Kirk Durston and Misrepresentation of Avida

Kirk Durston wrote in his “Introduction to Intelligent Design”:

Recent computer simulations have failed to generate 32 bits of functional information in 2 x 10^7 trials, unless the distance between selection points is kept to 2, 4, and 8-bit steps.

The 2003 Lenski et al. paper on Avida is cited by Durston as supporting the quoted statement. The 2e7 number comes from the section describing how 50 runs failed to evolve the EQU function if no less complex functions were rewarded, and the 2e7 number refers to the 2.15e7 unique genotypes evaluated in those 50 runs (p.143). But the remaining numbers don’t match up to stuff in the paper. 2, 4, 8, and 32 are mentioned in the paper as values of merit awarded to organisms based on the number of NAND operations required for the task completed. Those aren’t measures of “functional information”. Durston also left out the “16” number, which corresponded to the level of merit of two other tasks that were not rewarded in that experiment, and thus their absence is misleading.

Getting to bits isn’t difficult. I’ll be using a simple approach since the Avidian programs at issue all utilize a set of 26 instructions. Any instruction could be in any position in an Avidian genome, so each instruction in an Avidian genome can be considered to contribute

– log2 (1/26) = 4.70

bits to the Avidian.

If one were trying to express “functional information” of an Avidian in bits, one might assert that five NAND instructions being necessary out of an instruction set with 26 instructions in it would give you 23 bits, or that the minimal number of instructions needed for EQU of 19 gives 89 bits, or that the reported value of 35 instructions that were necessary for EQU in the knockout experiments reported yields 164 bits, or that the 60 instructions in the first Avidian to perform EQU yields 282 bits. None of those match the “32 bits” Durston mentions, and trying to assert the 23 bit figure would require using an idiosyncratic measure. All the other possible assertions yield more bits than Durston’s quote states.

If one is trying to specify how high the bar was that Avida failed to clear in that experiment, the lowest that one might reasonably argue for would be the 89 bits that one may derive from the minimal known program using 19 instructions. That’s a lot more than the 32 bit figure asserted by Durston.

If one is trying to figure out how large an informational difference exists between programs that accomplish the various logic tasks in Avida, Durston’s statement about “distance between selection points of 2, 4, and 8-bit steps” doesn’t seem to correspond to anything there. Any single insertion or deletion changes the information content of an Avidian by 4.7 bits (when one uses the standard instruction set of 26 instructions), not the even powers of two stated by Durston. Further, there is a handy table of the shortest known hand-coded Avidian programs.

Task Shortest known program length Bits Size of Merit Reward
NOT 6 28 2
NAND 5 23 2
OR_N 6 28 4
AND 9 42 4
OR 15 70 8
AND_N 10 47 8
NOR 19 89 16
XOR 15 70 16
EQU 19 89 32

Based on the minimal program lengths, all the logic tasks are substantially more complex than Durston admits. The “32 bit” barrier Durston discusses was not reported in the experimental results that he cites. NOT and NAND tasks, at 28 and 23 bits, evolve exceedingly rapidly in Avida populations whether they are rewarded or not. I just pulled up Avida-ED, turned off all rewards but EQU, and let a 3,600 max population run go. AND_N, at minimum a 47 bit task, turned up by update 196 in the very first run I made. That it was unrewarded does not mean that it did not evolve. The only task reported not to have evolved without other tasks being rewarded was EQU, an 89 bit task. Besides not being based on anything in the cited paper, it is easy to do some runs of Avida or Avida-ED and see that Durston’s primary claim of that sentence is demonstrably false: logic tasks of greater complexity than 32 bits do evolve in Avida even if less complex tasks are unrewarded. I tried that directly in Avida-ED by turning off rewards for all sub-32-bit complexity logic tasks (NOT, NAND, and OR_N) and running it. My first run had AND_N and AND evolve by update 800, OR by update 1200, XOR by update 1345, and NOR by update 1549. A second run fixed on a population mostly doing ANT and NOR. My third run showed evolution of EQU by update 2400. All the logic tasks rewarded were over 32 bits in complexity in those runs, and none of the less complex tasks were rewarded as “steps”. There isn’t a handy tally of unique genotypes, but it can’t possibly hit 2e7 such until after 5555 updates, anyway.

Expanding on Durston’s erroneous discussion on functional distance rewarded, the differences between minimal length programs for different tasks are in {0, 5, 14, 19, 23, 24, 28, 42, 47, 61, 66} bit distances, not “2, 4, and 8 bits” as Durston mistakenly asserts. The knockout experiment reported in the paper discusses the case where a single point mutation changed an Avidian program that had previously performed the AND task into one that performed EQU instead:

Besides EQU, this genotype performed five of the eight simpler logic functions; AND was lost as a side-effect of the EQU-producing mutation, and NAND had been eliminated by the one-step-prior mutation.

Based on minimal program lengths, the step from AND to EQU is a distance of 47 bits. The Avidian also performed the NOR task both before and after the mutation that permitted it to perform EQU. A transition from performing NOR to performing EQU could be claimed to be a 0 bit distance, given that both have shortest program lengths of 19 instructions, but that was not what was observed in that case. The very source Durston cites as support rebuts his assertions.

The implication of Durston’s “unless” phrasing is incorrect as well. The 50 run experiment where only EQU was rewarded did not try out an alternative reward structure to get to EQU. Durston cannot be referring to the outcome of experiments where all nine logic tasks were rewarded because he specifically used the unique genotypes figure from the “only EQU is rewarded” experiment, and not the significantly smaller number of unique genotypes explored in getting to EQU in the main experiment (1.22e7) where all nine logic tasks were rewarded.

So about the only thing Durston managed to get right in that sentence was copying one number from the original paper, where he limited himself to one significant digit. That seems excessively non-functional.

The Lenski et al. paper does a lot more than repudiate Durston’s dolorous-but-derelict assertions, though. It demonstrates via evolutionary computation that complex functions can arise from modification of simpler precursors. Avida removes the usual mainstay of antievolutionist argumentation, that there isn’t enough information about a lineage of interest to demonstrate that only evolutionary processes need be invoked as efficient causes to get to the result. Durston essentially gives us an instance of response #4 from my 1998 essay on objections to evolutionary computation:

Natural selection might be capable of being simulated on computers, and the simulations may demonstrate good capacity for solving some problems in optimization, but the optimization problems are not as complex as those in actual biology.

This objection typically appears once the first three above have been disposed of. Computer simulation, once held to be either a potential indicator of merit or an actual falsifier of natural selection, is then treated as essentially irrelevant to natural selection. It is certainly true that computer simulations are less complex than biological problems, but the claim at issue is not that EC captures all the nuances of biology, but rather that EC gives a demonstration of the adaptive capabilities of natural selection as an algorithm.

Durston’s attempt to misrepresent a single Avida experiment of modest extent and use that misrepresentation to make a proscriptive negative claim about evolutionary processes in biology is risible.

A point to be made, though, is that evolutionary processes don’t have to be good at “poofing” things together all at once; that’s the special creation hypothesis. Many religious antievolutionists get stuck on this, thinking that unless evolutionary processes have the asserted capabilities of omniscient, omnipotent creative deities that they can’t be credited with the history and diversity of life on earth.

Update: Other places Google thinks Kirk Durston’s erroneous conclusions have propagated:

Evolution under fire? — Part 1

Mathematically Defining Functional Information In Biology

Does God Exist? – Part 2 of 3 (about 6:20)

Re: Kirk Durston on information theory

Intelligent design question

Wesley R. Elsberry

Falconer. Interdisciplinary researcher: biology and computer science. Data scientist in real estate and econometrics. Blogger. Speaker. Photographer. Husband. Christian. Activist.

8 thoughts on “Kirk Durston and Misrepresentation of Avida

  • 2009/02/23 at 7:04 pm

    Amazing how a man with so many letters after his name can get so many simple things so badly wrong.

  • 2009/02/23 at 8:10 pm

    You wrote

    The Lenski et al. paper does a lot more than repudiate Durstonís dolorous-but-derelict assertions, though. It demonstrates via evolutionary computation that complex functions can arise from modification of simpler precursors.

    It’s actually stronger than that. It demonstrates that complex functions can arise from the modification of simpler precursors that are performing different functions.

    One of the canards about the Lenski, et al, study that Salvador and others — IIRC, Royal Truman — spread is that the fitness landscape is a sort of monotonic slope leading straight up to EQU. They neglected the fact that control conditions in which one or two ‘simpler’ functions were unrewarded also evolved lineages that performed EQU, as do conditions in which three simpler functions are unrewarded (I did some of those years ago just for the fun of it). There is no straight-line hill-climbing path from just replication to EQU via some fixed set of stair steps. Get a population in which critters are performing some simpler functions, and pretty much regardless of what those simpler functions are evolution can make use of (parts of) them to create more complex functions if the selective environment rewards them with reproductive resources for doing so.

  • 2009/02/23 at 8:18 pm

    And I’ll add that Durston’s remark about the merit steps is also wrong. Since Lenski, et al didn’t vary the proportional merit awards in the 2003 paper one can conclude nothing about whether that particular merit allocation scheme was necessary. Once again back when I was debating Salvador about it on ARN I did some runs in which the merit awards were much more modest, barely sloping as a function of the number of nands per function, and while it took longer, lineages there also evolved to perform EQU.

  • 2009/02/23 at 8:55 pm


    Itís actually stronger than that. It demonstrates that complex functions can arise from the modification of simpler precursors that are performing different functions.

    Coming from the biological side, the addition there looks redundant. I certainly didn’t intend to imply anything else.

  • 2009/02/25 at 1:17 am

    Coming from the biological side, the addition there looks redundant. I certainly didnít intend to imply anything else.

    I wouldn’t suggest you did. I merely wanted to emphasize the italicized phrase for the benefit of other readers who might use the material in your post to rebut local ID creationists. Change of function in lineages that evolve complex functions is ubiquitous in evolution and the results of the 2003 paper nicely demonstrate it.

  • 2009/06/01 at 2:39 pm

    I just noticed this via Sandwalk, thanks very much for the post Wesley. Durston seems like he might be one of the bigger new kids on the ID block and has more sciency sounding arguments than , so it’s nice to see a comprehensive refutation. Will you also be dealing with the Dembski and Marks AVIDA related claims?

  • 2009/06/01 at 2:41 pm

    should have been “more sciency sounding arguments than most”.

  • 2009/06/01 at 6:46 pm

    Dembski and Marks don’t really have Avida-related claims, other than vague opinions. It’s tough to do more than point out that they based their opinions on garbage-in, garbage-out and there leave it. I’ve begun a series on the Dawkins “weasel” program to explain how it works, since the antievolutionists seem to perpetually not get it. I’m not sure that I will get enough time in a block to get to post #3 in that until I finish up my current projects here at MSU.

Comments are closed.