Wednesday, July 13, 2005

Information From Randomness?

At the end of a previous post The Development of Information Processing, I segued into a discussion of DNA and stated:
Those who ascribe to the faith called Evolution have convinced themselves that all this complex machinery [DNA], which we have only begun to decipher, came into existence through random processes. They would like to believe that somehow information can arise out of randomness, but we who design computers know better.

Making information out of nothing is like the pseudoscience of perpetual-motion machines. These were proven to be impossible, because energy cannot be perfectly stored or transmitted. Always a little bit leaks out of the machine -- typically friction creating heat -- lost energy. In computer science and information theory, we know that likewise, information cannot be perfectly stored or transmitted. Always a little bit (or more) of error creeps in, and the data erodes.
One evolutionist, Richard Dawkins, wrote a book, The Blind Watchmaker, in which he describes a computer program and results that he claimed demonstrated that evolution was virtually inevitable. His program, sometimes called the Dawkins' Weasel Algorithm, essentially claims to create information out of randomness. (According to information theory, pure randomness is zero information.) But like those putative perpetual-motion machines once made by quack inventors, we can show that the claim is fraudulent.

We will show that (1) the program does not create information out of randomness; and (2) the program does not simulate evolution, as claimed.

Suppose I put a stencil with the words STOP HERE in front of a piece of paper and sprayed paint in the general direction of the stencil. The drops of paint fly in random directions, and some drops go through the stencil, coloring the paper behind it, and other drops are blocked by the stencil. Eventually, the words STOP HERE appear on the paper behind the stencil. Now suppose that I argue that the random paint drops have evolved into a meaningful message -- that I have created information out of randomness. You wouldn't be convinced, would you? No, it is obvious that the information on the paper came from the stencil, not from the randomness of the paint drops.

Dawkins' Weasel Algorithm essentially uses a similar principle, but is cloaked in evolutionary terminology and other description that hides the deception. It's not quite as simple as the stencil illustration, but it also uses a template that forces the random actions to create the desired result. And it's simple enough that you don't have to be a computer programmer to understand it.

The object of the program is to create, using random selections, the following phrase taken from Shakespeare's Hamlet:

METHINKS IT IS LIKE A WEASEL

It has a length of 28 characters, including the spaces. (We need to say 'characters' instead of 'letters', because space isn't a letter.) So the random process begins with a random string of 28 characters (randomly chosen from the upper and lower case letters and space), such as:

trial 01: zYODhPvNZUhwMGOBzik LTqJipFB

The first step is to compare the random string with the 'target' string (the template), and make new random choices for characters that do not match the target in corresponding positions:

template: METHINKS IT IS LIKE A WEASEL
trial 01: zYODhPvNZUhwMGOBzik LTqJipFB
trial 02: OgoRTm dnNdCvPeLJmP aWQsgraU

Notice that because one of the spaces was matched on the first trial, that position is not changed. The process described above for the first step is repeated until all positions are matched. Notice that once a match is obtained in some position, that position is no longer changed.

We show a typical sequence below, where, to save space, we show only every 10th trial after the first 10 trials:

trial 01: zYODhPvNZUhwMGOBzik LTqJipFB
trial 02: OgoRTm dnNdCvPeLJmP aWQsgraU
trial 03: QhgYWG TAZLjAshLVhU OvSylTIH
trial 04: jHQszcLeGLXy kYLXmg kzZNynvO
trial 05: mkIlsHmMjdjjzRlLcWL KyEsSZSA
trial 06: ky WAEphIFBhFfiLCnz MIwQYCJC
trial 07: kebPbNxUhLBj yoLwXC sjAOJCbk
trial 08: wCjseNNsuNWqeqULJjS ZaVDbTil
trial 09: aEAqLNMXaAhlIjTLrwx euhfhgEf
trial 10: xEfIpNNHIfSXIZZLhGL TELvHdEx
trial 20: yEQNONKa sJdIspLTrJ HhudZREA
trial 30: PEUdWNKV plNIERLhbE enWPISEG
trial 40: MEOpyNKs BVIIcSLLOE IRWEASET
trial 50: METJgNKz hTkISRLNpE v WEASEU
trial 60: METHINKF YTYISALfJE O WEASEN
trial 70: METHINKo hTNISOLuKE A WEASEx
trial 80: METHINKX DTvISfLgKE A WEASEC
trial 90: METHINKb iTyISwLvKE A WEASEk
trial100: METHINKQ IToISbLTKE A WEASEC
trial110: METHINKp ITAISrLaKE A WEASEz
trial120: METHINKS ITjISrLIKE A WEASEC
trial130: METHINKS ITMISpLIKE A WEASEx
trial140: METHINKS ITGISeLIKE A WEASEn
trial150: METHINKS ITvISELIKE A WEASEB
trial160: METHINKS ITEISwLIKE A WEASEL
trial170: METHINKS IT ISjLIKE A WEASEL
trial177: METHINKS IT IS LIKE A WEASEL

Notice that there is nothing in the rules that involves any interaction between character positions (columns). Therefore, every column is an independant random process where random choices are made until the target character is chosen. And, of course the target character is eventually chosen, else the choices are not truly random. (For example, if a die never fell with face 6 up, we would know that there was something wrong with the die.)

It is obvious where the information is coming from. In the program that I wrote to generate the above data, there is the line:

Target: string28 = 'METHINKS IT IS LIKE A WEASEL';

The information is coming from this part of the program. If I change this program line to read --

Target: string28 = 'METHINKS IT IS LIKE A RABBIT';

-- then the result of the program will change accordingly.

In Dawkins' description of the program, each step is said to simulate one generation of an evolutionary process, and the random choices are said to simulate random mutations. But the program does not simulate evolution, and isn't even an evolutionary algorithm.

No evolutionist claims that evolution proceeds until a given animal is acheived. No, the basic theory is that the random choices (mutations) survive when the animal is more 'fit' for its environment.

So let's say that grammatically correct strings with correctly spelled words simulate 'fit' animals. Spelling and grammatical errors will simulate flawed animal designs, or at least a need for improvement. So the line above labeled "trial170" might plausibly simulate such a case.

But what about trials 1 through 50? (Go look at them.) They have so many flaws -- they are so badly mangled -- that they don't even look like dead animals! So here's the key question: How did they reproduce to get to generation 50?

Furthermore, note that the data above is biased in favor of evolution because the mutation rate is way too high, and because the choices are guided by a grammatically correct template. In spite of this, the first 50 'generations' don't resemble anything that can plausibly represent an animal that can reproduce.



Sometimes it helps to look at a subject from multiple viewpoints. The discussion is continued in Dawkins' Weasel Algorithm, Revisited, which looks at it as a Markov process.

7 comments:

Anonymous said...

You are correct, Dawkins' demonstration is not a very good example. In fact, it's not even like evolution in the biological sense. It's just a hill-climbing algorithm.

Check out ev:

http://www.lecb.ncifcrf.gov/~toms/papers/ev/

It's a better example, as are Tierra, Avida, Archis, and many others.

JC said...

It's not a hill-climbing algorithm, either. I've written some of those -- one of them climbed hills in seven dimensions.

Thanks, I've downloaded ev.p and will check it out.

The biggest problem with evolution is not hill-climbing, but getting from one hill to another, and even finding the first hill.

Kagehi said...

Dawkin's uses a sentence as an example, because its is something 'everyone' can recognized as having a valid pattern. You missed the point by the simple inability to comprehend it is an analogy, not a literal version of evolution. See, the problem here is that you are using a know word as an example, just like he did, then claiming that no information was 'created' because you know its a word. So, lets say we take something else. Wander out into the woods, pick up a bunch of sticks, drop them all in a pile against the same rock, *then* spray your paint. I wouldn't be real surprised if the result didn't form a pattern roughly the shape of the outline of the sticks. Well, guess what, the real world has gravity, sunlight, rocks, mountains, oceans, and more recently, even various moving types of scenery we call 'life'. All of these things determine where a lifeform can go, what it can do, where it must find food, water, shade and protection from rain, to floods, lighting, forest fires and even stray meteorites. All of these things act like the sticks in the pile and leave behind and indelible mark on the forms of life. If a new mountain rises, cutting off them from a food source, they either find new food, evolve a means to climb the mountain or they eventually die. And this does happen, even today, producing dozens of different species, which before some natural event broke them up, where identical, but which not look significantly different, but can no longer breed with each other. See, evolution isn't, like you idiots persist in claiming, about random chance, its interactive. Animals are changed by what they can't alter, like a volcano or land slide suddenly cutting them off from half their population or their favorite food, but also in some cases by alterations to the environment 'by' them, such as wearing trails through forests or the like, which tends to make it either harder or easier for other species to move about. Evolution is genetic adaptation to a changing environment, in which all invalid changes, which don't make them survive better, get deleted.

Do you understand the analogy Dawkins tried to make? He used a 'known' and 'obvious' pattern, to show how a world that promoted that pattern would inevitably produce it. Just as a planet with land was inevitably going to produce something that walked on it, but not one that never had a land mass. ***Its not random.*** The only 'good' example would have required a book about 50 times longer to describe or a software simulation like Avida: http://dllab.caltech.edu/avida/

I suspect the real reason why pro-creationists can't grasp it is that its even easier to believe the simple idea, which requires no thought, knowledge or understanding at all, that some supreme being did it, than to understand something as simple as the mechanism Dawkins describes.

But water doesn't run uphill, why would nature or the laws of physics require something more complex? Well, it is, but only in the number of variables. Dawkins had only one in his example, "it must eventually produce the same pattern." What if the rule was, it must produce the same pattern, all synonymous patterns, in every known language (and in the correct script). Ah, see, now you are not just dealing with a simple climbing algorythm. Now its complex, and made more so since you have introduced not only the requirement for different 'sub-species' of the target phrase, but a requirement that the means by which its it displayed, i.e. the letters, also must change to fit the final requirement. But... Here is the trick, it only has to produce the correct pattern once to be successful and have it sufficiently similar to be accepted (or to survive), it doesn't matter if some or all of the intermediaries die out.

And what we see is both a lot of species with common traits that make no rational sense, like how all cats share mutations that prevent them from tasting sweets, in the exact same places (but not 100% identical changes), which leaves the genes non-functioning. Not doing something else, not partly function, not missing, but disabled. How does that make any sense? God get lazy and instead of deleting the whole thing, so cats couldn't taste sweets, or even altering 'one' gene it the start that disabled it, he not only made three changes that serve no purpose other than render the gene non-functional, but did it multiple ways, just for the hell of it? Random changes, which 'eventually' produce such a disbling effect can happen. Creating one that doesn't do a damn thing but render certain taste buds defective *and* doing it in a way that is inefficient, complicated and totally pointless way, doesn't.

The biggest problem with creationists and ID proponents is that they talk a lot about flaws, but don't know a damn thing about most of the science they claim refutes evolution. I can find crazy people that claim their are magic electromagnetic vortexes in the desert, with great healing powers, that have a more consistent understanding of science than people who believe in ID. And the creationists by themselves 'only' have, "Well, I am too ignorant to understand and too lazy to seriously study the subject, so I am just going to repeat the same BS everyone like me does, after all, my Bible says I am right!", as their basis for anything. Never mind the fact that more than half the other Christians think they are wrong.

I love jc's comment though, which amounts to, "Well, evolution is too complicated for me to comprehend and Dawkins example is too simple to be a good example, so I am by default going to go with the one explaination I don't get a head ache thinking about."

Anonymous said...

Your point about hill climbing is valid. I had my terminology confused.

The responder that discussed the fact that Dawkins was making an *analogy* also makes a valid point. Beware of reading too much into analogies. Analogies are explanatory aids; they don't prove anything, and they are often subject to misinterpretation.

On a side note, I still think Dawkins' sentence-constructing program is a shaky analogy.

On a more general note:

This "conservation of information" stuff that has come out of the creationist mileau is just silly, for a number of reasons.

I've created sequences of information in many different types of evolutionary computer simulations that exceeded Dembski's "universal probability bound" of ~500 bits. Not all of these had externally applied fitness functions. Search for Archis for example, and look at ev.

Dembski would say that these systems are synthetic. I agree. They don't prove that evolution is responsible for the biodiversity found on Earth. However, they do disprove any *strong* claim to a "law of conservation of information" or anything of a similar sort. To claim a *law* is to claim that *no system* can break that law. It doesn't matter whether the system is natural or synthetic.

Also, Dembski and others like to use SETI as an example of a scientific project that uses the "design inference" by looking for "complex specified information" (whatever that means). The only problem is that SETI hasn't worked. I like the SETI project in principle, but unfortunately I think it is fundamentally flawed. In my opinion, SETI hasn't heard anything because it is making anthrpomorphic assumptions. You can't look for evidence of design without knowing something about the nature of the designer. An alien designer is an absolute unknown. We do not know what sort of signals it might send, if any.

Creationist nonsense is actually one of the things that led me to become an atheist. It's not the only thing, but it's one of the "data points" that eventually led me to reject religious claims of truth. The arguments are so transparently dishonest that I must conclude that the individuals making them are being consciously deceptive. This, in turn, discredits religion in my eyes. What use does God have for lies?

Anonymous said...

Oops... I accidentlly hit "enter" and posted the last comment before I was done. Sorry to flood your blog with comments.

I don't want you to misinterpret the last paragraph as an attack. I actually liked your post, as you seemed to be genuinely trying to understand something. I'm not accusing you of lying, just some of the peddlers of ignorance out there who like to confuse people such as yourself who are trying to learn about how our universe works.

JC said...

Thanks, all, for your comments, even the long ones. I'll be addressing this subject further in the near future, but my computer is long overdue for serious repairs, maybe even replacement, and I've got a one-week trip ahead of me, so there may be a delay.

Anonymous said...

Calling somebody an idiot is not a scientific argument and proves that the name caller can't actually make the scientific argument. ... one tries to say randomness is not needed to drive evolution but they assume the first cell and that it can adapt - where did the ability to adapt come from - wait, where did the first cell come from? Please read the following, if you really trust the science process (including peer review) then consider the following critique:
http://www.arn.org/docs/odesign/od171/rnaworld171.htm
http://www.arn.org/docs/odesign/od171/ribo171.htm