Uncommon Descent Serving The Intelligent Design Community

Gil Has Never Grasped the Nature of a Simulation Model

Share
Facebook
Twitter
LinkedIn
Flipboard
Print
Email

Tom English challenged me with this:

I say categorically, as someone who has worked in evolutionary computation for 15 years, that Gil does not understand what he is talking about. This is not to say that he is trying to mislead anyone. It is simply clear that he has never grasped the nature of a simulation model. His comments reflect the sort of concrete thinking I have tried to help many students grow beyond, often without success.

The reason for Tom’s lack of success is that he, and Darwinists in general, try to explain everything with an overly — indeed catastrophically — simplistic model. Here’s what’s involved in a real-world computer simulation:

My mathematical, computational, and engineering specialty is guided-airdrop technology. The results of my computer simulations, and their integration into the mechanics of smart parachutes, are now being used to resupply U.S. forces in Afghanistan. C-130 and C-17 aircraft can now drop payloads from up to 25,000 feet MSL, out of range of enemy small-arms, shoulder-launched missile, and RPG fire, and the payloads autonomously guide themselves to their targets within a CEP (circular error probable) of approximately 26 meters. Did I do all of this highly sophisticated mathematical and software simulation without ever having “grasped the nature of a simulation model”?

One small part of developing this technology involves mathematically and computationally simulating the descent rate of a parachute and its payload at various altitudes. This includes the following: the drag coefficient of the parachute, the chute reference area, the density of the air at various altitudes (not only determined by altitude but lapse rate — the rate at which air temperature changes with altitude), and other subtle considerations, such as the flow-field effects of the payload which changes the drag characteristics of the parachute.

If any mathematical, computational, or real-world assumptions about any of these factors are wrong, or if any unforeseen factors are left out (and what I described above represents a small percentage of what’s involved), the simulation breaks down. We do our best, but we never know for sure until we throw the thing out of an airplane, see where it lands, and tediously analyze the telemetry data recorded by the in-flight computer.

Based on these observations and computer simulations that can be tested in the real world, what confidence can anyone have that biological evolutionary computer simulations have anything to do with reality?

The answer is: none. It’s all fantasy and speculation, masquerading as science.

Comments
DvK While these peptides were less than 30 units long, you’re arguing against Moore’s Law if you don’t expect longer results in the future. This is already within an order of magnitude of the 100 unit protein of Dembski’s analysis. Great! Now we're talking. Wake me up when it produces 40 odd proteins, millions of which are assembled in a precise manner to generate a flagellum. Not all things are scalable, DvK. Just because I can pile rocks to the roof of my house doesn't mean I can eventually pile them to the moon. You Darwinists have a penchant for demonstrating the simple and extrapolating to the complex like everything just scales up without limits. Those of who know things don't always scale like that need better proof of concept than "Poof! Chance did it."DaveScot
October 3, 2006
October
10
Oct
3
03
2006
04:03 AM
4
04
03
AM
PDT
Tom asks that I define "trial and error". No problem. Since this is about genetic algorithms and biological evolution I'll just quote wiki's examples of trial and error processes. It says it all.
Examples Trial and error has traditionally been the main method of finding new drugs, such as antibiotics. Chemists simply try chemicals at random until they find one with the desired effect. The scientific method can be regarded as containing an element of trial and error in its formulation and testing of hypotheses. Also compare genetic algorithms, simulated annealing and reinforcement learning - all varieties for search which apply the basic idea of trial and error. Biological Evolution is also a form of trial and error. Random mutations and sexual genetic variations can be viewed as trials and poor reproductive fitness as the error. Thus after a long time 'knowledge' of well-adapted genomes accumulates simply by virtue of them being able to reproduce.
Thanks for playing, Tom. There's a lovely consolation prize waiting as you exit stage left. It's an Avida generated EQU instruction autographed by fellow chance worshipper/ professor-in-denial Richard Dawkins.DaveScot
October 3, 2006
October
10
Oct
3
03
2006
03:44 AM
3
03
44
AM
PDT
Actually Tom, they're mosfets if you want to get techincal about it, and there are two mosfets in the most basic logic gate (inverter). A NAND gate requires four mosfets. Even assistant professors of computer science at Texas Tech should know that all other logic gates can be constructed from NAND gates. What assistant computer science professors at Texas Tech probably don't know is that microprocessor simulations, prior to creating the first mask, absolutely have to model at the gate level because of something called propagation delay which can result in something called race conditions. I was whipping out the fuse programming for programmable logic arrays while you were still in high school and I didn't have the benefit of simulators way back then. Prop delays had to be calculated by hand to eliminate race conditions just as they had to be when designing with discrete TTL logic which I did for many years before logic arrays were invented. In 1991 I implimented the core logic for an 80486 motherboard in 19 discrete PALs with nothing but PALASM and hardware design genius. Google it in all the spare time you have now that you've been booted off Uncommon Descent for your nasty habit of getting personal.DaveScot
October 3, 2006
October
10
Oct
3
03
2006
03:24 AM
3
03
24
AM
PDT
DaveScot:
We already know that trial and error can find solutions to problems.
Define trial-and-error.Tom English
October 3, 2006
October
10
Oct
3
03
2006
02:55 AM
2
02
55
AM
PDT
I’m curious, do any evo-sims model anything that has been observed directly?
They have for many years. Consider J. L. Crosby (1967), "Computers in the Study of Evolution," Sci. Prog. Oxf., Vol. 55, pp. 279-292 (not the earliest example I could give you, but a striking one). Among other biological phenomena, he considers the 45 alleles of a single gene in a population of just 500 plants. Why are there so many alleles? Wright, Fisher, and Moran had given the question different mathematical treatments, and had never agreed on the answer. Crosby obtained a convincing answer through simulation.
Oe. organensis is a long-lived perennial, and there have been insufficient generations since the catastrophe for this number [of alleles] to have fallen very much. The mathematical arguments may have been very interesting, but they had little relevance to the biology of the problem. Wright had also attempted to consider spatial distribution [as had the computational model]; this was necessarily rudimentary because of the limitations of mathematics, and had led him to a conclusion quite different from that derived from the more realistic computer model.
The best place to find Crosby's paper is in David B. Fogel (ed.), Evolutionary Computation: The Fossil Record. It may interest some of you that Prof. Bob Marks, a friend of Bill Dembski at Baylor, and I both served as technical reviewers of the volume.Tom English
October 3, 2006
October
10
Oct
3
03
2006
02:48 AM
2
02
48
AM
PDT
BC The problem with Avida isn't that it fails to implement a genetic algorithm more or less along the lines of organic rm+ns. The problem is that it didn't create anything non-trivial. An EQU instruction cobbled together out of microcode is trivial. Decades ago we were using GA (but we didn't call it that) to create optimized printed circuit board layouts connecting thousands of points in 3 dimensions. And even that I wouldn't have the gall to say was anything remotely approaching the complexity of even the simplest bacteria. Avida is child's play in more ways than one.DaveScot
October 3, 2006
October
10
Oct
3
03
2006
02:44 AM
2
02
44
AM
PDT
DaveScot:
Tom English writes: “There are people working in computational chemistry (e.g., simulation of protein folding), but proteins are no more the right level of granularity for evolutionary simulation than are transistors the right level for microprocessor simulation. If you truly know anything about microprocessor simulation, you know what I am saying.” Your ignorance is showing again.
Considering that a one million gate design requires 5-8 verification engineers, the task of verification is dominating project costs.
I would have thought a Dell Millionaire would know the difference between a transistor and a gate. Furthermore, I would have thought he might know that verification of a large microprocessor design is not accomplished by simulation of the entire microprocessor at the gate level, let alone the transistor level.Tom English
October 3, 2006
October
10
Oct
3
03
2006
02:01 AM
2
02
01
AM
PDT
DaveScot:
Tom English write “It does not matter one whit whether the simulations fit biological observations.” A damn good thing too because if you tried, they wouldn’t.
I have read a great deal of the ID literature, and somehow attempts to fit ID-theoretic predictions (as opposed to "postdictions") to the biological observations have eluded me. Would you please point me to appropriate references?Tom English
October 3, 2006
October
10
Oct
3
03
2006
01:09 AM
1
01
09
AM
PDT
Gil, Karl wrote:
Gil would have us believe that after three days of being criticized for confusing the simulator with the simulated, and after writing a new post which evades the issue altogether but attempts to establish his credentials, that he is just now getting around to mentioning that his first post was “sarcasm” which noone should take seriously.
This certainly had crossed my mind.
The results of my computer simulations, and their integration into the mechanics of smart parachutes, are now being used to resupply U.S. forces in Afghanistan.
Google gives me 386 hits for the Affordable Guided Airdrop System (AGAS), but none with an instance of "Dodgen." Browsing some papers, it seems that your firm has used simulations to develop the technology, but that the deployed AGAS system does not make use of simulation. One paper indicates that accurate delivery of the load is a matter of 1) determining atmospheric characteristics prior to the drop and planning a descent trajectory, 2) making the drop at precisely the right point, and 3) controlling actuators that pull on the risers, subject to constraints on fuel consumption. The controller could in principle compute and re-compute simulations of the descent to plan trajectory corrections, but I would expect a design team to choose a relatively simple and robust controller. Have I got this wrong?
Did I do all of this highly sophisticated mathematical and software simulation without ever having “grasped the nature of a simulation model”?
Hmm. I think you do not understand how concrete your practitioner's perspective is -- all particulars and no principles. "Well, golly, we had to model the 'chute down to the stitches in the seams to make it land in a certain place, so the best way to gain an understanding of evolution is to model right down to the nuclear membrane. Yup. It's obvious." The objectives of controlling an entity and understanding an entity often do not lead to the same sorts of models. For instance, good control can often be achieved with a good statistical model, but statistical models, unless structured in certain ways, do not make good scientific models. (I understand that you must be using first-principles physical models.) Test drops are expensive, and I am sure simulation was essential to economical development of the system. But in what sense does your simulation get the load onto the target? And how much of the first-principles modeling did you do yourself? Generally a software engineer involved in such a project would implement models developed by others. Is the controller model-based? Did you develop the controller? Did you develop its model?Tom English
October 3, 2006
October
10
Oct
3
03
2006
01:00 AM
1
01
00
AM
PDT
Jason:
We can’t even predict hurricane paths or basic weather patterns correctly half the time, but whatever takes place in a GA simulation equals what has happened in the real world, even if real world (biologically speaking) time frame equals millions of years? Far fetched doesn’t even begin to describe the scenario.
Apparently you have not read the thread. I commented:
The simulations are qualitatively correct, but not quantitatively.
Or have I made a mistake in assuming that you know the difference between qualitatively correct and quantitatively correct preditions? Just ask if you do not understand the distinction. Under the reasonable assumption that hurricanes are chaotic, it is impossible to predict their tracks over more than the short term. This does not mean that we cannot model the general behavior of hurricanes. Similarly, we cannot predict exactly how evolution will proceed, but we can conceivably capture aspects of how it works. To understand a system is not necessarily to predict its long-term behavior, but to model its general behavior.Tom English
October 3, 2006
October
10
Oct
3
03
2006
12:45 AM
12
12
45
AM
PDT
Tom English An evolutionary computation is an evolutionary process in the real world. If the computation does something, an evolutionary process in the real world has done it. If you cannot grasp this, then you are not grasping what much of evolutionary computation is about. So by your logic a book of fairy tales are fairy tales in the real world.
Is this argumentum yo mama?Tom English
October 3, 2006
October
10
Oct
3
03
2006
12:21 AM
12
12
21
AM
PDT
It’s claptrap. The Avida algorithm used in the Lenski paper is not a Darwinian process. Darwinian processes are blind/dumb/purposeless.
The reason Avida is compared to darwinian process, even though Avida is given an explicit goal whereas Darwinian ones have no explicit goal has to do with the structure of Genetic Algorithms. What happens in Genetic Algorithms is this: Start with some organisms. There is some mutation. The mutated organisms are scored according to how well they accomplish the (predefined) goal. The ones which don't score well are killed (or at least not allowed to reproduce). The ones that do best are allowed to have the most children. Repeat the process. The end result is that each generation accumulates a better and better genome for survival (where survival is based on accomplishing the predefined goal). Remember: the ONLY role the goal plays in GAs is to create a survival differential. Naturalistic Evolution doesn't employ a goal. Instead, what happens is this: Start with some organisms. There is some mutation. The mutated organisms that survive in their environment (sometimes because they have the best genes) tend to produce the most offspring, and the ones with bad genes and genetic defects die off (producing little or no offspring). The end result is that each generation accumulates a better and better genome for surviving (where survival is based thriving in their environment). Now, people will get hung up on the fact that Genetic Algorithms use a goal. But, I don't think it's a very big deal. GAs use the goal to provide a survival differential. The survival differential gives direction to the genome's evolution (in this case, towards accomplishing the goal). Natural Selection, by the competition for food, mates, survival, has it's own built-in survival differential which gives directionality to the genome's evolution. In both cases, genomes evolve in whichever direction favors survival. I'll say that again because it's an important point: in [genetic algorithms and in real-world evolution], genomes evolve in whichever direction favors survival. Now, people complain that "evolution has no goal!" That's true, naturalistic evolution has no externally-defined goal, but that's different from saying that it has no direction - the direction of genome evolution is towards better survival. So, using the "goal"/"no goal" complaint about Genetic Algorithms sounds like splitting hairs to evolutionists. Sure, GAs don't mirror competition and environments, but they do mirror mutation, genomes, and survival differentials which give rise to directionality in genome evolution. GAs mirror traits x,y, and z of real-world evolution (where x, y, and z are some of the major hangups over the evolutionary mechanism), but because they don't mirror traits u and w, people want to throw them out entirely. Isn't it obvious that GAs do legitimize *some* of features of naturalistic evolution - features which some people have erroneous hangups about? GAs do legitimately illuminate features of real-world evolution. Don't throw them out because of some insistence that they don't perfectly mirror everything or because you misunderstand them.BC
October 3, 2006
October
10
Oct
3
03
2006
12:02 AM
12
12
02
AM
PDT
I wrote:
Avida has shown that a Darwinian process is capable of producing irreducible complexity.
Scott wrote:
Nonsensical statements like this will quickly get you unselected from this blog. And the ID proponents who post here, know better than to buy such claptrap. Consider this a warning.
Scott, If you have an argument to make, make it. Otherwise, feel free to take your bluster and bold fonts elsewhere. I am prepared to justify my assertion. Are you able to do the same?Karl Pfluger
October 2, 2006
October
10
Oct
2
02
2006
11:47 PM
11
11
47
PM
PDT
PaV, There's much to disagree with in your last few comments, but let me concentrate on your most egregious statements. You wrote:
Karl, you seem to be missing the point that if the proteins don’t fold properly, then biochemical function comes to an end. I hope you understand that. And, of course, NS can’t act on something that is not biologically active. Hence, the massive amount of computer power required to search out the proper solution to folding is an undertaking that “random mutation and NS” would in some way have to deal with.
PaV, Here's the source of your confusion: you think that NDE must "understand" or "compute" protein folding in order to find proteins that fold "properly" and enhance the fitness of the organism possessing them. That is a complete misunderstanding of how NDE works. NDE will "try" any protein that is produced by a mutation, without regard for how it folds (the mutations are random, remember?). The proteins that happen to fold "properly" will enhance fitness and will therefore be retained by selection. They're not retained because of how they fold; indeed, NDE doesn't "know" how they fold. It keeps them because, and only because, they enhance fitness. It's analogous to the use of soap. Soap was invented long before anyone knew the chemistry behind it. How? People found that a certain combination of ingredients produced a substance that was an effective cleaning agent. It didn't matter how it worked; the fact that it did work motivated people to keep making it and using it. In exactly the same way, NDE doesn't need to know why a protein works (which depends, among other things, on how it folds). It simply keeps the ones that work and discards the ones that don't. Folding computations are superfluous in such a system.
Dave, If Tom is a mathematician by training, and you, like me, an engineer, then he’ll never understand our practical side, and we’ll probably never understand the abstract side. Here’s how I see it: if you love equations, you become a mathematician; if want to explore equations, you become a physicist; if you want to use equations to do something, you become an engineer. Does that pretty much size things up?
How sad to imagine that a mathematician's training precludes one from understanding or appreciating practicalities, or that an engineer's training precludes one from understanding abstraction. I'm happy to report that it's not that way at all in the real world. Abstraction is part and parcel of my work as a computer engineer. All of the following are useful abstractions in the computer world: 1. Programming languages. 2. Instruction sets. 3. Virtual memory. 4. Java virtual machines. 5. Virtual servers. 6. Object interfaces. 7. APIs. 8. Standard cell libraries. 9. Filesystems. ...and so on. And what sort of an engineering drudge would simply plug numbers into equations with no curiosity for why they work?
You seem to be suggesting–and quite earnestly it appears–that Avida is more “real” than nature, and that if Avida can produce IC, then this proves that nature ought to be able to do it as well, leaving IDers in the position of having to prove that nature can’t produce IC.
I'd be curious to know where you got that impression, for I've said nothing of the kind. Here's what I did say:
Again, the only ingredients required for a Darwinian process are reproduction, heritable random variation, and selection. Avida has all of these. Avida has shown that a Darwinian process is capable of producing irreducible complexity. Does that prove that biological evolution can also do so? Of course not. But what it does do (and this is extremely significant) is to show that nothing about IC is inherently unreachable by a Darwinian process. If you want to argue that biological evolution cannot produce IC, you can no longer simply say “Of course evolution can’t produce IC, because no mindless Darwinian process can ever produce IC.” You have to come up with specific reasons why biological evolution, unlike Avida, cannot generate IC.
To reiterate: 1. Avida fits the definition of a Darwinian process. 2. Avida has been shown to produce irreducible complexity. 3. Therefore, to show that a system in nature is unreachable by Darwinian evolution, it is not sufficient to show that the system is irreducibly complex, unless you can explain why evolution, a Darwinian process, is incapable of producing IC, while Avida, another Darwinian process, is quite capable of doing so.Karl Pfluger
October 2, 2006
October
10
Oct
2
02
2006
11:45 PM
11
11
45
PM
PDT
Darwinian processes are blind/dumb/purposeless.
While, yes, the search is guided by the simulation parameters the search pathways are not predefined and the end goals are more generalized and not explicit. So AVIDA is blind/dumb to a certain extent. But I agree that without intelligently setting up the model it wouldn't produce anything too interesting.Patrick
October 2, 2006
October
10
Oct
2
02
2006
08:28 PM
8
08
28
PM
PDT
Karl: "Avida has shown that a Darwinian process is capable of producing irreducible complexity." Tom English: "Well, I don’t know it to be nonsensical claptrap, and I just reread Bill Dembski’s criticism of Avida in his expert rebuttal for the Kitzmiller case (pp. 18-20, available at designinference.com). If you do the same, you will see that Bill never says that Avida did not give rise to irreducibly complex structures." It's claptrap. The Avida algorithm used in the Lenski paper is not a Darwinian process. Darwinian processes are blind/dumb/purposeless. Tom English (31): "We often see here claims of what random mutation and natural selection cannot do, and evolutionary computation puts the lie to those claims." Now you are calling ID a lie? Provide an example of such an evolutionary computation program. One that demonstrates that random mutation & natural selection (alone) can produce CSI.j
October 2, 2006
October
10
Oct
2
02
2006
07:50 PM
7
07
50
PM
PDT
DS: Make us all sit up and take notice by getting a computer simulation to reveal a biochemical pathway, based on nothing but random mutation and simulated natural selection, where a flagellum can be produced. OK, it is no flagellum, but EC systems are being used to discover peptides that would be patentable, if a human had found them. http://www.genetic-programming.org/hc2005/peptides.pdf Now these model results can and probably will be synthesized and tested against the real world of real bacteria. While these peptides were less than 30 units long, you're arguing against Moore's Law if you don't expect longer results in the future. This is already within an order of magnitude of the 100 unit protein of Dembski's analysis. The GA was compared in this study to other chance-based search procedures. It was found to be much more efficient. The take-away point is that not all search procedures are equally efficient for this task (which is not in conflict with the No Free Lunch Theorem). The GA did not have any knowledge of biology. That was encapsulated in a neural network that repersented the fitness landscape. Based on the description in the paper, the GA could have been used equally well for optimizing air drop parameters.David vun Kannon
October 2, 2006
October
10
Oct
2
02
2006
07:36 PM
7
07
36
PM
PDT
Scott:
Karl:
Avida has shown that a Darwinian process is capable of producing irreducible complexity.
Nonsensical statements like this will quickly get you unselected from this blog. And the ID proponents who post here, know better than to buy such claptrap. Consider this a warning. Well, I don't know it to be nonsensical claptrap, and I just reread Bill Dembski's criticism of Avida in his expert rebuttal for the Kitzmiller case (pp. 18-20, available at designinference.com). If you do the same, you will see that Bill never says that Avida did not give rise to irreducibly complex structures. You can count on it that if Avida had not, he would have said so. Instead he argues that Avida is not tied to biological reality. That is, Avida's evolution of irreducible complexity implies nothing about biological evolution. If you have read this long thread, then you know that Karl and I have emphasized that evolution can be studied in the abstract, apart from biological systems. Bill Dembski has tacitly admitted that Avida evolved irreducibly complex structures, and Karl himself warned against extending the Avida results to biological evolution:
Avida has shown that a Darwinian process is capable of producing irreducible complexity. Does that prove that biological evolution can also do so? Of course not.
I don't see the problem with this.Tom English
October 2, 2006
October
10
Oct
2
02
2006
07:22 PM
7
07
22
PM
PDT
DS: "Of course there is a target. Does the term “differential reproduction” ring any bells?" What I intended is that there are no targets in the sense of pre-envisioned end states - particular organisms, features/functions of organisms, etc., relative to which the process can take a positive (or negative) direction. Not in the standard scheme of variation and selection.Reciprocating Bill
October 2, 2006
October
10
Oct
2
02
2006
06:44 PM
6
06
44
PM
PDT
This isn’t a mathimatical proof, but if you reread this sentance, you’ll see that there were two “mutations” that neither increased, nor decreased information. There are such things as “neutral mutations”.
Yes, but look at it again. Are there ANY mutations that cause a reduction in information from sequence A (whatever it happens to be) to sequence B? You can argue that my example involved neutral mutations, but you have to argue that ALL possible mutations to sequence A are neutral in order to debunk my example showing that mutations can produce information. If there are ANY sitatuations where A->B involves a decrease in information, then B->C must involve an increase. Essentially, you have to argue that all mutations to all possible sequences are neutral. If any mutation to any sequence decreases information, then, by virtue of the fact that all mutations can be reversed within a certain probability, some mutations must increase information. To DaveScot: You keep quoting me on things I didn't say. Those are actually quotes from Karl and Bill.BC
October 2, 2006
October
10
Oct
2
02
2006
06:18 PM
6
06
18
PM
PDT
BC However, there are no preset targets or “directions” guiding natural selection Of course there is a target. Does the term "differential reproduction" ring any bells?DaveScot
October 2, 2006
October
10
Oct
2
02
2006
04:11 PM
4
04
11
PM
PDT
It does not matter one whit whether the simulations fit biological observations. --Tom English
Just the other day I was wondering if you supported this idea. I think that about says it all... Karl:
Gil, is it really that painful to admit that you were wrong? Even if it was pointed out by (gasp) Darwinists?
Hello??? I distinctly remember pointing out to you that he was likely joking.
The significance of Avida, in particular, is as an example of how a Darwinian mechanism can produce irreducible complexity. Honest critics can no longer claim that NDE cannot in principle produce IC. They must show that a particular IC structure cannot be produced because of the particular local genomic and fitness landscapes. This is a blow to the many ID advocates who saw the existence of IC as proof of design.
This was probably only a "blow" to those new to ID. While, as you agree, Avida does not simulate real-life biology, it does show that an IC system can evolve in tightly constrained environments under certain conditions of replication, variation, and selection. This is important, as some ID proponents seem to regard "irreducibly complex" as tantamount to "unevolvable in principle" WHICH IS NOT TRUE. Fortunately you conceed this:
Avida has shown that a Darwinian process is capable of producing irreducible complexity. Does that prove that biological evolution can also do so? Of course not.
But then you make another common goof.
But what it does do (and this is extremely significant) is to show that nothing about IC is inherently unreachable by a Darwinian process.
IC primarily deals with DIRECT Darwinian Pathways; always has. Behe has always stated that INDIRECT Darwinian pathways are another matter. And we're talking about Darwinian processes in Biological Reality, which is not not nearly as constrained... Anyway, in the "The Evolutionary Origin of Complex Features," published in Nature in 2003 by Lenski, the selective forces that have 100% probability affixed are those for various simple binary arithmetic functions, which are ultimately used to build the "equals" (EQU) function, and for the EQU function itself. What's more, the more complex the function, the greater the reward given to the digital organisms for it. There is no analogy for such selective forces in nature. Nature doesn't care whether something is more or less functionally complex, it only cares whether it can survive in a particular environment. And what happens when no step-by-step rewards are given for functional complexity? An article on Avida in Discover magazine last year (Feb. 2005) stated, "when the researchers took away rewards for simpler operations, the organisms never evolved an equals program." By building rewards into the system — ie providing a highly constrained fitness function — the programmers gave the system a purpose. Hence its creative power: dynamics.org/Altenberg/FILES/LeeEEGP.pdf "Both the regression and the search bias terms require the transmission function to have 'knowledge' about the fitness function. Under random search, the expected value of both these terms would be zero. Some knowledge of the fitness function must be incorporated in the transmission function for the expected value of these terms to be positive. It is this knowledge — whether incorporated explicitly or implicitly — that is the source of power in genetic algorithms." But let's break the discussion down even further. I think of fitness functions as a "funnel" that must be properly constrained in order to provide results. The design of this funnel must be balanced; it can either be too constrained or not constrained enough. The programmer's goal is to find a balnace which the stated goal can be reached. In my opinion, there really isn't such a thing as a "generic" GA program which can solve anything thrown at it--each program has to be designed to fit a purpose. Let's say I have a Chess GA program. Assume abiogenesis and start off with an AI script that recognizes the environment (the chess board) and knows how to move the pieces (survive in the environment) and has a certain basic strategy. At startup this script is duplicated many times without any mutations. The scripting system making up simulated life cannot be abnormally simplistic, like with AVIDA, and the scripts must have the ability to replicate themselves. The functionality for replication must not be protected. The replication process is capable of producing AI scripts that no longer recognize how to play certain elements of chess or they cannot compile at all (death). As in, replication is not limited to producing fully functional chess strategies. Unfortunately the rules of chess are static so the environment doesn't change. Now let's say I applied a very broad constraint in my fitness function: if the script still retains the ability to compile (aka play chess) then it survives. "Old" scripts eventually die. "Lower lifeforms" are afforded a niche where they thrive instead of arbitrarily being eliminated in favor of "higher lifeforms" based upon a constrained process. As in, winners of games get duplicated more often, and with a larger population comes more processor time for this subsection of the population, but losers are not necessarily eliminated in an arbitrary fashion. They just need to be capable of basic survival. Thus a group of "winners" may eventually be modified to the point they start losing horribly or they split off. That's it. Now let's say I applied a very narrow constraint in my fitness function: the script must not only compile but it must win its game in a small number of moves in order to survive. This is tantamount to the environment being overly hostile. I wonder what you could expect from these approaches.Patrick
October 2, 2006
October
10
Oct
2
02
2006
04:04 PM
4
04
04
PM
PDT
DS: "Feedback loop does indeed capture the essence of rm+ns. RM generates trial balloons. NS is a feedback mechanism that informs the trial generator whether or not the last trial was a step in a positive direction." Cybernetic feedback entails homeostatic adjustment to a target by means of feedback. This is implied in your phrase "positive direction" (which entails the notion of a correct vs. incorrect direction) and may be exemplified by the feedback that directs a guided missile to its target my means of correction. However, there are no preset targets or "directions" guiding natural selection, only selection pressures that are strictly contingent and local. So the analogy to cybernetic feedback and course correction breaks down.Reciprocating Bill
October 2, 2006
October
10
Oct
2
02
2006
04:04 PM
4
04
04
PM
PDT
Natural selecton never had anything to do with creative evolution except to preserve the status quo long enough for extinction to make way for the next step in a determined goal-directed process, a process no longer going on. Neither natural nor artificial selection of micromutants will ever create a new species. All new species and the higher categories came from within the relatively few forms that were capable of producing descendents fundamentally different from themselves. The environment played no role in those events except possibly to act as a simple stimulus for an endogenous prescribed potential. Chance never had anything to do with either ontogeny or phylogeny just as Berg claimed 83 years ago. Some folks are just slow learners. They are known as Darwinians. Evolution is finished. Get used to it. Robert Broom did, Julian Huxley did, Pierre Grasse did and so did I. "A past evolution is undeniable, a present evolution undemonstrable." John A. DavisonJohn A. Davison
October 2, 2006
October
10
Oct
2
02
2006
03:58 PM
3
03
58
PM
PDT
BC Feedback loop does indeed capture the essence of rm+ns. RM generates trial balloons. NS is a feedback mechanism that informs the trial generator whether or not the last trial was a step in a positive direction. You wrote: Perhaps it is more accurate to say that we find so many feedback loops in nature because natural selection builds so many of them. It is dead accurate to call your statement a tautology.DaveScot
October 2, 2006
October
10
Oct
2
02
2006
03:33 PM
3
03
33
PM
PDT
Tom English wrote:
It makes just as much sense for us in evolutionary computation to seek abstract principles of evolution.
Tom, your last post was written in a very open, forthright fashion for which you're to be commended. Having said that, and sincerely meaning it, I, however, just don't get what you mean by "abstract principles of evolution". Can you list two or three? I think it was Karl, perhaps you, who said of a Darwininian process that it involves reproduction, variation, selection and an accrued benefit (something like that). Is that the kind of thing you're talking about?PaV
October 2, 2006
October
10
Oct
2
02
2006
03:20 PM
3
03
20
PM
PDT
PaV said: "The term selectionist causation sounds almost mystical. Why not just say: “We find nature involves feedback loops”? Because "feedback loop" does not capture the logic of selectionist causation. My boiler and thermostat are tied together in a negative feedback loop that causes my old home's winter temp to cycle around a setting that I select - but there is no selectionist causation in play in that process. Perhaps it is more accurate to say that we find so many feedback loops in nature because natural selection builds so many of them. Nor is the operation of variation and selection the least mystical; it is easily described and modeled. In fact, the mindless simplicity of the process if probably why so many balk at the claim that it has built most of the complexity we see in biology.Reciprocating Bill
October 2, 2006
October
10
Oct
2
02
2006
03:12 PM
3
03
12
PM
PDT
Karl:
If you want to argue that biological evolution cannot produce IC, you can no longer simply say “Of course evolution can’t produce IC, because no mindless Darwinian process can ever produce IC.” You have to come up with specific reasons why biological evolution, unlike Avida, cannot generate IC.”
You seem to be suggesting--and quite earnestly it appears--that Avida is more "real" than nature, and that if Avida can produce IC, then this proves that nature ought to be able to do it as well, leaving IDers in the position of having to prove that nature can't produce IC. As I say, you seem to be earnest, but this is kind of a wild understanding of the "real". Why so strong a conviction? What makes you so sure of all of this?PaV
October 2, 2006
October
10
Oct
2
02
2006
03:01 PM
3
03
01
PM
PDT
Karl Avida has shown that a Darwinian process is capable of producing irreducible complexity. Scott didn't tell you why that was nonsense. Any complexity produced in a stepwise fashion by a computer is by definition not irreducible. Make us all sit up and take notice by getting a computer simulation to reveal a biochemical pathway, based on nothing but random mutation and simulated natural selection, where a flagellum can be produced. I remain quite unimpressed by Avida finding pathways where higher level operands are produced by trial and error tinkering with microcode. Even a blind squirrel finds an occasional acorn.DaveScot
October 2, 2006
October
10
Oct
2
02
2006
02:34 PM
2
02
34
PM
PDT
BC:
Thus, I’ve shown in a mathematically rigourous way, that mutation CAN increase information. Now, what about those other 17 mutations?
This isn't a mathimatical proof, but if you reread this sentance, you'll see that there were two "mutations" that neither increased, nor decreased information. There are such things as "neutral mutations".PaV
October 2, 2006
October
10
Oct
2
02
2006
02:34 PM
2
02
34
PM
PDT
1 2 3 4 5 7

Leave a Reply