Monday, October 5, 2009

Resolving a moral dilemma

I made a promise to Bob Marks that I would not divulge my correspondence with him regarding drafts of the paper that IEEE SMC-A published last month. But I did not know that he and Dembski would resort to trickery to get the paper through peer review, and, after considerable agonizing, I've decided that the better course is to break my word.

(Bob has done a great deal of admirable work in computational intelligence, and I don't count him as a personal enemy. In fact, I recently joined with him in recommending an IEEE member for elevation to a higher rank.)

This is from a July 31, 2008, note that went, at his request, to his Yahoo address rather than through the (no doubt archived) Baylor email system:

Dawkins' weasel program implements a (1, λ)-ES, not partitioned search. In each generation, there are λ offspring. All points in the parent string are subject to mutation in all generations. The fittest of the offspring replaces the parent. Fitness is the number of positions containing correct letters. It is possible for fitness of the parent to decrease from one generation to the next. The mutation probability is "low," so offspring are concentrated in the neighborhood (naturally defined in terms of Hamming distance) of the parent, not uniformly distributed on a subspace of the solution space as you indicate.

I think your analysis of partitioned search makes for a good example, but you should not suggest in any way that you're analyzing Dawkins' algorithm. I would like to see you change the target sentence. In any case, the Bard (and Dawkins) wrote "methinks" as one word, not two.

It would be interesting to see an analysis of the active information of the (1, λ)-ES Dawkins used. There are plenty of published analyses of the (1, λ)-ES as a Markov process for λ = 1 [actually 2]. I'm not sure about greater values of λ. What you need is the probability that the process enters the state corresponding to the target string in Q or fewer time steps, expressed as a function of mutation probability and λ. Of course, you could fall back on simulation.

Marks did not respond to this particular note, but "ME*THINKS" turned into "METHINKS" in later drafts, and section III-F-2 of the published article addresses a (1,2)-ES [evolution strategy; sometimes it's instead "EA" for "evolutionary algorithm"] solving a restricted form of the problem solved by the Weasel program. Furthermore, section III-F-3 deals with the closely related (1+1)-ES.

It's not plausible, I think, that Marks ignored my note - he's responded to others since - and that he and Dembski subsequently happened to think of analyzing the evolutionary algorithms. The article neither gives the established names of the algorithms nor cites prior analyses in the literature. You have a good idea now of why I consider this to be academic misconduct.

7 comments:

  1. This is a Dembski standard operating procedure. It is sad that he has corrupted Marks, though.

    ReplyDelete
  2. Speaking of standard operating procedure, if Dembski is forced to concede a point you've made publicly, look for it to be treated as if either he thought of the correction himself, or that he'll find some IDC advocate he can credit with bringing the matter to his attention.

    ReplyDelete
  3. Since the man himself is posting, I'll quote myself:

    "Incidentally, neither Dembski nor Marks have any excuse for their mischaracterization of Dr. Dawkins' work. Wesley Elsberry informed Dembski of the issue back in 2000. Marks has to have known since the date of that post, October of 2007, at the latest."

    ReplyDelete
  4. For all I know, Wesley, Marks did not tell Dembski that he was getting suggestions from me, just to keep him from refusing to consider them.

    ReplyDelete
  5. Patrick,

    I'm certainly not making excuses for Marks at this point. He and Dembski both should be made to squirm a bit, if not more, for what they've done. See my new entry.

    ReplyDelete
  6. On a similar venue: I tried to contact R. Marks and W. Dembski over my concerns on their follow-up article The Search for a Search. Unfortunately, I got no reaction: I like to document that they know that there are some problematic passages in this article before it will be published.

    I wonder whether they tried to get the next article to be published in the same journal. In his pod-cast interview with Casey Luskin, William Dembski states that the article should have been already published, but that there have been delays...

    ReplyDelete
  7. I don't believe that "The Search for a Search" is within the scope of IEEE SMC-A. On his blog, Jeff Shallitt tossed off the IEEE Transactions on Information Theory as a possibility.

    ReplyDelete