Evolution Icon Evolution
Intelligent Design Icon Intelligent Design

Methinks He Is Like a Dawkins: Prominent RNA World Researcher Michael Yarus Commits Famous “Weasel” Blunder

Casey Luskin


Recently I was at a library at the University of Washington doing some literature research, when I came across a table of books that library staff had set aside to highlight for undergraduate students. One of them was Michael Yarus’s short 2010 Harvard University Press paperback Life from an RNA World: The Ancestor Within. Yarus is one of the world’s foremost researchers of the RNA world hypothesis, and his RNA “direct templating” model has been forcefully rebutted by a paper in BIO-Complexity by Stephen Meyer and Paul Nelson. It wasn’t surprising that someone connected to the UW library liked Yarus’s book and hoped undergraduates would take notice of it. What surprised me was what I read after picking it up, and how simplistic and inaccurate Yarus’s arguments are about the origin of information in DNA.

Chapter 7 of Yarus’s book is devoted to rebutting intelligent design. Titled “Tornados in a Junkyard,” it refers to the famous comment by cosmologist Fred Hoyle about the Big Bang — a quip long misquoted as allegedly demonstrating the unlikelihood of the chemical origin of life. Yarus didn’t misquote, but the fact that he would allude to this dated argument through the title of his chapter shows how outdated his own arguments are. A few pages into the chapter we find Yarus taking on William Dembski’s arguments that “Darwinian mechanisms cannot generate the ‘specified complexity’ that we see everywhere in living things.” (p. 63) Yarus then aims to set the reader straight:

To suggest how Darwinian evolution can surf across supposed oceans of improbability, I have set up a mutation-selection demonstration. The exercise is based on an original devised by Richard Dawkins in The Blind Watchmaker (1986), but I have instead used an implementation by Rob Knight and Steve Freeland, two research evolutionary biologists.

(Michael Yarus, Life from an RNA World: The Ancestor Within, p. 64 (Harvard University Press, 2010).)

That exercise, of course, is Dawkins’s “METHINKS IT IS LIKE A WEASEL” evolution simulation, except Yarus instead uses Dobzhansky’s creed, “Nothing in biology makes sense except in the light of evolution.” Yarus explains:

We spell [Dobzhansky’s phrase] out using the 26 American letters plus a space, and we place these randomly in a string of 63 symbols. The stunning fact is that there exist 2763 = 1.5 X 1090 such possible strings … many, many more than the number of elementary particles in the known universe. So finding any one statement seems an unimaginable task. (pp. 64-65)

Yarus then triumphantly demonstrates how in a mere 120 generations, from a starting point of gibberish, he is able to evolve Dobzhansky’s entire dictum. I won’t reprint the entire 3+ pages of the stepwise evolution of Dobzhansky’s statement that Yarus’s book tediously reprints, but for demonstration purposes here are a few lines:

Generation 1: xabqwjknnefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof
Generation 2: xabqwjkniefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof
Generation 3: xabhwjkniefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof
Generation 4: xabhw kniefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof

Generation 92: n thing in biology makes sense except in the li ht of evolution
Generation 93: nothing in biology makes sense except in the li ht of evolution

Generation 120: nothing in biology makes sense except in the light of evolution

In Yarus’s own words:

How did we manage to go about 1090 times as fast as intelligent design critics suggest should be possible? The secret — as you have already realized — is in the selection, which winnows the random mutations. In particular, we have no need to, cannot, and do not take the time that would be required to visit every one of the possible states of our system of letters. Instead, we head “toward” the functional statement by accepting successively better approximations to it. This simplifying directional tendency makes our otherwise remote goal startlingly accessible. (pp. 68-69)

Folks who follow this debate closely are thinking “Really?,” and are already well-aware of the fallacy in Yarus’s argument. In short, both he and Dawkins intelligently pre-programmed their simulation to evolve certain targets. The problem isn’t the mere fact that there’s a target. Nonetheless, there’s an extremely severe problem. The problem is how he gets to the target, for the simulation wrongly assumes that all intermediate stages would provide some advantage to be preserved by natural selection, and thus gives natural selection an an all-important boost it wouldn’t have in nature.

(Also note: nobody is saying that his simulation has to go through all 1090 sequences, so he’s tearing down a straw man argument.)

Yarus’s Simulation Unwittingly Shows Why Natural Selection Fails
Everyone agrees that Darwinian selection has an easy time evolving some target sequence when each successive small change provides an advantage. But the objection to natural selection from ID proponents is not as Yarus states it. ID proponents recognize that in many biochemical systems, there may be no advantage to select until many parts (e.g. amino acids in a protein sequence) are already in place. In fact, there might not be any function at all, period, until many amino acids are fixed. This means that you can’t always break severe improbabilities into more manageable ones. Yarus’s own evolution simulation unwittingly illustrates this fact.

Consider the starting point for Yarus’s evolutionary scheme:

Generation 0: xabqwjknnefjflwlsd yeykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof

Dr. Yarus promises us that this is a “random string” of text. It sure looks that way, as it resembles no meaningful words. Spend all day trying to find meaning or function in this text, and I doubt you’ll find any. But this has drastic implications for natural selection: In a Darwinian world, this sequence would have no reason to be preserved, because it carries no meaning and provides no advantage whatsoever to the organism.

If it were preserved, that would be due to random forces like drift — but random forces do not help in the quest to reduce the odds of stumbling upon highly improbable sequences. There is no statistical reason why natural selection should preserve that sequence.

Yarus hopes, of course, that small changes leading to the target sequence will be preserved. So let’s take the next five steps in Yarus’s series:

Generation 1: xabqwjknnefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof
Generation 2: xabqwjkniefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof
Generation 3: xabhwjkniefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof
Generation 4: xabhw kniefjflwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof
Generation 5: xabhw knief flwlsd ymykwqsoxrsgrkodeqqjcfxtot yold npjloxylqpof

In each of these five steps, the text is nothing but functionless gibberish, and it carries no meaning. In the unguided world of natural selection, there would be no reason for nature to preserve any of these changes. In contrast, Yarus’s simulation is intelligently programmed to retain any changes which match the target sequence.

In fact, there is nothing resembling any meaning in the sequence until dozens of generations have passed. Darwinian evolution requires that each small step along an evolutionary pathway confer an advantage improving the odds of survival and reproduction, but here we see no meaning is conferred by the changes in these first five steps of the pathway. Many mutations are necessary before there is any meaningful function to select for. The fact that these intermediate stages are functionless gibberish demonstrates that in nature, Yarus’s evolutionary pathway would not retain these intermediate stages, and the pathway would not be viable.

Michael Behe explains why this is a problem for Darwinian selection:

[I]f only one mutation is needed to confer some ability, then Darwinian evolution has little problem finding it. But if more than one is needed, the probability of getting all the right ones grows exponentially worse.

This was the problem with Dawkins’s original simulation, and Yarus’s current one: both wrongly assume that some functional advantage exists at each small step along their evolutionary pathway, when in reality they have not demonstrated any reason for unguided natural selection to retain many of the evolutionary steps. The fact that these early steps resemble gibberish for so many generations shows precisely why Darwinian evolution cannot select for structures that provide no useful function or benefit to the organism. Natural selection might be able to fine-tune sequences which are already highly functional, but it has great difficulty evolving new functional sequences of code.

Dembski’s Peer-Reviewed Analysis of Dawkins
In 2010 William Dembski co-published a paper-reviewed paper in a mainstream computer / engineering journal elaborating on exactly what was wrong with Dawkins’s Weasel program. According to their paper, Dawkins’s algorithm can best be understood as using a “Hamming Oracle” as follows: “When a sequence of letters is presented to a Hamming oracle, the oracle responds with the Hamming distance equal to the number of letter mismatches in the sequence.” The authors find that this form of a search is very efficient at finding its target — but only because it’s preprogrammed with large amounts of active information needed to quickly find the target. (Much of this is spelled out in more detail at the Weasel Ware page on the Evolutionary Informatics Lab website.)

But as we have seen, Darwinian searches are blind to distant targets. They only see what’s right in front of them. If a slight change to the current sequence can provide an advantage, perhaps Darwinian evolution will preserve it. But when many changes are required to confer an advantage, then in the blind and unguided world of nature, Darwinian evolution gets stuck.

Yarus Responds to Objections
At first glance, Yarus seems to grasp these objections. He doesn’t cite Dembski’s paper or respond to it. (Perhaps it was published after his book went to press.) Instead, Yarus replies by stating that ID proponents “sometimes claim that this kind of result is faked because we included the target statement in our program and then intelligently chose the intermediates.” (p. 69) Well, at least he got the objection partly right. But his response makes you wonder if he really understands the problem.

He responds as follows:

This objection gets the argument backward. Even our starting statement, as indicated earlier, is one out of 1.5 x 1090. Or, to put it in other words, by induction we get a similar result no matter which statement we pick as the target. The Chomsky gibberish on the opening page of this chapter — written long before and in complete disregard of our present purposes (in order to faithfully emulate English text, but without any meaning) — is one of the statements of our Dobzhansky system and accordingly could be evolved if we so chose. (p. 69)

Oh, I see. So if you start with any gibberish and pre-program your simulation to evolve new targets, then it can evolve those targets. Yarus’s “response” doesn’t respond to the objection — namely he doesn’t explain why natural selection would preserve intermediate stages if they are gibberish, and thereby confer no benefit upon the organism.

Yarus continues:

Furthermore, we can eliminate human choice as a factor: the first English word surrounded by spaces occurs quickly, at the 19th step. We could evolve words and compose sentences without any initial target instead of targeting Dobzhansky’s aphorism. Mutation and selection achieve adaptive texts (texts that make sense) without any target.

Unfortunately Yarus provides no demonstration to show that this assertion is true. That’s probably because any realistic simulation would refute his point. The only simulation we have is his Dobzhansky simulation, where words and sentences evolve only because the simulation is pre-programmed to preserve those changes that match the target sequence. But as we saw, the fact that the intermediate stages in his simulation are functionless gibberish demonstrates that in nature, this evolutionary pathway would not be viable. Thus, the one simulation we do have refutes his point.

Rather than demonstrating the efficacy of natural selection, Yarus’s demonstration shows precisely why natural selection fails in the real world. How ironic that one of the world’s foremost researchers into the RNA world should make these elementary mistakes.


Casey Luskin

Associate Director, Center for Science and Culture
Casey Luskin is a geologist and an attorney with graduate degrees in science and law, giving him expertise in both the scientific and legal dimensions of the debate over evolution. He earned his PhD in Geology from the University of Johannesburg, and BS and MS degrees in Earth Sciences from the University of California, San Diego, where he studied evolution extensively at both the graduate and undergraduate levels. His law degree is from the University of San Diego, where he focused his studies on First Amendment law, education law, and environmental law.



intelligent designMichael Yarusmultiverseorigin of lifePaul NelsonRichard DawkinsRNA worldScientific AmericanStephen Meyer