Can Evolution be Tested on Computers?



 

Can Evolution Be Tested On Computers?

Or What You Haven’t Been Told About ‘Simulations’ of Evolution

By Les Ey,   Mar 9, 2010

There is a type of computer program that is modelled on evolutionary ideas and is known by various terms including, ‘genetic algorithm’, ‘evolutionary algorithm’ and ‘artificial life’. In this article I look at whether it is possible to use evolutionary algorithms to test Darwin’s theory of evolution. I think you will be surprised to learn that the designers of evolutionary algorithms have ‘problems’ designing software that will behave in the way that they would ‘expect’.

The Missing Engine

One of the memories that I have as a young boy growing up in Australia is of being able to buy fireworks and let them off all without adult supervision. Looking back as an adult I am not sure that was a good thing but one thing I do remember is what it is like when every so often I would light the fuse and it would hiss and glow and let off a little smoke and I would run and wait and wait and wait some more and ... nothing; just the disappointing realization that my pocket money had been spent on a dud.
That is what I think it would be like working on a genetic algorithm and hoping it would take on a life all of its own.

So to illustrate why I think that, I invite you imagine how you would feel if you went out looking to buy a car and a salesman directed you to a car that looked reasonable on the outside but when you asked to look under the hood the salesman refused to on the grounds that it would only confuse you.
What would you think if after twisting his arm, the salesman reluctantly lifted the hood only to reveal that the car had no engine?
Would you tolerate any excuses the salesman made? (Excuses like insisting that the car would get you where you wanted to go if you gave it enough time or were willing to push it there.)
Would it matter how plausible or rational the salesman tried to make his excuses sound?
What would you say if the salesman treated you like a fool when you refused to buy the car?

That is effectively what is happening when evolutionists expect you to buy the idea that Darwinian evolution has the engine able to drive the kind of change that can turn an amoeba into a man. Some evolutionists (such as Dawkins) like to emphasize that evolution is ‘not just a theory’ but a very well established ‘fact’. 1
So how can someone like me (a computer programmer, not a biologist) make the claim that evolution is like a car that has no engine?
Simple, by observing what happens when the principle of evolution is tested on computers. Just like the car that has no engine, for evolutionary ‘simulations’ to get anywhere at all requires some pushing, in fact it requires a lot of pushing.
The only problem is that the programmers who use evolutionary ideas as a model for their software don’t seem to realize just how large a push they are giving their simulations.
To make matters worse, programmers are conditioned to look for shortcuts without evaluating the relevance of those short cuts to the real world. Anyone who has encountered simulated ‘physics’ in a virtual reality computer game (especially an early one) should get my point.
To put it bluntly they are cheating and they seem to be unaware of that fact. That is a minor annoyance if you are playing a computer game but downright misleading (intentionally or not) if someone is presenting it in support of a scientific theory.
To put this in everyday terms, think about what you might do in order to test the idea that pigs can fly. You could try just observing pigs or if you think that that method is way too slow, you could try giving them a helping hand like pushing them off a cliff (with a safety net at the bottom - just in case they didn’t fly) or you could try strapping them into a remote controlled ultra-light aircraft.
Please note, I am not advocating cruelty to animals - this is a ‘thought experiment’ and it is not meant to be taken literally.
It should go without saying that my proposed experiment would be unscientific and tell us absolutely nothing about whether pigs can fly except to reveal that they are not known to fly without unreasonable interference by the investigator.
In terms of the car analogy, the car is being pushed in order to prove it really has an engine.
Well that’s still a bold assertion for a computer programmer to make. Can I back it up with relevant observation?
To answer that question, let’s start by looking at some real world observation.

If you look up at the sky just after twilight you might be lucky enough to spot a satellite. If you have navigation software on your mobile phone (or in your car), then it is working thanks to satellites. Satellites were made possible thanks to the scientific discoveries of people like Kepler, Galileo, Newton and others. Those discoveries are underlined by their real world applications. Now when we want to talk about real world observations and applications, evolutionists will insist that evolution has many real world applications. My point (at least for the purpose of this article) is not to dispute that evolution has any real world applications but that (in terms of the car analogy) there are no real world applications that demonstrate that evolution has an engine. Let me elaborate. The whole issue comes down to what creationists actually dispute. I am not aware of any informed creationists who dispute that living things change. It is the direction of the change that is disputed. Evolutionists like Richard Dawkins seem to think that creationists don’t get the concept of what an accumulation of change can do. 2 I would politely suggest that creationists do in fact realize that change accumulates but that they disagree about the direction of the change. Creationists interpret the evidence as pointing to a change that is heading in the wrong direction.
The point is that life is devolving not evolving; genetic information is being lost not gained.
The clear evidence from computer ‘simulations’ that are supposedly based on evolution is that they don’t create brand new complexity without a lot of help/interference from intelligent designers; they are cheating.
So what do I mean by cheating?

The on Switch

The cheating starts when the programmer reaches for the ‘on’ button. In the real world life has to obtain energy the hard way through complex systems like photosynthesis.
For the most part programmers of evolutionary simulations on computers don’t even try to simulate processes like that.

The ‘Life Forms’ are Tiny

Just one of the proteins used in photosynthesis would be many times larger (measured in bits of information) than an entire virtual life form in a typical evolutionary ‘simulation’.
Tierra 4 is a ‘simulation’ written by Tom Ray (a biologist). Tierra is a virtual computer that hosts virus like computer programs that are meant to ‘evolve’.
In reference to the ‘creatures’ in Tierra, Dr Rodney Brooks; the Panasonic Professor of Robotics (on leave) at MIT, wrote, ‘The genomes of his programs were four orders of magnitude [10,000 times!] smaller than the genome of the simplest self-sufficient cell.’ 6

Virtual Life Support

‘Artificial life’, (if it is fair to call it life at all), would be the equivalent of life that is wrapped up in cotton wool, placed on a drip feed and put on an iron lung. Evolutionary algorithms are intentionally designed to be ‘non-brittle’ (very hard to break) in order to avoid the possibility of problems such as global extinction.

Optimized Mutation Rates

When a programmer designs an evolutionary algorithm he/she will adjust the mutation rate so that it is not too large (a situation that would eventually cause infertility/extinction in the real world) or too low (the simulation would take too long). Our DNA has genetic information that an amoeba does not; to go from an amoeba to a man requires new genetic information. When the mutation rate is too high, it guarantees that genetic information will be lost.
Recent work by Dr John Sanford has shown that, for human populations in the real world, the mutation rate is way too high.
It is very easily demonstrated by computer models. A program called Mendel’s Accountant shows that selection can not keep up with a high mutation rate and the result is a steady decline in ‘fitness’, (in real life that would certainly result in a loss of useful genetic information). That occurs for a simulation of the human genome even using very conservative settings, based on data from research done by evolutionists.

Cheep Function

The raw computer language that a CPU understands is called ‘machine code’. It is a series of bits represented by ones and zeros which are grouped into words or ‘OP codes’. A 64 bit processor has words that are made up of 64 bits. Tierra has 5 bit words. Each word can represent data or machine instructions. A CPU’s instructions tell it to do things like add two numbers, compare two numbers or move data to a location in the memory (RAM). Even a 64 bit instruction is much smaller than the smallest protein.
For cell division, a real life organism requires many genes (each of those genes containing more bytes of information than the entire ‘genome’ of a typical artificial ‘life’ form). To do the equivalent of cell division an evolutionary algorithm might only require a single op code. Many ‘simulations’ don’t require the organism to do anything at all, they are copied free of charge by the ‘simulation’. In Tierra, a single op code called ‘DIV’ makes a ‘virtual CPU’ (presumably out of thin air) and dedicates the CPU to the child ‘life form’. All for the extremely low cost of guessing a number between 0 and 31 (a 5 bit number) and it should also be noted that these ‘simulations’ often get thousands or perhaps millions of chances to guess the right number, (while they have little chance of extinction).
In case that’s not easy enough, the ‘parent’ life form already has the DIV instruction embedded in its code and except in the case of mutation, that instruction will be copied to its ‘offspring’. By comparison it would be practically impossible to get a single functional protein in just one million random trials let alone all the proteins required for cell division. 10 (If that is what the DIV instruction is supposed to represent).

Fitness Functions

What is a fitness function? A fitness function is a part of the ‘simulation’ that is meant to play the role natural selection.
Natural selection judges a creature’s ability to survive (in an environment with limited resources) long enough to pass on its genes (hopefully more of copies of them). But what actually happens with fitness functions is that they test virtual life forms for how close they are to the ultimate objective that the evolutionary algorithm is trying to achieve.
Generally speaking, they let the ‘fittest’ ‘life forms’ live and kill off the rest of the population. In other words it gives the simulation a chance to plan ahead. As Dawkins said in reference to his own program that uses a fitness function, ‘[it] is misleading in important ways; ... Life isn’t like that. Evolution has no long-term goal’. 3
So, does that sound like natural selection? To me, it sounds more like what human breeders do when they breed livestock. As far as I am concerned, programs that use fitness functions are not testing the principle of Darwinian evolution; they are testing a cross between artificial selection and genetic engineering (on life support).
In real life selective breeding does have limited success, but only to the extent that the required genes already exist somewhere in the breeding population.
See Variation and natural selection versus evolution
The results of evolutionary simulations, when you understand them, are no more impressive than supposedly ‘self’ replicating molecules, (molecules that replicate in a laboratory with a lot of help/interference from intelligent investigators) and creationists have no trouble explaining them .

See ‘Further Reading’ at the end of this article for more detailed information on the problems with genetic algorithms.

Using Genetic Algorithms to Test Evolution

So now that I have criticized the relevance of evolutionary algorithms to the real world, can I actually expect you to think of them as a test for the principle of evolution?
Well yes, I’m convinced that when you understand what is actually happening, they reveal evolution’s lack of creativity. They reveal a failure of evolution at least in principle. That is they can test the ideas behind evolution to see whether or not the car has an engine, (i.e. if Darwinian evolution can produce complex information all by itself without unreasonable help from an intelligent designer).
In terms of my testing pig flight analogy, we unstrap the pig from the ultra light a put it back in the pig pen. We can leave it on life support because as we shall see, even under extremely favorable conditions, evolution, in principle, will fail to create any new complexity all by itself.

It turns out that evolution has already been tested and it has already failed. Tierra is an evolutionary algorithm that does not have an extremely unrealistic fitness function. 5 In Tierra it is all about virtual life forms competing for limited resources within its artificial environment, it is not really about matching an arbitrary concept of fitness.
In Tierra when the simulation started, the life forms were making copies of themselves and doing nothing else, at the end of the run the virtual life forms were making copies of themselves and apart from optimizing their genome size and copying themselves in ways that were not predicted by Tom Ray, were not doing anything new. 9 They had definitely not generated any brand new function. In other words, Tierra gave no indication that evolutionary algorithms could generate complex new function without a highly specified fitness function.
However, due to the limitations of what is considered scientific ‘proof’, I cannot categorically state that evolution has been falsified by Tierra’s failure to generate complex new function any more than I can claim that not seeing a pig fly is scientific proof that they can’t.
Tierra is not scientific ‘proof’ that evolution can’t generate new complexity but it’s a powerful observation that it doesn’t, even when it is treated very kindly indeed. There isn’t even a hint that Tierra could generate open ended evolution or new complex function (without unreasonable interference).
By now you might be wondering if there is a real problem with evolutionary algorithms, why anyone hasn’t said anything about this, like an evolutionist with a PhD. Well they have, they just don’t see it as a failure of Darwinian evolution as much as a failure to simulate it in an ‘open ended’ fashion. So as far as they are concerned the problem is with the software not the theory.
It all comes down to the kind of language that is being used.
While critics of evolution talk about new complex specified information, complex function and irreducible complexity, the writers of evolutionary algorithms are talking about ‘evolvability’ and ‘open ended evolution’.
Here is what Rodney Brooks a world leader in robotics had to say about ‘the problem’.

But somehow a glass ceiling was soon hit [by Tierra]. Despite many years of further work by Ray and others, and experiments with thousands of computers connected over the Internet, nothing very much more interesting than the results of the first experiments have come along.

...

The problem is that we do not really know why they [various evolutionary simulations] do not get better and better, so it is hard to have a good idea to fix the problem.
In summary, both robots and artificial life simulations have come a long way but they have not taken off by themselves in the ways we have come to expect of biological systems. 7

Brooks then lists and discusses a number of suspected reasons for the difficulties including lack of computing power (that was nine years ago!). 8

I am still doubtful that more computer power, by itself, will be sufficient, however. I think we probably need a few Einsteins or Edisons [intelligent designers!] to figure out a few things that we do not yet understand.

...

My hypothesis is that we may simply not be seeing some fundamental mathematical description of what is going on in living systems.

It seems that Brooks himself does not understand at least some mathematical aspect of evolution. That makes it all the more harder to take when Dawkins writes

‘It is absolutely safe to say that if you meet somebody who claims not to believe in evolution, that person is ignorant, stupid or insane (or wicked, but I’d rather not consider that).’ 11

How can people like me be expected to understand it when people like Brooks do not? But then again, what is there to understand? You take something that makes copies of itself, add slight variation (not too large) and put it in an environment where it has to compete for limited resources. None of that is missing from Tierra.

Conclusion

Even though I can’t use evolutionary algorithms to falsify evolution, there are some things that I can say with reasonable certainty.
I see this as quite a blow for ‘Chemical Evolution’, (the idea that simple self replicating molecules evolved into cells). If simple self replicating artificial life form fail to ‘[take] off by themselves in the ways [Brooks and others have] come to expect of biological systems’ then why should the public accept Chemical Evolution as plausible science. I am not sure that Eugenie Scott would want High School students to read Brooks’ book. 12

Another issue that can be noted is the amount of effort required to get genetic algorithms to work at all. Newtonian physics was proven to be an incomplete theory when Einstein discovered relativity and yet we have satellites flying overhead. Even incomplete theories have practical applications yet when it comes to simulating evolution on a computer the applications are very limited and will not do anything without a lot of unreasonable help from intelligent designers. The bottom line is if you want a genetic algorithm to do something that it has not done before, it will require a programmer to do a lot of hard work in order to achieve a modest outcome. In other words it’s much easier and efficient to write even extremely simple programs manually than it is to try to get it to evolve on a computer (and it will not work at all without a fitness function).

I can also say that even if Brooks was right and there was some magic concept that could be used to engineer evolutionary algorithms that were capable of independent design, they would do so only under very specific circumstances and as Brooks has admitted, would require ‘a few Einsteins or Edisons’ to engineer those specific conditions. All that begs the question, ‘who engineered our environment if we evolved?’ It seems obvious to me that we were created.

Imagine if Newton had been seriously mistaken about the way objects travel through the vacuum of space. I wonder how long it would have taken to abandon Newtonian physics if every single rocket kept falling out of the sky.
With all the incredibly fast computers (and clusters of computers) that are available today and after all the hours that University students and professors have spent trying to write a genuinely open ended genetic algorithm that can create complex design without the need of a human programmer, I think it is about time that people started facing up to reality. Darwin presented evolution as an explanation of how complex design in living things could exist, (without the help of an intelligent designer, i.e. God). Simulations of evolution on computers are failing to generate the very thing that Darwin was trying to explain.
What I see happening in ‘evolutionary’ simulations on computers actually gives me great confidence that we were created by an intelligent God. (Other evidences such as history give me great confidence that he introduced himself to this world in the person of Jesus Christ).
As far as I’m concerned evolution has been tested and it has failed the test, is still failing the test and will keep failing.
So next time you use your GPS navigation device, please just spare a thought for all those computer science graduates toiling away in vain.
I can’t ‘prove’ scientifically that one of them will never write a genetic algorithm that proves me wrong, but then again, pigs might fly!

Further Reading

Response to ‘The Greatest Show On Earth’

Weasel words - Creation Magazine (critique of Dawkins)

Weasel, a flexible program for invest. deterministic computer ...

Genetic algorithms - do they show that evolution works?

Genetic Algorithms and Robotic Folly

References and Notes

1. Dawkins, R. & Coyne, J. (Sep 1, 2005) One side can be wrong, The Guardian   back

2. Dawkins R. (1986) The Blind Watchmaker, Penguin Books
Accumulating change is the point that Dawkins was trying to illustrate with his ‘Monkey/Shakespeare model’   back

3. Ibid   back

A longer quote from this section follows

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

4. Ray, T. (accessed on Mar 9, 2010) Evolution, Ecology and Optimization of Digital Organisms, http://life.ou.edu/pubs/tierra/   back

5. Tierra had what Ray referred to as ‘the reaper’; it killed off older life forms and those that were not making copies of itself. Various versions had some selection based on genome size. But it is obvious that all versions failed to do anything really creative.   back

6. Brooks, R. (2002) Flesh and Machines: How Robots Will Change Us, Pantheon Books   back

7. Ibid, Brooks lists four reasons for ‘the problem’.   back

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 discounts but does not entirely rule out 1, 2 & 3 and then states a preference for point 4.]   back

8. Ibid.   back

9. Ray refers to what he sees as a ‘parasite’ and ‘host’ relationship between some of the ‘life forms’ in Tierra. In a protected operating system (like Windows) there is strict control over the way that applications interact with each other. Because Tierra was coded to be ‘non-brittle’ the programs can interact with each others in ways that would cause something called a ‘protection fault violation’ error and crash the program that attempted to ‘illegally’ access the memory of another application. While some people might consider the ‘parasite’ / ‘host’ behavior of Tierra to be a bug, in some ways it might be a fair analogy of real world hosts and parasites but comparing the complexity of biological hosts and parasites with what is happening in Tierra makes it hard to take this analogy very seriously. back

10. Materialists try to dismiss arguments like this as ‘Hoyle’s fallacy’. Yet the question still remains how the first life could come about (without a Creator), especially in the light of the difficulties that Artificial Life programmers are facing.   back

11. Dawkins R. (April 9, 1989) Review of Blueprints: Solving the Mystery of Evolution, New York Times   back

12. From Wikipedia

Eugenie Carol Scott (born October 24, 1945) is an American physical anthropologist who has been the executive director of the National Center for Science Education (NCSE) since 1987. She is a leading critic of young earth creationism and intelligent design.

Eugenie Scott is also opposed to ‘Teaching the Controversy’ (that is teaching criticisms of evolution) because it will supposedly ‘confuse the students’.   back