Can Random Mutation Create Codes?

The following discussion shows why random processes by definition to not create information.

Main Points:

  1. The Random Mutation generator accurately simulates random mutation, using real English characters, and demonstrates the futility of creating information with random processes. Skeptics have been invited to use it to generate evolved sentences but nobody accepts the challenge.
  2. My statement that noise never improves the signal is challenged, with stochastic resonance as an example. I demonstrate that while noise in this case may boost reception of a signal it does not increase the information; in fact it decreases it.
  3. The statement is made that I am unfamiliar with evolutionary computing. I respond by asking whether random mutation is applied to the entire source code or only certain parts of the code in these programs (it’s always the latter) but I receive no response.
  4. In Claude Shannon’s information theory, noise is mathematically identical to entropy. Its damage to a signal is irreversible.
  5. All evolution is driven by intelligent processes – no exceptions.

Your random mutation generator is idiotic. Its random mutations have nothing to do with evolution. Evolution is not a random process, though it does involve some random steps. I cannot fathom how some people are unable to perceive this. Darwinian evolution is a theory of natural selection, not random selection. If you’d ever been to school you’d know this. The random mutation generator, as its name implies, simulates random mutation. It does not simulate natural selection. You’re free to provide that component yourself, and in my talk ” If you can read this I can prove God exists ” I provide real world examples where this is testable and verifiable. Google ads are a particularly useful application, and I would encourage you to use them to conduct some tests of your own. You can do this experiment at home and it’s not even dangerous. We take some text and mutate it. We take another copy of the original text and mutate it too. We do it 100,000 times. Then we take each of these once-mutated texts and mutate them. We let the fittest survive. How do they survive? Apparently the text that makes the most sense in English is the fittest. We would need somebody to decide what makes sense; in Darwinian evolution, nature decides. So out of 100,000 mutations we take 10 which actually make sense. For example ‘The quick brown box jumped over the lazy dog’ may survive. We select 10 fit versions and continue to evolve. I have no doubt that if I peform these iterations for long enough the text will mutate into the first sentence from ‘Anna Karenina’ and I have no do some natural living organism might mutate into a better one, given enough time. Try it, and do report the findings to us when you’re done. You need to examine Tom Schneider’s work showing how information can be developed by a stochastic process. The reason this works is natural selection, which is not accounted for in information theory. I quote Warren Weaver, Shannon’s co-author in “The Mathematical Theory of Communication” 1998: “Two messages, one of which is heavily loaded with meaning and the other of which is pure nonsense, can be exactly equivalent, from the present viewpoint, as regards information. It is this, undoubtedly, that Shannon means when he says that ‘the semantic aspects of communication are irrelevant to the engineering aspects.” But this does not mean that the engineering aspects are necessarily irrelevant to the semantic aspects…” “If noise is introduced, the received message contains certain distortions, certain errors, certain extraneous material, that would certainly lead one to say that the received message exhibits, because of the effects of the noise, an increased uncertainty. But if the uncertainty is increased, the information is increased, and this sounds as though the noise were beneficial!”… “Uncertainty which arises by virtue of freedom of choice on the part of the sender is desirable uncertainty. Uncertainty which arises because of errors or because of the influence of noise is undesirable uncertainty.” … “It is thus clear where the joker is in saying that the received signal has more information.” It is not reasonable to say that natural selection has no parallel in communication theory. Any signal can pass or fail based on certain criteria. The only thing one must be extra careful about is to make sure that if one is simulating natural selection, that the experiment does mimic nature and is natural, not artificial. Perry Marshall, given his background, ought to know that noise does NOT always destroy a signal. There are situations where adding noise to a system enhances its performance. See here, here, and here for examples. All of this is within the framework of communication systems – he ought to be familiar with Stochastic Resonance. As you perhaps expect, I am familiar with stochastic resonance; two more examples which you might have mentioned to further your point would be dither, used in digital recording (adding noise to the signal at the lowest bits of resolution) and bias in magnetic tape recording (mixing a high frequency with the signal when recording). A closer examination shows that stochastic resonance functions in a different context, and that strictly speaking, dither and bias result in a loss of information, not a gain. Stochastic resonance uses noise to increase the energy level of a very weak signal so that a sensing system, operating at its lowest threshold limits, can be triggered. This is entirely different from taking a signal that is already encoded and adding noise to it, so that more useful information can allegedly be derived. Strictly speaking, the encoding process here represents, if anything, an extreme loss of information, since most of the original analog data is lost in the process. The sensor can barely detect the data in the first place, and in many cases only triggers a 1 or a 0. From analog signal to mere 1 and 0 – that’s a significant loss of information, not a gain. In the case of dither, the lowest bit of the encoding system is toggling on and off, which to the human ear sounds like severe clipping distortion. (You can hear this on CD’s and digital recordings made in the early 80’s, if you turn up the volume while the song fades out.) Adding noise makes it much less objectionable to the human ear. Nonetheless, net information has been lost, not gained. In the case of tape bias, it is randomizing the hysteresis of the magnetic tape to make it more statistically linear and have less distortion. It is compensating for a weakness in the recording medium. Thus the bias is indeed useful, but it certainly doesn’t add valuable meaning to the signal that wasn’t there before. So when I say in my talk that there’s no place in the field of communication where noise is added to a signal to improve it, I admit I do have to be a little more careful. Better to have said “there is no place in the field of communication where random noise is added to a signal to increase the quality of its information.” On my website this is transcribed from a talk to a lay audience, where a discussion of stochastic resonance, dither or magnetic tape bias is not only over their heads, it is irrelevant to the concept of truly RANDOM mutation as means of increasing the useful content of coded information. Mr. Marshall should familiarize himself with Evolutionary Computing. Random mutations together with selection generate useful non-random outcomes. My own company uses evolutionary algorithms which use random mutations (a Mersenne twister generates them) and selection to produce very non-random adaptations of artificial agents that must function successfully in a complex, real world adaptive system. My clients would be very unhappy if this did not work. RBH, are these random mutations applied truly randomly, i.e. anywhere in the Source Code of the entire program, or… are they deliberately restricted to specific parts of the code? Are you mutating the source code itself, or are you mutating certain specified variables within the source code? [RBH never responded to this question. -Perry Marshall] Communication theory is not information theory. In communication theory, mutation is bad. But in biology, mutation is not always bad since natural selection is doing its work. I make no distinction between the terms “information theory” and “communication theory.” Math is math. Languages evolve over time. Of course French, Spanish and Portuguese have Latin roots, but who was the first person to speak French? At what point did they stop speaking Latin? How did one language evolve into four? When you deal with this you’ll begin to understand biological evolution too. Never in any of my posts or on my website have I made any claim that evolution did not happen, does not exist, etc. What I have said is that it is not driven by a purely random process. Certainly the evolution of French and Spanish from Latin are not random, blind processes, they’re all directed by intelligent beings. Furthermore I’m quite open to the possibility that the Antelope evolved into the Giraffe in much the same way as Latin evolved into French. I discuss this in my talk If you can read this I can prove God exists and The Atheist’s Riddle and The case for intelligent evolution. So if biological evolution works the way you say it does – the same way as the evolution of human ideas, markets, technologies and languages, then most of Neo-Darwinism is tragically wrong. Because all evolution that we experience in everyday life is driven by intelligence, not random mutation. To the extent that evolution happened, the capacity to evolve was designed in, not accidental. G.K. Chesterton, the well-known 20th century intellectual, said “The Christian is quite free to believe that there is a considerable amount of settled order and inevitable development in the universe. But the materialist is not allowed to admit into his spotless machine the slightest speck of spiritualism or miracle.” Chesterton is suggesting that battle over “creation vs. evolution” is, to some extent, a false dichotomy – except for atheists! The real issue, both scientifically and philosophically, is naturalism vs. design. Personally I am open to a number of views on the origins question. In particular I am open to propositions that can be settled by experimentation and/or mathematical principle. Random Mutations cause birth defects, tumors, cancer, death and extinction; NOT helpful adaptations. The current dogma which says random mutations drive evolution is 100% false. So the faster we can discard this wrongful notion, the faster we can get on with real research and understanding. [Note: I added most of this response after the fact, because this particular topic deserves clarification. -Perry Marshall, January 2006] Quote: “The weak part of this argument is that it “may” evolve (troublesome word, that ‘may’) and there are “theories on how it came about.” (Troublesome word, ‘theories’.) I’m asking for evidence, not theories. Note that the evidence I’m asking for here should be a whole lot easier to produce than, say, full blown abiogenesis. The example could theoretically come from any branch of science or math – and I’m only asking for one.” You have missed the point of my last post. I have put up papers that show in science, it is agreed that a code CAN evolve. An existing code certainly can evolve when guided by an externally directed process or pre-defined goal of some sort, as the Genetic Algorithm examples you cite clearly show. Perry is still claiming that mutations don’t improve information, and I have shown this argument is wrong:’t_add_information Evolution, however it may be understood, still does not explain the origin of DNA or the information in DNA. See Yockey , p. 93-113, and the quote just above.