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
David vun Kannon wrote:
...to reconnect to biology, you have to ensure that you didn’t abstract away the wrong things in the first place. One way to do that is to do very little abstraction.
Less abstraction isn't necessarily better. The more concrete the model, the longer it takes to run. A hurricane model that tracked the path of every air and water molecule would be quite concrete but worthless for forecasting, because it would run so slowly that its forecasts would turn into retrocasts. Less abstraction doesn't necessarily mean better accuracy, either. A model of the soybean market that attempted to track the activity of every neuron in the brain of every farmer, trader, and consumer would degenerate into an incoherent mess, whereas a more abstract model could yield useful results. The evolutionary models that are out there are not attempts at "photorealistic" simulations of evolution. Nobody expects to see one of them reproduce the marsupial/placental split, for example. They are instantiations of abstract Darwinian processes (replication, variation, selection), and their purpose is to yield information about the capabilities and limitations of Darwinian processes. Biological evolution is just one instance of a Darwinian process. Whatever we learn about Darwinian processes in general applies to biological evolution in particular. 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.Karl Pfluger
October 1, 2006
October
10
Oct
1
01
2006
04:42 PM
4
04
42
PM
PDT
todd:
It seems to me Dembski isn’t really modelling, is he?
Let's focus on "Searching Large Spaces." He gives a model of search for a small target in a large space. He calls that model assisted search. Late in the paper, he indicates that natural evolution must have been an assisted search. That is, his claim is that one can model natural evolution as assisted search.
Is WD really modelling when he takes what is there - eg, a 100 amino acid protein sequence - and determines the statistical search space accordingly?
He's not trying to solve that particular problem. It's just an example to motivate his analysis of assisted search. His mathematical results are applicable to a large class of problems.Tom English
October 1, 2006
October
10
Oct
1
01
2006
04:21 PM
4
04
21
PM
PDT
Roger, Yes, Bill gives a made-up problem of searching for a particular sequence of amino acids. He does not go at all into the complexities of how amino acids specify proteins. And he certainly does not mention that in reality many permutations of an amino acid sequence may represent the same protein, and that numerous proteins are represented by amino acid sequences of length 100. So what Bill does is no more than to indicate that search for some biological structures takes place in combinatorial spaces. The mention of amino acid sequences does not embue his work with biological realism. The paper would be unchanged if the problem were to search for a sequence of 100 letters (the size of the search space would go from 20 ^ 100 to 26 ^ 100).Tom English
October 1, 2006
October
10
Oct
1
01
2006
03:54 PM
3
03
54
PM
PDT
Tom, It seems to me Dembski isn't really modelling, is he? I'm no math geek, nor am I an uber programmer geek, so I assume I'm missing something obvious and hope you'll correct my error. Is WD really modelling when he takes what is there - eg, a 100 amino acid protein sequence - and determines the statistical search space accordingly? That seems much different than setting up an digital 'environment' to demonstrate evolution. What am I missing?todd
October 1, 2006
October
10
Oct
1
01
2006
03:47 PM
3
03
47
PM
PDT
Tom English says:
If biological realism were as crucial as you say, Bill would be in bad shape.
Well, let us look at the first Dembski article you reference.
1 BlindSearch Most searches that come up in scientific investigation occur over spaces that are far too large to be searched exhaustively. Take the search for a very modest protein, one that is, say, 100 amino acids in length (most proteins are at least 250 to 300 amino acids in length). The space of all possible protein sequences that are 100 amino acids in length has size 20100, or approximately 1.27×10130.
Actually, he references the biological context right up front. Now certainly this doesn't prove anything about the underlying issue, but it does frame the difficulties for a simple RM&NS. Now, biological realities may include some convenient arrangement of the landscape, or maybe some sort of life-friendly self-organization that remains as yet undiscovered. I'll let the reader decide whether those possibilities favor a claim of "victory" by either side of the debate, but will merely point out that with regard to simple claims of RM&NS, known biological practicalities make Dembksi's arguments the easier road to travel, so he is unlikely to forsake them as you claim.Roger
October 1, 2006
October
10
Oct
1
01
2006
02:31 PM
2
02
31
PM
PDT
Earlier I wrote:
And if simulation models suffer the defects you say they do, then Bill Dembski’s abstract models of evolution in Searching Large Spaces and The Conservation of Information are in even bigger trouble.
Some of you are avoiding the crucial point that the leading theorist of the ID movement focuses much more on probability and information than he does living things. If you look at the papers I linked to, you will see that Bill says nothing about protein folding and genes and chromosomes and predators and plagues and earthquakes and meteors. He models information flow that has never been observed directly. There are no data for his models to fit, and he gives no guidance as to how to validate the models. If biological realism were as crucial as you say, Bill would be in bad shape.Tom English
October 1, 2006
October
10
Oct
1
01
2006
01:26 PM
1
01
26
PM
PDT
Natural selection is very real. Today, as in the past, it PREVENTS change. That is all it ever did. Don't take my word for it. You never do. "The struggle for existence and natural selection are not progressive agencies, but being, on the contrary, conservative, maintain the standard." Leo Berg, Nomogenesis, page 406 "Animals are not struggling for existence. Most of the time they are sitting around doing nothing at all." John A. Davison Any student of the living world knows that. Accordingly, Darwinians are not students of the living world. How does that grab you Darwimps? I bet it smarts a little eh? I certainly hope so. I l0ve it so! "A past evolution is undeniable, a present evolution undemonstrable." John A. DavisonJohn A. Davison
October 1, 2006
October
10
Oct
1
01
2006
12:56 PM
12
12
56
PM
PDT
Tom English says:
Proof of principle does not require a fit to data. We often see here claims of what random mutation and natural selection cannot do, and evolutionary computation puts the lie to those claims.
And what is that principle? The implication that EC in general involves "natural selection" seems to be quite a stretch.Roger
October 1, 2006
October
10
Oct
1
01
2006
11:18 AM
11
11
18
AM
PDT
Tom English,
bFast, You’re a great example of why I can’t make ID-NDE into a litmus test of the decency of a person. Thanks for your posts.
Good trick. I, an IDer, have been suggesting that a couple of other IDers on this site are viewing an issue with an opposite moral position to my own. If IDers hold opposite moral positions on any given issue, then your litmus test is bogus, isn't it. Further, I personally agree with many issues held by the religious right. I only fail to understand how the religious right has adopted a love for war. They certainly didn't get it from the New Testament that I read.bFast
October 1, 2006
October
10
Oct
1
01
2006
08:50 AM
8
08
50
AM
PDT
Sorry, my comments start with "Tom, this is a bit shocking..."russ
October 1, 2006
October
10
Oct
1
01
2006
08:02 AM
8
08
02
AM
PDT
Tom English // Oct 1st 2006 at 7:12 am bFast, You’re a great example of why I can’t make ID-NDE into a litmus test of the decency of a person. Thanks for your posts. Tom, this is a bit shocking. I've been to Pandas Thumb and seen the sneering insults that are smattered about. But it never occured to me that belief in ID-NDE might be a litmus test for "the decency of a person" (I know it's not your litmus test, but apparently the idea occured to you). The only realm in which I see this kind of litmus test for decency is liberal politics, in which liberals often believe conservatives are evil, but conservatives generally believe liberals are merely mistaken or foolish. Since I believe you identified yourself as a college professor (i.e. someone toiling deep in the heart of political liberalism), that explanation for your comment to bFast seems to fit. Am I mistaken?russ
October 1, 2006
October
10
Oct
1
01
2006
08:01 AM
8
08
01
AM
PDT
Tom, I think that DaveScot is making the following point. EC has abstracted evolution out of one description of biological reality. It has been successful in using that abstraction to attack problems. Only some EC researchers attempt to explain anything about biological reality, most are happy with having a useful tool to use on other problems. But to reconnect to biology, you have to ensure that you didn't abstract away the wrong things in the first place. One way to do that is to do very little abstraction. DS, am I close? There ar people using to GP to work on protein folding problems.David vun Kannon
October 1, 2006
October
10
Oct
1
01
2006
07:54 AM
7
07
54
AM
PDT
It seems I am wasting my time with this thread also. I did manage to flush out a Darwimp in response to my several papers. Just click on John A. Davison on the side bar and join the fun. The more the merrier I always say. SOCKITTOME! I love it so! "I'm an old campaigner and I love a good fight." Franklin Delano Roosevelt "There is nothing more exhilirating than to be shot at without result." Winston Churchill "Darwinians of the world unite. You have nothing to lose but your natural selection." after Karl Marx "A past evolution is undeniable, a present evolution undemonstrable." John A. DavisonJohn A. Davison
October 1, 2006
October
10
Oct
1
01
2006
07:39 AM
7
07
39
AM
PDT
Dave Scott,
I’m growing very frustrated by you and others’ inability to grasp the fact that models of reality need to be testable.
I have long been frustrated with certain parties' inability to recognize that neo-Darwinian evolution is an abstract process, not necessarily implemented in biota. In 1995, I ran an evolutionary computation on a massively-parallel computer and obtained more than 20 thousand models predicting annual sunspots counts. The series of sunspots counts is chaotic, and has been a challenging benchmark problem in time series prediction for many years. With a combination of models, I obtained far better prediction accuracy than anyone else ever had (“Stacked Generalization and Simulated Evolution,” BioSystems, 39(1), pp. 3-18, 1996). My question for you is how I did that. I had and still have zero knowledge of sunspot formation. To my knowledge, nobody understands the blasted things. How could I have front-loaded the evolutionary computation when no physicist or statistician had ever managed to give a good model of the time series? I can see no way that you can attribute the success of the evolutionary process to my intelligence. The point I want you and others to get here is that such successes of evolutionary computation demonstrate the efficacy of abstract neo-Darwinian processes. Again, if this says nothing about evolution, then neither do Bill Dembski's models of assisted search and added information. How are you going to rule out my work without ruling out his? Direct question.
I’ve given you examples of real models in biology (protein folding), mechanical design (aircraft), and electronics (microprocessors). These all model the real world and can be tested by seeing if they duplicate the results obtained in the real world.
Modeling always entails a choice of granularity. No one, but no one, simulates microprocessors at a fine level of granularity. The computational demands are too great, as you should know. 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. As for aeronautics, let's consider airfoil design instead of mechanics. Simulation is very useful in that domain, even at high Reynolds numbers. But if you ask someone to give you a precise prediction of airflow, he or she will not be able to give it. The simulations are qualitatively correct, but not quantitatively. In other words, simulations used to design airfoils cannot give the kind of predictions you demand of evolutionary simulations.Tom English
October 1, 2006
October
10
Oct
1
01
2006
07:04 AM
7
07
04
AM
PDT
DS: "Here’s yet another real, testable model that interests me: Stellar Evolution." Your distinction between models that are testable against empirical observations of some reality (actual stellar behavior) and those that are theoretical explorations certainly legitimately points to different sorts of simulation that serve different sorts of purpose (although the question of to which class of simulation evolutionary models belong is another debate). But your legitimate distinction doesn't speak to the problems with Gil's original proposal. Gil insisted that random modifications (a feature of a computational model of NS) extend all the way down into the hardware on which that simulation is run (the computational substrate) before the evolutionary simulation is complete. This is just plain wrong. Nor does it accomplish what I think Gil and maybe you are reaching for - a test grounded 3-space reality of an evolutionary simulation. His proposal simply does not pose that test.Reciprocating Bill
October 1, 2006
October
10
Oct
1
01
2006
06:50 AM
6
06
50
AM
PDT
DS said: "The substrate doesn’t matter if you’re not modeling reality. If you’re modeling reality then the substrate of reality is a benchmark by which all others are measured." You miss what I intend by substrate, and are using "substrate" with a different referent than am I. I am referring to the computational system capable of hosting the abstract collection of algorithms that compose the simulation. As pointedly illustrated above, this hardware may be composed of Tinker Toys, paper and pencil, hand calculator, ropes and logs, tunneling quantum events, etc. The logic of the algorithm that runs on these various physical substrates remains constant as the substrates vary. Using "substrate" in the sense I intend, as defined above, the assertion that an algorithm is independent of the particulars of the computational substrate is as true when simulating a system that exists in reality as when modeling hypothetical events for theoretical purposes. To return to my original example, the computational algorithms that compose a model of an approaching, very real hurricane are as independent of computational substrate as is any evolutionary algorithm run as a theoretical exploration. In the instance of a hurricane simulation you will indeed want to test the predictions issued by your model against the behavior of the actual hurricane, not the least for the purpose of improving your model. Same with Gil's simulation of the process of dropping guided packages. Nevertheless, these simulations, whether of hurricane or guided package, run at a computational level that is utterly independent of the hardware that hosts those simulations. The computational substrate is not part of the simulation - nor is it the reality against which the results issued by the simulation are tested. You use substrate in the sense of the "system that is being simulated and the reality within which it is embedded." This is not the substrate to which I refer. As you say, this reality is of obvious relevance when testing for accuracy the results issued by a simulation. But this is not the computational substrate to which we are referring. BTW, the assertion that it is necessary and desirable to simulate every detail of the system and every level of the system being modeled is misleading. This depends upon the purpose served by the simulation. For example, if I am simulating airfoil shapes to determine which yield the least drag and the greatest lift, it is perfectly legitimate to omit modeling of the interior structure of the airfoil. OTOH, if I am modeling the performance under flight stress of an airfoil destined to be built and installed on actual aircraft that internal structure becomes relevant. So the level of detail required is determined by the purposes served by the simulation. And, often times, one wants to pare one's simulation down to simple terms for the purpose of better understanding the phenomenon being modeled.Reciprocating Bill
October 1, 2006
October
10
Oct
1
01
2006
06:22 AM
6
06
22
AM
PDT
Tom English I understand your frustration and but I'm not going to allow your ad hominem attacks to stand. Two were deleted. Knock it off. I'm growing very frustrated by you and others' inability to grasp the fact that models of reality need to be testable. They need to make predictions that can be tested against reality. I've given you examples of real models in biology (protein folding), mechanical design (aircraft), and electronics (microprocessors). These all model the real world and can be tested by seeing if they duplicate the results obtained in the real world. Here's yet another real, testable model that interests me: Stellar Evolution. If you don't see the difference between these, testable models of real world processes, and Avida which creates artificial laws for a world that doesn't exist in nature, then I just don't know what more I can say. You give me tinker toys and I give you the stars. The examples speak for themselves.DaveScot
October 1, 2006
October
10
Oct
1
01
2006
06:15 AM
6
06
15
AM
PDT
The computational machinery and information content of biological systems is inherent in, and quintessentially critical to, the function of the system being modeled, and therefore cannot be excluded from the effects of mutations, without the simulation being rendered completely meaningless.
Ah, Gil, but what you seem not to comprehend -- this, I think, is a genuine misunderstanding -- is that there is no absolute distinction between an analytical model and a simulation model. And if simulation models suffer the defects you say they do, then Bill Dembski's abstract models of evolution in Searching Large Spaces and The Conservation of Information are in even bigger trouble. By the way, consider that I published on conservation of information in search in 1996, ten years in advance of Bill's dissemination of "The Conservation of Information." I can tell you that the analytic models are not as detailed as the simulation models. The very reason we implement some models as computer programs is that they are not amenable to mathematical analysis. You cannot dismiss simulation models of evolution as simplistic without doing the same for Bill Dembski's and my own analytic models. Sure you want to do that? It doesn't seem wise to me.Tom English
October 1, 2006
October
10
Oct
1
01
2006
05:54 AM
5
05
54
AM
PDT
If it flies, the simulation was good. If it crashes, the simulation was bad. Everything else is irrelevant.
An extremely limited notion of what a simulation can teach us. Read post 31.Tom English
October 1, 2006
October
10
Oct
1
01
2006
05:30 AM
5
05
30
AM
PDT
It’s all well and good to model it but you still have to build the real thing to see if it works. That’s how models work. You don’t seem to understand that. Once you build it and it works as the model predicted then you have a working model, until then you have an untested model. Capisce?
Proof of principle does not require a fit to data. We often see here claims of what random mutation and natural selection cannot do, and evolutionary computation puts the lie to those claims. It does not matter one whit whether the simulations fit biological observations. If you read the writings of Bill Dembski, you will see that he understands as well as I do that the essential questions regard informational physics, not biosystems per se.Tom English
October 1, 2006
October
10
Oct
1
01
2006
05:21 AM
5
05
21
AM
PDT
bFast, You're a great example of why I can't make ID-NDE into a litmus test of the decency of a person. Thanks for your posts.Tom English
October 1, 2006
October
10
Oct
1
01
2006
05:12 AM
5
05
12
AM
PDT
The point is that the logic of algorithmic computation (including simulation) is independent of the substrate on which it is instantiated.
As an undergrad at MIT, Danny Hillis and teammates implemented a tic-tac-toe player in Tinker Toys. It was guaranteed never to lose a game. I am sure there are some here who would say, "How can tic-tac-toe be played with Tinker Toys? You have to write on a piece of paper to play the game." This reflects a common deficit in abstraction. One can as represent the game state in Tinker Toys as well as one can with marks on a piece of paper.Tom English
October 1, 2006
October
10
Oct
1
01
2006
05:04 AM
5
05
04
AM
PDT
And my last sentence was sarcasm, in case anyone misses that.tribune7
October 1, 2006
October
10
Oct
1
01
2006
04:40 AM
4
04
40
AM
PDT
bfast -- In general I agree that Gil has overstepped when he suggests that the computer running a simulation must in itself be subject to random mutation. Wouldn't that depend on what degree of evolution that you are trying to model? If one is trying to demonstrate how life occurred without intelligent design, it should be recognized as axiomatic that you can't you do so using intelligent design which obviously includes hardware and software. OTOH, if one is trying to demonstrate evolution with intelligent design, then simulation programs will make sense. Further, the more the ID the more rational the evolution. “We support the decision of our leader. Bush is a man of God who is making righteous war on our enemies” is the cry of the religious right. Where the heck are you getting this from??? Anyway, isn't the war was supposed to be some kind of neocon (Jewish) conspiracy tribune7
October 1, 2006
October
10
Oct
1
01
2006
04:38 AM
4
04
38
AM
PDT
I am beginning to think that evolution WAS entirely the loss of potentiality. There is no evidence that any contemporary organism is capable of ever becoming anything different from what it is right now. Furthermore, there is no evidence in the past for any such event even though I know it must have taken place. I conclude that the "evolvers " are all gone by the wayside and we are left with only the products, all of which are doomed to extinction. How does that grab you? "A past evolution is undeniable, a present evolution undemonstrable." John A. DavisonJohn A. Davison
October 1, 2006
October
10
Oct
1
01
2006
12:48 AM
12
12
48
AM
PDT
StephenA-
Here’s my idea for a simulation that might mimic the way that RM & NS are supposed to produce new information
In some sense, this suggestion is not very useful as long as Gil and other IDists believe that software and hardware must undergo mutation. If we treat their suggestions seriously, no simulation at all can occur. Sure, we know they're wrong, but it's also important to show them that they are wrong. Failure to do that means that they will continue to erroneously play the "simulations don't mirror actual reality" card, which may be convincing to people who don't know better. I've thought about creating evolutionary simulations which involve entire ecosystems (letting natural selection rather than an explicit goal act as the selector). Part of me thinks that it's a waste of time because I know that ID advocates will always come up with some reason to dismiss the software. To put it another way: there is no conceivable simuation which could satisfy ID advocates. IDists may disagree with this, but the suggestions that involve destruction of the computer which runs the simulation or suggestions that are far outside of our compuational power (protein folding) end up preventing any "accurate" simulation from happening at all. DaveScot -
An example of a real honest to God biological simulation is protein folding. It’s something of a holy grail.
But, that's really overkill. Sure, you can say that we need to mirror the real world all the way down to the protein-folding level. (Nevermind my earlier point that we have nowhere near the computational power to do so - the "Folding at Home" project has massive computing power being put into this, and they're nowhere near the needed computational power. An evolutionary simulation that does this multiplied by billions of organisms over billions of years is clearly out of our reach. Further, even if it were done, I suspect it would be attacked with claims that information was subtlely smuggled in, that the simulation was "intelligently designed" and therefore useless, or (as Gil claims) that the software and hardware must be subject to mutations as well.) But, it's overkill to do this if you simply want to show that RM+NS is capable of producing information. That point can be shown through simple genetic algorithms; no need for simulation down to the protein-folding level if you want to bust the argument that RM+NS can't produce information. If your argument is that something else confounds the evolutionary process somewhere else, then maybe a system capable of doing protein-folding would be useful. But, to break through the claim that "RM+NS can't produce information/CSI", it's overkill. As far as genetic algorithms are concerned, I think it's already clear that they do produce information - the main question is how do we show that to people who resist this conclusion or have intellectual hangups or misunderstandings somewhere?BC
October 1, 2006
October
10
Oct
1
01
2006
12:46 AM
12
12
46
AM
PDT
recip The substrate doesn't matter if you're not modeling reality. If you're modeling reality then the substrate of reality is a benchmark by which all others are measured. Say we're modeling an aircraft. Do you get an FAA certification based upon it flying in a computer simulation? Of course not. The model may be flawed. An actual simulation of evolution would need to begin by modeling biochemistry just as Gil's simulation begins by modeling the atmosphere. By modeling reality there then becomes a benchmark against which the simulation can be tested. An example of a real honest to God biological simulation is protein folding. It's something of a holy grail. We don't have a model that produces the folds reality produces. You have to walk before you can run. A real testable model of evolution is a long way off. We're still working on the biochemistry part. When we get protein folding licked then we can start plowing through sequenced genomes getting accurate 3D models of the proteins they produce and how those proteins behave. When we get there things will be getting really interesting in biological simulations. Avida and other digital organism programs are silly in comparison to these which actually model (or attempt to model) something real where the model can be tested. And yes, disabling the computer the simulation is running on is a silly suggestion for how to make a better model of life. The simulations he's talking about are silly to begin with since they're nothing but fantasy worlds. That's why I made a joke out of it saying another way to make it more real would be an asteroid that smashes the hardware to smithereens periodically. Are you forgetting the computer maxim - silly in, silly out?DaveScot
September 30, 2006
September
09
Sep
30
30
2006
10:50 PM
10
10
50
PM
PDT
On the other thread, Tom English wrote:
Teaching computer science students from the undergraduate to the doctoral level, I encountered quite a few who were excellent programmers, but who could not begin to comprehend the notion of a model. The concept is simply too abstract for some people. They never catch on to it.
Tom, You're absolutely right. As Gil demonstrates, some people never get it, even after having it repeatedly explained to them. Gil wrote:
One other obvious point: A simulation must accurately depict the system being modeled. The computational machinery and information content of biological systems is inherent in, and quintessentially critical to, the function of the system being modeled, and therefore cannot be excluded from the effects of mutations, without the simulation being rendered completely meaningless.
Gil, Let me make this as clear as I can: 1. Yes, the reproductive apparatus of life is (in part) an information processing system. 2. Yes, the computer and software running the simulation are also an information processing system. 3. No, they are not the same system. In an evolutionary simulation, one of them (#2) is simulating the other (#1). 4. Yes, a completely realistic simulation of evolution should include mutations of the reproductive apparatus being modeled(#1). 5. No, this does not mean there should be mutations of the computer and software running the simulation (#2). Do you see the difference? Let me try again using your example of autonomously guided airdrop payloads. 1. The guidance computer and software on one of your airdropped payloads form an information processing system. 2. The computer and software you use to do an airdrop simulation also form an information processing system. 3. They are not the same system. In an airdrop simulation, one of them (#2) is simulating the other (#1). 4. If you wanted to simulate the effects of hardware or software errors on an airdrop, you would introduce the errors into the model of #1. 5. You would not introduce errors into the computer and software running the simulation. Errors in the hardware and software of the simulator only serve to produce nonsensical simulation results. If you introduce an error into the simulator's operating system, you won't be able to trust the results of your simulation. This is just as true of an evolutionary simulation as it is of an airdrop simulation. Finally, just to really hammer the point home: the simulator and the model are separate. I can simulate a broken microprocessor chip on a perfectly functioning computer. You can simulate a failed airdrop on a perfectly functioning computer. We can simulate DNA copying errors using a perfectly functioning computer. The simulator and model exist at different levels. Errors in the model do not require errors in the simulator. Sorry to go on so long, but it really seems that the message won't get through if it isn't spelled out ultra-explicitly.Karl Pfluger
September 30, 2006
September
09
Sep
30
30
2006
09:55 PM
9
09
55
PM
PDT
StephenA - If you Google "genetic programming maze" you'll find several sites devoted to this subject. At least one seems to let you design your own maze. bFast - what's important to NS is differential survival. That is usually modelled as scoring well at some task. If only two states are allowed, you'll still get progress, but slower than with a finer grained scale. It is also possible to claim that there is front loading in the choice of instruction set (this is in referenece to GP systems, such as StephenA was speculating on). A way to avoid this is to give a GP system a random instruction set which may or may not include the operators that the researcher knows a priori are useful. All this does is slow down the system, it doesn't stop it from eventually finding better and better solutions. DS - Avida isn't a CA, AFAIK.David vun Kannon
September 30, 2006
September
09
Sep
30
30
2006
08:36 PM
8
08
36
PM
PDT
One other obvious point: A simulation must accurately depict the system being modeled. The computational machinery and information content of biological systems is inherent in, and quintessentially critical to, the function of the system being modeled, and therefore cannot be excluded from the effects of mutations, without the simulation being rendered completely meaningless. There is nothing analogous in guided-airdrop simulations. My sims have been proven to work in the real world. Bio-sims have not.GilDodgen
September 30, 2006
September
09
Sep
30
30
2006
08:28 PM
8
08
28
PM
PDT
1 4 5 6 7

Leave a Reply