strongly against that assertion. It is my position that all language systems, to include English as well as genetic and protein language systems of living cells are not lined up nice and pretty like at all and that the clustering that does indeed exist at lower levels of complexity get smaller and smaller and more and more widely spaced, in and exponential manner, as one moves up the ladder of functional complexity. This assertion is not only mathematically valid, it has also been experimentally supported by many thousands of experiments that have never show anything to evolve in any language system beyond the lowest levels of functional complexity. For example, there are no examples of protein functions evolving that require a minimum more than a few hundred fairly specified amino acids working together at the same time. And, this is despite well over 10^14 individual organisms working on this problem under close observation for millions of generations.

Dated: 4/29/2004 https://groups.google.com/forum/message/raw?msg=talk.origins/TdfZ8CC9Bb0/X24ZX8is6xoJ https://groups.google.com/forum/#!msg/talk.origins/TdfZ8CC9Bb0/X24ZX8is6xoJ You see, we've been over all of this before. Now, if you have some valid reason for believing that your "pathways" really do exist within high levels of sequence space (like beyond the level of 1000 saars), by all means, present your evidence. Certainly your evolution algorithms do no such thing as they aren't based on functional selection, but on template matching without respect to beneficial function. seanpit

Zachriel, You wrote:

You said, “If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words.” But you don’t have to swim through meaningless sequences to cross over to the next meaningful word. Your claim is false.

And you took this statement as me arguing that the Darwinian mechanism would stall out at 7-character sequences? Really? I’m sorry, but that’s not what I said here. Again, nowhere did I say that evolution would stall out at the level of 7-character sequences. Why else would I specifically draw the line where the Darwinian mechanism stalls out at “1000 specifically arranged amino acid residues (1000 saars)”? – well before I had any conversation with you? While 7-character sequence space is a small “ocean” of around 1.28 billion sequences, the ratio of potentially beneficial vs. non-beneficial sequences is only about 1 in 300k. That’s a very small ratio when you’re talking about an algorithm that analyzes tens of thousands of sequences per “generation”. Also, take into account that at such low levels of functional complexity functional sequences for clusters that are connected to each other by extensive bridges that link the clusters throughout sequence space (see link below). http://www.educatetruth.com/wp-content/uploads/2014/01/Sequence-Space.png Yet, you write:

Sean Pitman: All I was trying to demonstrate is how the ratio of potentially beneficial vs. non-beneficial changes with each increase in the minimum size and or specificity of a sequence – exponentially. Sure. However, the space is not random, but highly structured. There are selectable stepwise paths from short words to long words.

The problem here, however, is that with each step up the ladder of functional complexity the “structure” and “stepwise paths” and “bridges” between the island clusters of beneficial sequences with higher and higher level sequence space start to break down – quite rapidly in fact. So, by the time you read the level of 1000 saars, there is no “stepwise path” of closely-spaced steppingstones between your starting point(s) and the next closest potentially beneficial island within sequence space. Your starting-point island is completely surrounded, on all sides, by a truly enormous ocean of non-beneficial sequences. There is simply no way to cross over to the next closest island except by swimming blindingly within an ocean that is larger than the universe for a star that is so very far away that it isn’t even visible with the most powerful telescope. We’re talking trillions upon trillions of years, on average, to get from one “island” to any other within such an extremely sparsely populated sequence space... That, in a nutshell, is the fundamental problem for the Darwinian mechanism. There simply are no “stepwise paths” beyond very very low levels of functional complexity. They just don’t exist. Of course, you’re not alone in believing that they must exist at all levels of functional complexity. I had a debate not too long ago with a mathematician, Jason Rosenhouse, who made the same claim that you just made here. He argued that the exponentially increasing ratio of potentially beneficial vs. non-beneficial at higher and higher levels of functional complexity didn’t matter because there would always be thin little paths of steppingstones that could quickly and easily transport the evolving sequence across the vast oceans of non-beneficial sequences (see link below). http://www.detectingdesign.com/JasonRosenhouse.html#Steppingstones The problem, of course, is that this just isn’t true beyond the lowest levels of functional complexity. Such paths simply don’t exist at or beyond the level of 1000 saars. Why not? Because, it is known that functionally-beneficial sequences have an essentially uniform distribution within sequence space. It is also known that at higher levels of functional complexity the modifications needed to get from one island to the next requires more than the tweaking of just one or two residue positions. At the level of 1000 saars dozens of residue potions need to be modified to produce something qualitatively new that is also functionally beneficial to the organism. And, a non-beneficial gap distance that is a few dozen residues wide (the Levenshtein distance) is not crossable in what anyone would consider to be a reasonable amount of time. For further information on this topic see: http://www.detectingdesign.com/flagellum.html#Calculation But, of course, I've already explained this to you before during our original conversations over 10 years ago:

Sean Pitman "Well now, that also depends now doesn't it? The answer to this question is really the answer to your second question. Technically speaking, the English language system _could_ have been set up so that all 2-letter sequences surrounding the "at" sequence would be meaningfully defined." Zachriel: It wasn't. Sean Pitman: That is correct. It wasn't set up like this even though it could have been. Instead, it was set up very much like I claim it was. It is much more randomly diffuse in its setup than you and many other evolutionists seem to be capable of recognizing. At lower levels the islands and bridges are in fact quite common. But, as even you have discovered, these islands start moving rapidly away from each other and the bridges start narrowing and snapping completely, in an exponential manner, with each step up the ladder of meaningful complexity. https://groups.google.com/forum/#!msg/talk.origins/TdfZ8CC9Bb0/Ad8Fbww1TYAJ

seanpit

Zachriel is still conflating artificial selection with natural selection. Sean Pittman- trying to argue with zachriel is fruitless and just leads to aggravation. Virgil Cain

Sean Pitman: Where did I ever say that 7-letter words (or a good bit longer) can’t evolve in a reasonable amount of time? You said, "If I want to evolve a new 7-letter word starting with meaningful 7-letter word, I will have to swim through this ocean of meaningless words." But you don't have to swim through meaningless sequences to cross over to the next meaningful word. Your claim is false. You also said, "Getting from one meaningful 7-letter phrase to a different meaningful 7-letter phrase requires, on average, a fairly long random walk through 250,000 meaningless options." This reiterates your point that there is no selectable stepwise path, and that it requires a random walk. This is false. There are selectable stepwise pathways. Sean Pitman: All I was trying to demonstrate is how the ratio of potentially beneficial vs. non-beneficial changes with each increase in the minimum size and or specificity of a sequence – exponentially. Sure. However, the space is not random, but highly structured. There are selectable stepwise paths from short words to long words. Zachriel

Zachriel:

So you’ve abandoned your original contention that words much longer than seven letters can’t evolve per the algorithm you yourself provided above. That’s all you had to say.

Where did I ever say that 7-letter words (or a good bit longer) can't evolve in a reasonable amount of time? I've never made such a claim - ever. In fact, I've specifically said, many many many times (well before you came along with your Dawkins-like evolution algorithms), that my uncrossable line for evolutionary progress is at the level of 1,000 specifically-arranged characters. How is that remotely close to a very short 7-character sequence? All I was trying to demonstrate is how the ratio of potentially beneficial vs. non-beneficial changes with each increase in the minimum size and or specificity of a sequence - exponentially. You do realize, however, that a ratio of 1 in 350,000 is quite evolvable? There is no significant limitation to evolutionary progress at this level - especially given populations that run into the trillions (as in bacterial populations for example). For next time, why not try and read all of what I've written on a particular topic like this (as per the links I've provided in this thread) before you jump to conclusions? - and make claims about my position that I've never made? I simply challenged you to see how far you could get using a true Darwinian model of random mutations and natural selection. I never told you that the uncrossable line would end up at 7 or 10 or 12-letter words. I told you that you would eventually see an exponential decline in evolutionary potential as you moved beyond these very very low levels of functional complexity - until evolutionary progress completely stalls out before reaching the level of 1000 specifically arranged characters. So far, I've been right. You're algorithms have done absolutely nothing to support the Darwinian notion that random mutations and function-based selection can evolve anything beyond the low levels of functional complexity this side of a practical eternity of time. seanpit

seanpit: The ratio of defined vs. non-defined two-letter sequences is about 1 in 7. The ratio of defined vs. non-defined 3-letter sequences is about 1 in 18. The ratio for defined vs. non-defined 7-letter sequence is about 1 in 350,000… etc. That's right, which was the basis of your original argument. seanpit: Beyond this, just because a single word happens to exist within the English dictionary doesn’t mean that it has a selectable advantage over any other word in a given context. So you've abandoned your original contention that words much longer than seven letters can't evolve per the algorithm you yourself provided above. That's all you had to say. Zachriel

Virgil,

I was just telling you what evolutionists like Larry Moran will say about your claims.

I've had many discussions with Larry Moran over the years. The problem with neutral evolution, as Larry knows full well, is that with each linear increase in a neutral gap, the average time to cross this gap increases exponentially - because natural selection is completely blind within such a gap and cannot, therefore, aid in the process. So, you see, the argument that neutral evolution solves the statistical problems for Darwinian evolution is nonsense. It is the problem. seanpit

Zachriel, You wrote:

We’re not talking about phrases, but words as there is no disagreement as to what constitutes a valid word in the English language. Per your statement, words are functional, and per your statement longer words can’t evolve from shorter words because as the words get longer, they are more and more widely separated in letter-space. You had stated elsewhere that the limit was words of about seven in length.

First off, I never said that the limit to evolutionary progress was "words about 7-letters in length". That's not remotely true. My stated limit for evolutionary progress has been consistently placed at systems that require over "1000 specifically arranged characters" (be those characters letters or amino acid residues within proteins). Also, as you very well know, evolution is supposed to be able to start with the very simple and evolve the very complex - equivalent to starting with single words and evolving an entire Shakespearean play. You're fully aware of this. After all, didn't you write a "Phrasenation" algorithm to evolve entire phrases, poems, and longer works from Shakespeare? http://www.zachriel.com/phrasenation/ Why did you do this if you thought we were only dealing with single words in English? Beyond this, just because a single word happens to exist within the English dictionary doesn't mean that it has a selectable advantage over any other word in a given context. Again, in modeling real evolution, real natural selection, your selection process must be based on changes in function - not just matches to a pre-established target sequence. If you cannot do this, you're simply not modeling natural selection. You're not getting at the heart of the problem for the Darwinian mechanism. Of course, even without modeling natural selection, the ratio of all words within the English dictionary changes with their size. The ratio of defined vs. non-defined two-letter sequences is about 1 in 7. The ratio of defined vs. non-defined 3-letter sequences is about 1 in 18. The ratio for defined vs. non-defined 7-letter sequence is about 1 in 350,000... etc. And, this changing ratio is in regards to defined vs. non-defined - without respect to beneficial function. Still, one quickly gets the idea of what would happen to this ratio given the additional requirement of going beyond what is merely defined to what is also functionally beneficial within a given setting or environment. Clearly, the ratio would significantly decrease given such an additional requirement. The exponential nature of the problem becomes quite clear to the candid mind. Notice also that at lower levels there is a clustering effect within most language or information-based systems (to include the English language). For example, many longer words are comprised of shorter words or prefixes that end up being clustered within various regions of sequence space. Random point mutations within these clusters can move around fairly rapidly. However, such a clustering effect becomes less and less prominent with each step of the ladder of functional complexity. Gaps between clusters become wider and wider. This is not reflected in your "Phrasenation" algorithm, in particular, because you define all portions of phrases or sequences as "selectable" as long as they match your chosen target sequences. And, of course, that's a key problem with your algorithm. seanpit

Sean:

Random drift takes to long when it comes to making anything beyond very low levels of functional complexity (i.e., nothing that requires a minimum of at least 1000 specifically arranged amino acid residues). It just doesn’t happen and statistically is very very unlikely to happen this side of trillions of years of time.

I was just telling you what evolutionists like Larry Moran will say about your claims. Virgil Cain

Zachriel, There are two key problems with your algorithms when it comes to modeling the Darwinian mechanism of random mutations and natural selection:

1) Your algorithms don't select based on beneficial function. 2) Your mutations aren't random when it comes to where mutations take place within a sequence (i.e., in your algorithms, mutations never happen within the middle of words within a phrase for instance - unlike real mutations that affect DNA within organisms randomly in the middle of genes or other coding regions).

If you modify you're algorithms accordingly, I think you'll come up with very different results that demonstrate the exponential nature of the problem that natural selection faces in real life. Remember, it's all about the functionality of a sequence. If your algorithm does not select based on function, you haven't got a truly evolutionary algorithm. seanpit

seanpit: How can you argue that a random sequence of English words is more functionally beneficial compared to any other random sequence of English words? We're not talking about phrases, but words as there is no disagreement as to what constitutes a valid word in the English language. Per your statement, words are functional, and per your statement longer words can't evolve from shorter words because as the words get longer, they are more and more widely separated in letter-space. You had stated elsewhere that the limit was words of about seven in length. Zachriel

Zachriel, You wrote:

You defined function as a word that is defined or recognized as beneficial in a larger system, such as the English language.

Oh please. Words, or sequences of words, may have beneficial meaning, or no beneficial meaning, depending on context and how the words are arranged relative to each other. How can you argue that a random sequence of English words is more functionally beneficial compared to any other random sequence of English words? "are flabergasted figs tree dog" "stick wasp in weasel woods skunk" Which "phrase" is more functionally beneficial when spoken to an English-speaking person? You see, just because all of the individual words in a particular phrase may be found an English dictionary doesn't mean that the sequence is any more functionally beneficial compared to what came before. Yet, that is exactly what you have to determine if you're going to model how natural selection really works. You have to determine if the new sequence is functionally beneficial compared to what came before - i.e., that it produces some kind of functional advantage. It is not enough that it happens to match a portion of some predetermined target sequence - regardless of its own independent meaning/function. That's not now natural selection works. You have to demonstrate a functional advantage each step of the way. If you cannot do this, you're simply not modeling how natural selection really works. Now, I'll give you a starting point of any word you may find in an English dictionary. However, from that point onward you have to determine if the next evolutionary step is functionally beneficial in a given environment or situation compared to what came before. That's how natural selection works in real life. However, this isn't how your algorithms work. You write:

a at sat sate sated

Where is the change in beneficial function for each step in this sequence? While it is statistically easy to evolve between small words in the English language using single point mutations, and while it is more difficult to evolve between larger words, it is exponentially harder to evolve larger and larger sequences, to include multi-word sequences, when you have to make your selections based on functional benefits (regardless of the random search algorithm you choose to use). Also, in real life, mutations within a sequence cannot distinguish between whole "words" or portions of "words". The mutational breaks are randomly determined - unlike your algorithms. Now, this isn't a difficult concept to understand. Do you really not understand that natural selection is based on a functional advantage for the newly evolved sequence? seanpit

Virgil, You wrote:

Sean, Most evolutionary biologists would agree that natural selection alone is incapable of producing complex adaptations for the reasons you mention, mainly the fact of missing selectable steps. That is why drift and neutral (construction) theory have been given new life. They are the tinkerers behind the scenes. And sometimes adaptations can emerge from that.

Random drift takes to long when it comes to making anything beyond very low levels of functional complexity (i.e., nothing that requires a minimum of at least 1000 specifically arranged amino acid residues). It just doesn't happen and statistically is very very unlikely to happen this side of trillions of years of time. seanpit

seanpit: However, why do you keep missing the part where I said, “Very quickly you will find yourself running into walls of non-beneficial function“? You defined function as a word that is defined or recognized as beneficial in a larger system, such as the English language. seanpit: You do understand that the goal here is to model natural selection? – right? Do you not understand that your algorithm doesn’t do this? It's your algorithm. We're just implementing it. It's quite obvious that you meant that we take a word, such as "a", and then randomly change letters to find longer words. a at sat sate sated Zachriel

Sean, Most evolutionary biologists would agree that natural selection alone is incapable of producing complex adaptations for the reasons you mention, mainly the fact of missing selectable steps. That is why drift and neutral (construction) theory have been given new life. They are the tinkerers behind the scenes. And sometimes adaptations can emerge from that. Virgil Cain

One more thing Zachriel. You keep quoting my original challenge like I never mentioned that natural selection was based on beneficial changes in function:

Sean Pitman: say you start with a short sequence, like a two or three-letter word that is defined or recognized as beneficial by a much larger system of function, such as a living cell or an English language system. Try evolving this short word, one letter at a time, into a longer and longer word or phrase. See how far you can go. Very quickly you will find yourself running into walls of non-beneficial function.

However, why do you keep missing the part where I said, "Very quickly you will find yourself running into walls of non-beneficial function"? You see, I've always consistently pointed out to your that selection must be based on beneficial function. Yet, your algorithms aren't based on function at all, but on template matching to pre-selected targets where any match to any portion of your pre-selected target sequence is "selectable" by your algorithm. Natural selection cannot do what your algorithms do! seanpit

Zachriel, You do understand that the goal here is to model natural selection? - right? Do you not understand that your algorithm doesn't do this? Natural selection can only select based on changes in beneficial function within a given environment - that's it. Your algorithm doesn't select based on functional changes at all. Your algorithm selects based on template matching or the additional of small sequences without regard to their meaning or the overall change in function or meaning of the evolving sequence. That means that your algorithm does not actually model natural selection at all - not even a little bit. You're algorithm is doing exactly what Richard Dawkins "Weasel" algorithm did. Don't you see that? You haven't come up with anything new or helpful here at all. Not at all. In my initial discussions with you, I was trying to get you to understand the problems with a function-based selection mechanism when it comes to producing higher and higher level systems. The key problem, of course, is that as the minimum structural threshold requirements increase in a linear manner (i.e., either an increase in the minimum size and/or minimum degree of specificity) the ratio of potentially beneficial vs. non-beneficial sequences will decrease in an exponential manner. This is true for the English language system and any other information system you wish to name - to include systems based on DNA or proteins. For example, as the minimum size of a set of English characters increase from small words, to larger words, to phrases, to sentences, to paragraphs,... etc, the ratio of sequences that will be functionally beneficial will decrease in an exponential manner as compared to non-beneficial or meaningless sequences. I'm sure that even you can recognize the truth of this concept. Even Dawkins recognizes the truth of it. What happens, then, is that with each step up the ladder of functional complexity the next closest potentially beneficial island within sequence space gets farther and farther away - in a linear manner. And, with each linear increase in the minimum distance in sequence space, the average time it takes for a random search algorithm to find another qualitatively novel beneficial sequences grows exponentially. Now, go back and look at your "Phrasenation" program and notice that the vast majority of your intermediate steppingstone sequences make no sense - are not meaningful or functionally beneficial within a given environment. You're simply selecting any additional single "word", of whatever kind, and adding it to your evolving "phrase" without any consideration of if it makes meaningful sense or not - if it would be functionally advantageous in a given environment. And, eventually you end up with your "target phrase" - just like Dawkins did with his "Weasel" algorithm. That is why your algorithm "works" in such a rapid manner. However, it is also why your algorithm doesn't reflect what natural selection can do in real life. As I explained to you many years ago (see link below) your program is based on the notion that every part of a sequence in a collection of phrases like Hamlet is meaningfully beneficial as long as no partial words are present. This notion is simply ridiculous. I means that sequences like, "and in the" and ", no, not" and "is let the" are defined as meaningfully beneficial in your program. Basically, absolutely any addition to a string will be defined as beneficial as long as it is a complete word found as part of this particular sequence in Hamlet. It need not represent an intact thought much less an internally relevant thought. http://talk.origins.narkive.com/1Bty4KMM/zach-s-prasenation-evolution-program#post6 So, what are you trying to prove here? That evolution makes sense beyond very low levels of functional complexity? How have you done that any better than Richard Dawkins who admitted that his own "Weasel" algorithm doesn't really function like natural selection functions? seanpit

Zachriel: Like this: Or a bit more precisely, a h ka za b q az aa ... at ytn wt t uat ... at ... sat Or with a bit of recombination, this time leaving out the stillborns and cousins. a an can cancan Zachriel

Zachriel conflates artificial selection with natural selection. How typical... Virgil Cain

seanpit: the problem with your evolution algorithms is that nothing is selected based on beneficial meaning or function. Here's the challenge again:

Sean Pitman: say you start with a short sequence, like a two or three-letter word that is defined or recognized as beneficial by a much larger system of function, such as a living cell or an English language system. Try evolving this short word, one letter at a time, into a longer and longer word or phrase. See how far you can go. Very quickly you will find yourself running into walls of non-beneficial function.

Start with a two or three-letter word. That's easy. We can go to the dictionary and find one of those. Now, we are to "evolve this short word, one letter at a time, into a longer and longer word or phrase." Okay. We'll just stick with words for now, as there is no disagreement as to what constitutes a valid word in the English language. What does evolve mean? Well, it means randomly change a letter in a word, or randomly recombine parts of words that are already in the population. If it makes a new word, then we can add the word to the population. If you want, we could limit the population to just the longest words, but that isn't essential. That seems to be exactly what the challenge entails. ETA: Like this: a at sat sate sated Zachriel

Again, Zachriel, the problem with your evolution algorithms is that nothing is selected based on beneficial meaning or function. That means, of course, that your algorithms aren't doing what natural selection does in real life. Longer and longer functionally meaningful words, phrases, sentences, paragraphs, etc., become more and more separated from each other, in sequence space, so that the average number of required mutations to get from one to the next increases in an exponential manner. Not just any sequence of letters or words in English is functionally meaningful - producing some kind of advantage in a given environment. And, this is the very same problem that exists in DNA or protein sequence space. Functionally beneficial systems that require a greater minimum number and/or specificity of amino acid residues are exponentially harder to find in sequence space via any kind of random search algorithm. Surely you can understand that - if you only did the math or produced an algorithm that actually made selections based on beneficial functionality. In short, your algorithms do no "select" based on any kind of meaningful functional advantage beyond what already existed within the original "gene pool" of options. They select based only on a comparison to a pre-established target sequence without any evaluation of the functional meaning of the intermediate sequences. This is exactly the same problem Dawkins had with his "Methinks it is like a weasel" algorithm. http://www.zachriel.com/phrasenation/ http://www.zachriel.com/mutagenation/ At least Dawkins was honest enough to admit that his algorithm didn't truly reflect the Darwinian mechanism of natural selection...

Although the monkey/Shakespeare model is useful for explaining the distinction between single-step selection and cumulative selection, it is misleading in important ways. One of these is that, in each generation of selective 'breeding', the mutant 'progeny' phrases were judged according to the criterion of resemblance to a distant ideal target, the phrase METHINKS IT IS LIKE A WEASEL. Life isn't like that. Evolution has no long-term goal. There is no long-distance target, no final perfection to serve as a criterion for selection, although human vanity cherishes the absurd notion that our species is the final goal of evolution. In real life, the criterion for selection is always short-term, either simple survival or, more generally, reproductive success.

seanpit

seanpit: The problem with Zachriel’s evolution algorithms, as I’ve mentioned to him many times before, is the same problem Dawkins has with his evolution algorithm (“Methinks it is like a weasel”). Neither uses function-based selection where each mutation is functionally beneficial compared to what came before. Sean Pitman defined the function: "start with a short sequence, like a two or three-letter word that is defined or recognized as beneficial by a much larger system of function, such as a living cell or an English language system. Try evolving this short word, one letter at a time, into a longer and longer word or phrase." Zachriel

So, Dawkins and Zachriel need to go back to the drawing board and come up with a new evolutionary algorithm that actually reflects what we see in nature. If they do this, they will soon realize, if they are honest with themselves, that such algorithms stall out, in an exponential manner, with each step up the ladder of functional complexity. I wrote a Weasel program last night and it couldn't even reliably find even the first two letters 'M' 'E'. Mung

Not to mention the fact that natural selection is all about survival and reproduction and in all evolutionary algorithms that is granted from the start. That means natural selection is satisfied from the get go and it has nothing left to do. Virgil Cain

The problem with Zachriel's evolution algorithms, as I've mentioned to him many times before, is the same problem Dawkins has with his evolution algorithm ("Methinks it is like a weasel"). Neither uses function-based selection where each mutation is functionally beneficial compared to what came before. All of these algorithms use "target sequences" that function as templates. Each additional match to this target sequence is defined as "selectable" in these evolution algorithms. That is why they work so well and so quickly. The problem, of course, is that biological evolution does not and cannot work like this. Natural selection cannot preferentially select any novel mutation over any other until such a mutation comes along that actually produces some qualitatively novel functional change that also has a positive effect on reproductive fitness relative to all of the other individuals within that population. Using this Darwinian mechanism, finding novel functionality with greater and greater minimum size and/or specificity requirements becomes exponentially more and more difficult to achieve within a given span of time. http://www.detectingdesign.com/flagellum.html#Calculation So, Dawkins and Zachriel need to go back to the drawing board and come up with a new evolutionary algorithm that actually reflects what we see in nature. If they do this, they will soon realize, if they are honest with themselves, that such algorithms stall out, in an exponential manner, with each step up the ladder of functional complexity. seanpit

Dawkins' weasel is great support for What Dr Pittman wrote. The sentence "Methinks it is like a weasel" only works in one specific case, that is in the Shakespeare play that contains it. It is meaningless in every other piece of literature. It would only do any good if it arose and was properly integrated into that play. And not surprisingly Dawkins and the evo-minions seem totally unaware of that fact. Virgil Cain

Zachriel:

The origin of the life we know Just like this poem rose from simple forms, In meaning, and in kind, step-by-step.

What a nonsensical BS artist Zachriel is. Talk about closing one's eyes and blocking the sight- Zachriel is as blind as blind can be and just as mindless as natural selection. Virgil Cain

Sean Pitman: say you start with a short sequence, like a two or three-letter word that is defined or recognized as beneficial by a much larger system of function, such as a living cell or an English language system. Try evolving this short word, one letter at a time, into a longer and longer word or phrase. See how far you can go. Very quickly you will find yourself running into walls of non-beneficial function.

O Sean Pitman Beware a war of words ere you err. A man wins the crown, but lowers his helm. A kiss Is a kiss, and a war can be just, but a war of words Just irks the crowd and leads you far astray. Words, you know, can lead to a clash of swords. Why do you think that you alone have it Legit when sages aver another idea? Could it be that you could see the light But choose instead to close your eyes and block The sight? The origin of the life we know Just like this poem rose from simple forms, In meaning, and in kind, step-by-step. Zachriel