Does Evolution Need a Few Intelligent Designers?

This article is longer and more technical than most of my other articles but I hope that it will answer questions that people might have about so called “evolution” on computers (hopefully I will do so using reasonably simple language).
In this article I will be looking at genetic algorithms. A genetic algorithm is a type of computer program that is supposed to simulate evolution. The concept of generating new information is a key concept in this article, so I recommend that you read my introductory article on the origin of genetic information if you are unfamiliar with the concept. Throughout this article I will refer to open-ended evolution, so I will briefly explain what I mean by the term. Open-ended evolution would be evolution that could keep on evolving without hitting a brick wall (It should keep on generating new complexity indefinitely). If open-ended evolution could occur on a computer, it would not only be able to produce a wide range of “species” but it should also produce some very complicated designs (all without being given any hints of what those designs are like and without being aided by any hidden function). For an amoeba to evolve into a man you would need a considerable amount of open-ended evolution.

Designer evolution

Rodney Brooks is one of the world leaders in robotics and artificial intelligence and he is adamant that evolution is the only reasonable explanation for how we came to exist. In his book, Flesh and Machines: How Robots Will Change Us, Brooks said, “I think we probably need a few Einsteins or Edisons to figure out a few things that we do not yet understand.”
Both Einstein and Edison were very intelligent. So what is it that Brooks was referring to when he said that? Well it turns out that Brooks was talking about the problems of designing computer software in two related fields. One field was “open-ended” genetic algorithms and the other field was software that can think, reason and feel emotion like us (this is known as strong artificial intelligence).
In effect Brooks was saying that it would probably take a few (very) intelligent designers to figure out why open-ended evolution and human like intelligence is not yet seen on computers.
In this article I will mainly focus on the concept of “open-ended” evolution and look at why an ardent evolutionist would say something like that and what I feel the implications of that statement are for the theory of evolution.

Are we special?

To understand why the remark that Brooks made is a problem for evolutionists we need to understand Brooks’ point of view. In his book, Brooks has one chapter titled “We are special” which is followed immediately by another titled “We are not special”. That is a deliberate contradiction that Brooks is using to convey a paradox (the paradox exists in the way he thinks about human beings). He points out a lot of unique and remarkable things about us, but then he also puts forwards a view of people as nothing more than chemical machines that were designed by evolution (as far as Brooks is concerned).
Materialists believe that matter is all there is, that there is nothing outside of what we can see (so to speak) and so they believe that there is no spiritual realm and that we do not have a spirit that can live on after our physical brain is dead. Brooks believes that our mind, will, emotions, conscious awareness and intuition are only the result of neurons firing in our brain and nothing more. If however, we were just machines that were made by evolution then it would seem that it should be possible for us to make machines that are at least as intelligent as us. However, Brooks points out that there may be a limitation to our intelligence that prevents us from understanding our own intelligence enough to make something as intelligent as ourselves. But even if we don’t understand our own intelligence sufficiently - if evolution is as powerful as it is often portrayed - what would stop it evolving on a computer if we allowed it to do so?

Who wants to be a Billionaire?

Imagine if your computer had a system on it that could write complicated computer programs. Imagine if you were the person who engineered that system? (Or more importantly, what if you owned the royalties to it?) You could be rich, very rich! You would not have to employ people, you could just buy a few computers and let them do all the work and you would not have to pay them anything. If you were as good an entrepreneur as Bill Gates then no other software house could compete with you. So why hasn’t anyone tried this? All they would have to do is to write a computer program that could write programs of its own by either being able to think for itself or by helping other programs to evolve on computers by themselves. Well genetic algorithms are supposed to imitate evolution so lets take a look at genetic algorithms.

What is a Genetic Algorithm?

“Genetic algorithm” (GA) is another name for artificial life (ALife is an abbreviation for artificial life). For the purpose of this article I will use the terms Genetic algorithm (GA) and artificial life (ALife) interchangeably. I will avoid talking about scientific definitions of life because for this article my main concern is how information is generated and not how scientists define life.
An algorithm can be thought of as a sequence of logical steps or tasks that a mathematician or a computer program performs in order to solve a problem or perform a task. A genetic algorithm is a special type of algorithm that is written to solve problems using steps that are designed to act the way that evolution is supposed to. So a genetic algorithm is a computer program that is designed to solve problems in a way that is supposed to simulate evolution. The kinds of problems that can be solved by genetic algorithms include problems like designing efficient aircraft wings and other trial and error type problems. The program will have a population and the population will consist of a number of “solutions” or “virtual life forms” that “spawn” (or if you like give birth to) their own “offspring”. At each generation something a little different is tried and then the best solution is chosen (or the best of a limited number of solutions are chosen). All the other solutions die out or technically speaking, they are deleted. The solutions that are chosen spawn children of their own. This is repeated until the genetic algorithm finds a solution that works as it is required to.
I should emphasize the point right now that genetic algorithms do work, they do amazing things and some of those things - I wish I did not have to admit - made me feel somewhat uneasy until I thought through the real implications of what they actually did.
But in his book Darwin’s Black Box, Michael Behe puts it this way
‘You can get pretty pictures and nice games on the computer, but even most Darwinians recognize that these simplistic models are a long way from the real, complex world of biology and chemistry.’
Even though genetic algorithms can design the shape of aircraft wings and other things that would require an engineer to do a lot of trial an error type work, they don’t go anywhere near achieving the range or level of complexity that evolution is supposed to have done. I like to think of genetic algorithms as “guessing game” or “the price is right” programs. All you do is design the program so that it can calculate if something is “hot” or “cold” and let it keep trying something else until it zeros in on the answer. That’s fine for guessing games, computers can “optimise” by trial and error much faster than humans can but human engineers are much better at solving most problems than genetic algorithms are.
But, if genetic algorithms can do some of the work of a designer, doesn’t that prove evolution at least in part? Later in the article I will talk about the concept of “new information” and show that genetic algorithms don’t produce new information. Genetic algorithms use something called a “fitness function” to borrow information from the genetic algorithm itself. So that leads to the question, “What is a fitness function?”

What is a fitness function?

The concept of a fitness function is a technical concept but it is important to understand it because this is where information is hidden and, as I will show later, the problem of where the information actually comes from is a key part in answering the question of whether genetic algorithms support evolution or in fact help to question it.
So what is a fitness function? A fitness function is a module (a part of a computer program) that is used to play the role that natural selection is supposed to play in real life. The fitness function does the choosing so it decides which virtual life forms live to reproduce and which ones do not. (In most genetic algorithms it is the genetic algorithm that does the actual copying). In The Blind Watchmaker, Richard Dawkins describes a program that he wrote that he refers to as the monkey/Shakespeare model. He wrote the program to illustrate the concept of “cumulative selection”. He starts off by talking about “single step” selection. To explain single step selection, Dawkins looked at the chance that a monkey would have of typing a phrase on a specially designed typewriter that only has capital letters and a space bar. He chose the phrase, “METHINKS IT IS LIKE A WEASEL”, which comes from the Shakespearean play called Hamlet.
He calculated that the chance of a monkey matching the exact phrase in one go would be about 1/10⁴⁰. Well what if you could let the monkey keep right on typing until he matched Shakespeare’s phrase? In the book, No Free Lunch, Dembski states that to have a better than even chance the monkey would need 10⁴⁰ tries. Dawkins concluded, “To put it mildly, the phrase we seek would be a long time in coming, to say nothing of the complete works of Shakespeare.” Well let’s not put it so mildly, just how long would it take a monkey to type that reasonably short phrase? For a monkey to type 10⁴⁰ letters if it could type right around the clock at a rate of ten letters per second (with an unlimited supply of recycled paper) it would take about 3 x 10³² years, that is a 3 with 32 zeros following it. You could knock off 9 zeros of you had a billion monkeys to do your typing but that is still many times longer than evolutionists claim that the Universe has existed for.
That would be using single step selection but what is cumulative selection? That is where you might start with something random and change it a little each time and keep the changes that you like and discard the changes that you don’t like. So cumulative selection gradually builds up good changes. Dawkins shows quite clearly that it is extremely unlikely that haemoglobin can be generated by single step selection. He estimates that the chance making just one out of the 4 “chains” that make haemoglobin in one attempt would be about 1 in 10¹⁹⁰. That is an unimaginably smaller chance than the monkey had of typing Shakespeare’s phrase. It is worth noting here that a “simple” living organism is much more complicated than haemoglobin alone because it contains many times that amount of complexity. Scientist are yet to show how something could survive that is a lot less complicated than the simplest living organism that we know about today. Only “living” things can reproduce and you need reproduction for cumulative selection to work so the first living thing must have come about by single step selection (if you don’t want to accept that God created life). So for life to come into existence without God’s help would require something far more unlikely to happen than a monkey typing “METHINKS IT IS LIKE A WEASEL” by chance alone. Evolutionists try to avoid the origin of the first living cell by suggesting that there was a molecule that could make copies of itself that gradually evolved into a cell but they do not agree on what that molecule was and they don’t know how it came about. In the preface of his book titled The Origin of Life Paul Davies wrote

Many investigators feel uneasy about stating in public that the origin of life is a mystery, even though behind closed doors they freely admit that they are baffled.

The computer program that Dawkins wrote to demonstrate cumulative selection had a population of (probably) 100 “mutant ‘progeny’ phrases”. In Dawkins’ virtual world the ‘phrases’ took the place of a population of living things that can breed. Those phrases started out as a random collection of 28 consecutive capital letters and spaces. For example a run might start with the following “phrase”, “OSYTFJPOPOSICXURQZVBOOYGBWXP”.
The “phrases” do not make sense but Dawkins did not intend it to make sense at the start of a run. (Whether they should have or not was ignored by Dawkins).
That program had a “fitness function” and the purpose of the fitness function was to test to see which of the members of a virtual population of random ‘phrases’ were most like the same phrase that the monkey was supposed to type, namely, “METHINKS IT IS LIKE A WEASEL”.
Instead of trying to get the target in one go, Dawkins computer program was to accumulate changes over a number of generations until it matched the target phrase and it did that through the help of the fitness function.
Whichever one of the “mutant phrases” had the most letters that matched Shakespeare’s phrase would be chosen to breed the next generation even if it only had one letter that matched just so long as it was the best phrase.
So just supposing we started with the random phrase, “OSYTFJPOPOSICXURQZVBOOYGBWXP” and that phrase had 100 offspring and the following two offspring were compared with each other
“OSYTFNPOPOSICXURQZVBO YGBSXP” and
“OSTTFNPOPOSICXURQZVBO YGBSXP”.
When the two were compared the second one would win because the third letter is a “T” and that is a better match of “METHINKS IT IS LIKE A WEASEL”. If that “phrase” turned out to be the best match in the entire population then that would be the one that was chosen to produce the entire next generation of 100 mutant phrases. What would happen if we repeated this process over several generations?
The following shows the results of a run of a program designed to work in a similar way to Dawkins’ program

00: OSYTFJPOPOSICXURQZVBOOYGBWXP
01: OSYTFNPOPOSICXURQZVBOOYGBWXP

02: OSYTFNPOPOSICXURQZVBOOYGBSXP

03: OSYTFNPOPOSICXURQZVBO YGBSXP

04: OSTTFNPOPOSICXURQZVBO YGBSXP

05: OSTTFNPOPOSICXURQZVBA YGBSXP

06: OSTTFNPOPOTICXURQZVBA YGBSXP

07: OSTTFNPOPOTICX RQZVBA YGBSXP

08: OSTTFNPOPOTICX RQZV A YGBSXP

09: OSTTFNPOPOTICX LQZV A YGBSXP

11: OSTTFNPOPOTICX LQZE A YGBSXP

12: OSTTFNKOPOTICX LQZE A YGBSXP

13: OSTTFNKOPITICX LQZE A YGBSXP

14: OSTTFNKOPITICX LQZE A YEBSXP

15: OSTTFNKOPITICX LQZE A YEASXP

16: OSTTFNKOPITICX LQZE A YEASEP

17: OSTTFNKOPITICX LQZE A YEASEV

18: OSTTFNKOPITICX LQKE A YEASEV

19: OSTTFNKOPIT CX LQKE A YEASEV

20: OSTTFNKOPIT CS LQKE A YEASEV

21: OETTFNKOPIT CS LQKE A YEASEV

22: OETTFNKOPIT CS LQKE A YEASEL

23: OETTFNKO IT CS LQKE A YEASEL

24: OETTFNKO IT CS LQKE A WEASEL

25: OETTFNKO IT GS LQKE A WEASEL

26: OETTFNK IT GS LQKE A WEASEL

27: OETTFNK IT BS LQKE A WEASEL

28: METTFNK IT BS LQKE A WEASEL

29: METHFNK IT BS LQKE A WEASEL

31: METHFNKS IT BS LQKE A WEASEL

32: METHFNKS IT PS LQKE A WEASEL

33: METHFNKS IT IS LQKE A WEASEL

36: METHFNKS IT IS LCKE A WEASEL

37: METHINKS IT IS LCKE A WEASEL

39: METHINKS IT IS LMKE A WEASEL

41: METHINKS IT IS LFKE A WEASEL

42: METHINKS IT IS LYKE A WEASEL

44: METHINKS IT IS L KE A WEASEL

45: METHINKS IT IS LIKE A WEASEL

See how the fitness function was able to help the program to select the letters that would eventually match the target phrase. That looks impressive but there is a catch (actually there are a few but I will focus on one here). The problem is that this fitness function is nothing like “natural” selection. The fitness function knows in advance where it wants to go but natural selection does not know where it wants to go, it only knows what it has got to work with right now. That kind of a fitness function has what is called a predetermined target or goal. Dawkins himself said

“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 a distant ideal target the phrase METHINKS IT IS LIKE A WEASEL. Life is not like that. Evolution has no long-term goal.” [emphasis his]

So if Dawkins’ program was only useful for explaining cumulative selection then it is not an accurate model of what natural selection can do because natural selection does not have a crystal ball to tell it what will work better in the future. Any program that has a fitness function that drives it to a distant ideal target using a “deterministic” fitness function is “misleading in important ways.”
It’s a bit like a little boy with a toy car, having pushed his toy car; he then says with amazement, “Look Daddy, it moved!”
William Dembski often refers to “complex specified information.”
Complex specified information is a key point to this issue so I will give a quick and non-technical definition of it. The “complex” part of the term implies that the information content is not small. For instance if you were to tell me that a monkey had typed the word “CAT’, I would not be impressed because it is not complex enough since it is only three letters long and so the chances of that happening are not extremely low. Now if someone sent a radio signal that contained the Morse code for CATCATCATCATCATCATCATCATCATCATCAT, that would be far more complex but it would not be “specified” because it contains a repeated pattern. In fact the “phrase” that I referred to when I talked about
Dawkins monkey/Shakespeare model “OSYTFJPOPOSICXURQZVBOOYGBWXP” is complex because the odds of a monkey typing are actually the same as a monkey typing “METHINKS IT IS LIKE A WEASEL”. But the phrase “OSYTFJPOPOSICXURQZVBOOYGBWXP” is “unspecified” because it is “random” in other words it makes no sense and conveys no real new information.
The phrase “METHINKS IT IS LIKE A WEASEL” is specified because it is using meaningful (though somewhat dated) English words and grammar and they convey a message.
In chapter 4 of his book “No Free Lunch”, William Dembski uses mathematics and logic to show that genetic algorithms do not generate “complex specified information” but borrow the information from the fitness function. He also shows that the problem of explaining the fitness function itself is orders of magnitude more difficult to explain so there could be no way of generating a fitness function without the aid of an intelligent designer. Since the information is coming from the fitness function in a genetic algorithm, it is the fitness function that must be explained not the relatively simplistic target or goal. So genetic algorithms that have targets do not demonstrate evolution, they just hide the real issues from people who are unaware of what is really happening.
What Dawkins has failed to do is to demonstrate that there can be a “continuum” of gradual and progressive steps that can change an amoeba into a man with or without the help of a “distant ideal target”. If evolution is as powerful an idea as Dawkins would have us believe then it would be a simple matter for him to write an “open-ended” genetic algorithm that does not use either a distant idea target or a human user to decide what would be chosen by “natural” selection. So genetic algorithms that have fitness functions don’t help us if we want to learn what “natural” selection is really like.

Testing times for evolution?

One complaint against the theory of evolution is that there is no way to test it. How could you test it? After all, doesn’t it take millions of years for evolution to work? Who can wait around that long?
In his book “Darwin’s Black Box”, Michael Behe talks about a test for evolution that Darwin himself proposed,

“If it could be demonstrated that any complex organ existed which could not possibly have been formed by numerous, successive, slight modifications, my theory would absolutely break down.”

Michael Behe lists a number of Biochemical (microscopic) systems that don’t have any obvious step-by-step explanation for their existence.
Michael Behe calls these systems “irreducibly complex”, in other words they are too complicated to explain how they came about with “numerous, successive, slight modifications” because if one part is missing the entire system would not work so “natural selection” would not select such a system. Unfortunately evolutionists don’t want to admit that Behe is right so they make up “just so” stories about how the systems could have evolved without leaving any actual evidence of how they evolved.

However, isn’t all of this just academic? Isn’t there a mountain of evidence in favour of evolution such that it is no longer a question of if evolution happened as opposed to one of filling in the details of how it happened? The fossil record and things like anti-biotic resistant super bugs are two examples of the supposed evidences for evolution. I don’t have the space to directly response to those arguments but the CMI website addresses those issues and many other supposed proofs of evolution. All I will say here is that the fossil record is nothing like Darwin expected it to be like and there isn’t just one missing link, there are huge gaps in the fossil record. In diagrams of the evolutionary ‘tree of life’ dotted lines are often used to represent gaps in the fossil record. If you get some whiteout and paint over the dotted lines in the evolutionary “tree of life” the tree of life looks more like a “creationist orchard”.
Michael Behe has used the test of Irreducible Complexity to cast doubt on the theory of evolution and now I would like to suggest that there is another way that evolution can be tested. What if we could we watch evolution in action? Not just guessing game evolution but evolution that could think up new ideas all by itself and even create intelligence as smart or even smarter than human intelligence? What if we could write a computer program that could do everything that Darwin talked about? What if we could write one that did not have an artificial “fitness function”?
You would need a program that could

• Have things inside it that made copies of themselves.
• Each copy would have, on average, one slight change.
• Each “life form” would be given a certain amount of time to copy itself.
• The ones that were faster at making copies would be more likely to take over the population.

Those steps would be repeated until you saw something evolve that could do things that could not be done before the start of the program (or until there has been a reasonable number of generations for that to happen).
This could happen very quickly on a computer and soon you would have millions of years worth of generations occurring on a computer.
Sound familiar? That’s pretty much a repeat of how I described a genetic algorithm except I haven’t included a fitness function. Genetic algorithms are computer programs that are designed to simulate evolution on a computer. The main difference is that this GA would not be designed to solve problems because the only thing the virtual life forms would have to do is make copies of themselves and compete for the limited memory inside the computer.

Tierra

So now we can test evolution on a computer. We can break evolution down into its key components, “replication”, “variation” and “selection”. Replication is the part where something makes copies of itself. Variation is the part where small changes are made to the copies. Selection is where the ones that are better at copying themselves survive.
So here is a test for evolution to see if it can do what evolutionists claim it can do. We create a virtual world on a computer; in that world we introduce computer programs that make copies of themselves. We artificially introduce mistakes in the copying process (but not too many mistakes) and let them compete for limited computer memory and kill off the ones that are too old and the ones that are not copying themselves. We watch and see if they can think up new ideas all by themselves. Sound like a good idea? Well I’m sorry but it’s all a bit too late. Tom Ray has already thought of it. Who is Tom Ray? Tom Ray is a biologist who wrote a computer program called “Tierra” so that he could “study evolution” in “real time”. Tom Ray did not want to wait millions of years for evolution to take place so he hoped that he could watch it happen on a computer. If you think that Tom Ray was just a computer programmer who did not know much about evolution then you would be wrong. If you think that he is a biologist who is not a good programmer then you would be wrong about that also. Tierra is a lot more complicated than a quick and simple JavaScript or macro because Tierra simulates computer hardware. Tierra is extremely “low level”. It should be sufficient just to say that computer programmers are impressed when another programmer does something that is “low level”. Tom Ray has earned my respect as far as his programming skills go.
Well then, Tom Ray must be a billionaire, right?
Not exactly, Tom Ray has given up on Tierra and gone on to do other things.
Tom Ray’s dream of creating evolution on a computer has failed to live up to his expectations. Tom Ray wanted to create “open-ended” evolution but although Tierra would optimise the computer code by shrinking the size of the Tierran programs, (except during the runs where he deliberately rewarded the programs that grew larger), the Tierran programs would get so far and then stop doing anything new, they would hit what Rodney Brooks refers to as a “glass ceiling”. The programs that Ray placed in his Tierran world at the start of a run, would make copies of themselves and that’s all they did and by the time they had stopped “evolving” they were still making copies of themselves and that’s all they did.
Even the so-called “parasites” were just segments of computer code that copied computer code and they did that by cheating (they borrowed code from other programs and that is not really all that impressive when you understand the technical details of how they borrowed the code).
In other words Tom Ray’s Tierra programs did not think up any new ideas or what could be more correctly referred to as new “function”. It certainly did not generate what William Dembski refers to as “complex specified information”.
In case you think that my biased Creationist view can easily be ignored on this issue, I will point out that Brooks specifically mentions Tierra and refers to it as hitting a “glass ceiling”.
Brooks would not be inclined to be so pessimistic unless there were good reasons for him to be.

What’s missing?

So what is missing from Tierra? It has replication; the virtual “life” forms are computer programs that make copies of themselves. It has variation since the virus like programs gain a small number of “mutations” when they are copied. It also has (reasonably) unguided selection; the main role of the selection function is to check which life forms are making the most copies and killing off the ones that aren’t. So Tierra has all of the main components of evolution, “replication”, “variation” and “selection”. The only things missing are the open-ended evolution and an intelligent designer. Actually, I think Tom Ray is intelligent and he designed Tierra so it does have an intelligent designer. Well nothing is missing apart from the expected results, so why doesn’t Tierra work the way it should? Perhaps the environment is too simple, perhaps if you had a network of computers then the situation would be more dynamic and produce more impressive results. Nope, Tom Ray has already tried that. Well maybe there needs to be something like an environment that simulates chemistry. Tim Taylor has thought of that. In fact Tom Ray and others have tried many variations on Tierra and it is the failure of Tierra and systems like it that prompted Brooks to talk about the need for a few Einsteins and Edisons to figure out a few things.

The Alchemy of Computer Science

So will they ever have a real instance of evolution occurring on a computer?
The guys at Caltech (Adami et al) think they have and they would claim that the program that they wrote (Avida) is not a simulation of evolution but real evolution.
So are they right? Has Microsoft been put out of business yet? Have computer programmers been put out of a job? Something isn’t adding up so what is it? The whole problem has to do with “new information” or “new” function”. Creationists do not claim that things don’t change; they do not claim that things do not mutate and that the mutations are never advantageous. What they do claim is that information does not come out of nowhere. They claim that information always comes from intelligent sources and not from random mutations and they claim that the advantageous mutations that we do see really represent either a transfer of existing information or a decrease in genetic information. In other words the direction of the change is downhill and not uphill. Genetic Algorithms actually support this view. When a genetic algorithm solves a problem it doesn’t actually think up anything new by itself, it does not create new information, it borrows it from the “fitness function”. That is why people are still employed as computer programmers, they are still employed because genetic algorithms cannot come up with new ideas all by themselves, they have to be told what to do and it’s the computer programmers that have to write the “fitness function”.
Avida is not “open ended” because Avida works using complicated fitness functions.
Avida cannot keep thinking up new ideas all by itself in an open-ended fashion.
Avida is a step backwards (as far as undirected evolution goes). Avida was based on “Tierra” except they went back to using “distant ideal targets”. All that the “life forms” in Tierra had to do was to reproduce but since that did not achieve a great deal, the people at Caltech went back to using a deterministic fitness function. Bill Gates need not get too worried yet. Dembski equates “the fields concerned with the emergence of complex specified information” with alchemy. The goal to create open-ended evolution on a computer is the alchemy of computer science. You can make gold in a nuclear reactor but you will go broke doing it (It cost more to make the gold than the gold is worth). You can earn a living writing genetic algorithms that do have fitness functions but you will not get a genetic algorithm to write the fitness function for you. You could spend your entire life trying to simulate truly open-ended evolution on a computer but it will be a complete waste of time. To get a genetic algorithm to do anything useful requires a fitness function that is designed to achieve its goal and such a fitness function requires an intelligent designer.

Evolution on life support

Have the Avida guys and Tom Ray been kind enough to evolution? Perhaps the virtual environment is too harsh. Well not exactly, in fact they added a number of things that make life very easy for evolution. This is evolution wrapped up in cotton wool and put on a drip feed. The virtual life forms don’t need to digest food in order to gain energy. The entire population doesn’t have any reasonable chance of becoming extinct. The size of their “DNA” is thousands of times smaller than the DNA in “simple” bacteria.
That difference in size means it can tolerate a much higher rate of mutations than living things can and programmers exploit this by setting the mutation rates to a high setting to speed things up. If real life had a similar rate of mutation, as the ALife programs do, life would rapidly become extinct because our DNA would soon become corrupted beyond recognition.
Tierra provides us with a good example of how ALife programs make life easy for “evolution”. The Tierran environment is a virtual computer that has a population of virus like programs that run in it. Those programs use a type of “machine language” computer code. For those of you who are not programmers a “machine language” is like a very technical (or low level) nuts and bolts type of language that the computer chip itself understands.
There is a lot of hidden function available to the programs in Tierra that is accessed by the power of the instruction set that is used in Tierra’s machine language. The “Divide” instruction in Tierra tells the virtual computer to make a completely new virtual processor (a simulated computer chip) and to have the virtual processor take control of a child program. That is sort of like the equivalent of using just 3 letters of DNA to code to build all the machinery inside the cell nucleus that can make copies of the DNA and that can build important molecules inside the cell called proteins. In reality it takes thousands of genes to build all that and genes tend to be thousands of DNA letters long.
All a Tierra program needs to do to get all of that is guess a number between 0 and 31 (in the 5 bit version of Tierra). The chance of guessing the right number given a sufficient number of tries is extremely high and the reward is disproportionately high. The chance of generating the same level of function in real life using a similar number of attempts is for all practical purposes zero.
Do you think that any imaginable “primordial soup” would be anywhere near as kind to life as the ALife programmers are to their virtual programs?
As far as I’m concerned Tierra represents a test for open-ended evolution, in the most favourable conditions possible, and it has failed. I doubt that more complicated systems will produce genuine open-ended evolution since the Tierran programs failed to exploit the full potential of instruction set that was available to them. What I mean to say is that a human programmer would be able to write programs in the Tierran language that had far more complexity and that were of greater use.

What do Computer Simulations have to do with the real world?

By now it will probably have occurred to you that one major issue is that there is a large difference between what can happen on a computer and what really happens in the real world. If you have played computer games for any length of time you will have probably worked out that the best way to beat a computer is to work out the weaknesses in the AI of the computer game because you will be able to exploit those weaknesses time and time again. The computer never learns from its mistakes (at least not in any of the games that I have played) but most human players will get the idea eventually. Another issue with computer games is the “physics”, in some of the games and in some of the flight simulators the physics will be more realistic than it is in others. For instance in some flight simulators the introductory level will allow a “pilot” to do things like land an aircraft on top of a lake (and lots of other very unrealistic things). In other words the level of realism that is programmed into a simulation limits the level of realism that you experience when using a simulation. So how similar to real life are genetic algorithms to real life? The answer is obviously that genetic algorithms are not very similar to real life at all. So can we learn anything meaningful about evolution from genetic algorithms? I like to split the problem into two ways of thinking.
One of the ways of thinking is what I call “biological” relevance. The other is abstract relevance. Genetic algorithms are not very useful when it comes to using them as a support for evolution when it comes to their “biological” relevance. Even if they did manage to get open-ended evolution occurring in the ultra friendly virtual world of a GA, that would still be a long way from proving that it could happen in the much harsher real world. When I mentioned that Tierra like programs could be used to test the concept of open-ended evolution I was not thinking of its biological or (real-world) relevance, I was thinking more about its abstract relevance.
So what is abstract relevance? Abstract relevance is where you break a problem down into its key components and try to build a mathematical model. You do that in order to get a proof of concept. We saw that the key components of Darwinian evolution are replication, variation and unguided selection and that Tierra has all of those components and yet it fails to be open-ended. Tom Ray is a Biologist, he knows about the theory of evolution and he modelled Tierra on Darwinian evolution to the best of his ability, the only reason to question the accuracy of Tom Ray’s work, at least in an abstract sense, is because you don’t like the results it has produced.
If the abstract concept of evolution was ever going to produce complex specified information without the aid of a finely tuned fitness function it should have happened in Tierra. The Tierran life forms failed to think up complex ideas that were not handed to them effectively free of charge from the virtual CPU as well as the powerful instruction set. The real world is not so kind to evolution so why would one expect open-ended evolution to be more likely to occur in the real world than in a friendly virtual world?
If you would like to suggest that it is because the Tierran life forms are too simple to evolve then consider this. It is incomprehensible how extremely unlikely that even a “simple” bacterium could have arisen by chance alone so the chemical evolution people like to think of the first living thing as being a simple self replicating molecule.
The chemical evolution people want the first life form to be as simple as possible. The only problem with that is that the ALife people keep pushing the limit to how simple that can be. It seems to me that they are at odds with each other because if the ALife people can’t come up with something open ended that is relatively simple then the odds of a real-world equivalent forming by chance are extremely small. The more complex the ALife systems become (without being open-ended) the more of a concern it should be for Biologists.
Some would say that evolution only occurs in suitable environments. So that would then raise the issue of how suitable does it have to be? Is the environment as finely tuned to allow things to evolve as the physical forces (such as gravity) are? If that was the case then think about the irony in that, would that not imply that the environment was designed for evolution?
Research is being conducted into genetic algorithms and artificial intelligence in Universities all over the globe, brilliant minds like Rodney Brooks, Karl Sims, Chris Adami, Tim Taylor, Tom Ray and Paul Marrow have worked on the problems for years.
Is it not time to start asking some tough questions about evolution?

Conclusion - Never say never?

Naturally enough evolutionists have not given up on open-ended genetic algorithms and they are not likely to give up any time soon.
The theory of evolution strongly influences their thinking and they have an “a priori” commitment to a naturalistic explanation for life. The theory of evolution has had many challenges and every time evolutionists will just make up add hoc explanations to accommodate what is actually observed, so it is hard to imagine that much will change in the near future.
I want to re-iterate something here just in case I haven’t made the point clear enough. Genetic algorithms that have a fitness function can do some very useful things; in fact, if I had to write an application that favoured that approach then I would consider using a GA myself (obviously the design would include a fitness function). The only time I have a problem with genetic algorithms is when someone tries to use them as proof of evolution (whether the proof be stated or implied.) Since genetic algorithms do not generate new complex specified information, they cannot be used in support of molecules to man evolution.
But what will happen with ALife and artificial intelligence in the future? With the amount of time being spent on them, they will get better and the results will be more amazing and it will get harder and harder for creationists to analyse what is actually happening. The programmers themselves will think up new ways of introducing information (new function) into a system without realising the real implications of that to the theory of evolution. So I predict that (useful) genetic algorithms will always have fitness functions but that the programmers will get good at hiding it (even from themselves).
Two potential ways that it could be achieved is 1. Optimising the way that fitness functions are defined (using what programmers call a higher level language to describe what the goal of the GA is – this would make them easier to program) 2. Being able to introduce fitness functions as plug-ins while the GA is working in “real time”.
There may well be other ways that a GA can “learn” (extract ideas from an external source) while it is still running.
But can I categorically say that they will never invent a truly “open-ended” genetic algorithm that uses real evolution to design new ideas without any help from a programmer? The problem is that the question itself seems in some respects to be open-ended because evolutionists can just invent some mystery component that Darwin was not apparently able to recognise and tell us that a few Einsteins will figure it out in the future.

Here’s what Brooks says (In “Flesh and Machines: How Robots Will Change Us”) about the problems of building open-ended genetic algorithms and smart robots.

There are a few hypotheses we could make about what is lacking in all our robotic and Alife models.
1. We might just be getting a few parameters wrong in all our systems.
2. We might be building all our systems in too simple environments, and once we cross a certain complexity threshold, everything will work out as we expect.
3. We might simply be lacking enough computer power.
4. We might actually be missing something in our models of biology; there might indeed be some "new stuff that we need.

Brooks goes on to argue against the first three explanations and then talks about the “new stuff” that he latter refers to as some mystery thing called “the juice”.
He thinks that “the juice” might be a kind of mathematical algorithm that will require a few “Einsteins or Edisons to figure out”.

While the question about whether they can produce open-ended evolution on a computer seems open-ended in itself, it seems to me that the issue is already solved but that evolutionists don’t want to accept the implications.
Tom Ray’s Tierra or a more complex derivative of it (that does not have a fitness function) should have done something impressive by now if it was ever going to. The computing power of linked computers over the Internet or in super computers is unimaginable so that is not the issue any more. Tom Ray’s algorithm lacked none of the elements of Darwin’s theory. My conclusion is that at best Darwinian evolution is an incomplete theory and it has been demonstrated to be so on a network of computers.
Newton’s work on gravity and motion has been shown to be incomplete and yet we have satellites in space. Einstein’s theory of “General Relativity” is considered to be incomplete and yet some of those satellites are GPS satellites and GPS satellites take relativity into account when they perform their calculations. Newton and Einstein have come up with theories that work in the real world. Darwin’s theory of evolution is supposed to have explained how living things are meant to have evolved from a “simple” living cell into people and how it was able to think up new ideas along the way without the help of a few Einsteins or Edisons. Yet Brooks admits that evolution on a computer is not open-ended and he talks about the need for a few intelligent designers. How then could life have evolved in the real world without the aid of an Intelligent Designer?

In summary the situation as it stands is

No fitness function = no evolution.
No programmer = no fitness function

therefore

No programmer = no evolution.

Dr Don Batten and I wrote an article about Dawkins’ “monkey/Shakespeare” program that discusses more of the shortcomings of computer simulations of evolution.
See also “Weasel Words” and this article on Genetic Algorithms.
Dr Don Batten is a qualified scientist so he had the final say in the wording that was used in the article that he wrote together with me. I am not and have never been an accredited speaker for any creation ministry and the views that I express on this website are my own.

I believe that evolution is unnecessary since an all-knowing; intelligent Designer would not need to use evolution. I believe that the evidence points to God who created life as recorded in the Bible.
The lack of open-ended evolution in computer simulations is caused by the inability of genetic algorithms to generate new complex specified information. That confirms Dembski claims about complex specified information.
Don’t hold your breath waiting for the “new stuff” that Brooks talked about and don’t waste you life looking for the elusive pot of gold at the end of the virtual rainbow.

The only real alternative to evolution is creation. I believe that God created the Universe and that he did that so that we would have the opportunity to get to know him.
Please think about asking God to reveal himself to you. Knowledge is useful to help us understand many things about the “natural” world but we will never know everything because our knowledge has limitations. But God is not limited consequently he is able to reveal himself to you and to change your life for the better. The Bible says if you seek you will find; if you seek God with all your heart you will find him. There is evidence that Jesus was a real person and that he rose from the dead.
If you want to become a Christian please see my “How do I become a Christian?” page.

Acknowledgements.

I would like to acknowledge the inspiration and help of Dr Don Batten and Stephen Tuggy while I’ve been researching Genetic Algorithms.
William Dembski’s book “No Free Lunch” would be of interest to anyone with a high level of mathematics.