More Articles on Evolution
Robert Berwick Responds
The Dawkins-Dennett position on evolution reduces to a single doctrinal principle: evolution is natural selection. I dub their outlook "doctrinal" because they would have us believe that we must choose between Fisherian evolution-by-gradual-natural-selection and creationism (or a straggly band of fallen angels like Gould and Lewontin cast into outer darkness). This picture of modern evolutionary biology is far from accurate, and my review of Dawkins's Climbing Mount Improbable was intended as a counterweight to it.
Dennett's and Dawkins's opening statements in their letters underscore their doctrinal conviction. Both complain about misquotation, but neither takes issue with the "plain Anglo-Saxon prose" that Dawkins quotes: that "all questions about life have the same answer." I don't see how the missing words--"though it may not always be a helpful one"--affect the essential point: the issue was precisely whether natural selection is "the answer" to "all questions about life," not whether that sole answer is helpful.
The substantive problem, as James Shapiro notes in his contribution to this forum, is that the pace of genetic discovery has stretched gene-by-gene gradualism well past the breaking point. In my review I mentioned genes of large effect as one overlooked evolutionary factor. Let me briefly describe another, more current example to indicate how non-selective forces might turn out to be necessary (critically so) in evolutionary explanation. In the December 1996 issue of Genetics and Development,1 Arend Sidow proposes that the evolution of vertebrates proceeded via two stages of complete genome doubling. Each time, the doubling admitted a period of required neutral evolution, the chance accumulation of developmentally advantageous mutations, racing against the clock of chance degradation--to give us the "slack" developmental hardware that was exploited for the far richer developmental programs of (jawed) vertebrates. Sidow even works out a scenario with known regulatory genes as to how this system might lead to fins and vertebrates--a story that relies on neutral evolution, major-effect genes, and interaction. A "Just So" story? To be sure. But no more so than the ectoderm-to-eye simulations that Dawkins endorses.
Alongside the misquotation issue, Dawkins indicts me for a second hanging offense, with no more substance than the first. When I used the term "null hypothesis" I meant null hypothesis--in exactly the statistical sense that Dawkins recites. I did so to emphasize the point that, unlike Dawkins, working evolutionary biologists recognize that forces other than natural selection must be considered and rejected via explicit statistical methods before one sends out a paper to Evolution arguing for directed selection, lest it be returned rejected. There are nonselective forces in evolution. How strong they are remains a question.
Coming now to Dennett: once we cut through all the overgrown prose, we find two substantive points, one concerning my "strange novelty" about the computability of ontogenesis and the other on the "oft-heard but fatally ambiguous" claim that the rewound evolutionary tape won't play back the same way twice. Consider them in turn.
The computability of ontogenesis is neither novel nor strange--rather, it is an important open question of current biological interest. You needn't take my word for it. Developmental biologist Lewis Wolpert--who pioneered writing programs to build organisms and was a key figure in the discovery of morphogen--kicked off an "Open Questions" column series in the January 1996 issue of Current Biology by asking "what are the really interesting questions to answer in biology" and stating that if there were just one question he could have answered by an oracle, it would be this: "Is the egg computable?" He continues: "In other words, is there some computation that would enable us to predict the outcome of embryonic development? . . . I await the answer with little optimism." Responding to this column, E–rs Szathmáry--co-author with John Maynard Smith of The Major Transitions in Evolution--rejoined, "Wolpert is rightly sceptical about the computability of embryogenesis but I am not entirely sceptical." So are Wolpert and Szathmáry now to be cast into outer darkness?
The oracular Dennett thinks he knows the answer to Wolpert's question. Since the answer is worth a trip to Stockholm, we should see what evidence he provides. It's that DNA is a digital code and that "cell division is recursive." I have stared long and hard at the quoted phrase, but simply cannot figure out what it means. (I am a professional computer scientist; recursive functions are my business.) Perhaps Dennett is referring to our (still quite incomplete) understanding of part of the biochemical feedback/checkpoint system ensuring that the DNA of a (eukaryotic) cell will duplicate itself exactly once during mitosis, as James Shapiro describes in his letter. In yeast (Schizosaccharomyces pombe) there are sites sprinkled along the DNA coil every 86kb or so that can be "unwound" by attaching another chemical, and these "unwinders" are fired off by another chemical that must accumulate up to a certain critical level, with this triggering chemical becoming digested at the next phase of cell division. Nobody really knows if even this part of cell division is recursive. If we culture a single cell then, say, 20 minutes later, we generally have two. Those two divide again--but not exactly 20 minutes later, say 18 mins and 21.2 mins. And so on (sometimes--let's forget about cell death for now). Unless Dennett wants to now say something bizarre about time and the density of real numbers, we are simply in the same mathematical boat I described in my review: a full model of evolution has to include passing from the infinite but countable space of possible gene sequences, out to gamete formation and fertilization (in "Animal Space"), back again to genotypes and the cell, then out once more through ontogenesis to the organism that actually gets selected. No one knows what any of these mappings look like in detail, so no one knows whether they are computable--not even Dennett.
As to the "fatal ambiguity" about replaying the evolutionary tape: Dennett's remarks suggest that he has figured out all of developmental biology, and not only cell division. As Gould makes clear in his recent book Full House, "if we could replay the game of life again and again . . . the inhabitants of this region of greatest complexity would be wildly and unpredictably different in each rendition--and the vast majority of replays would never produce (on the finite scale of a planet's lifetime) a creature with self-consciousness. Humans are here by the luck of the draw. . . ."2
Now, if Dennett really thinks the great evolutionary VCR in the sky would replay anywhere close to creatures with self-consciousness, he should let us in on what he knows that is more than what every other developmental biologist in the world knows--it is worth a second trip to Stockholm. Consider what Dennett claims to understand: that an immensely complex dynamical system, whose properties we don't even know (we can't even write down the math for it) will always evolve along trajectories to roughly the same very, very small regions of an immensely large space. But why should we believe any of this? Aside from convergent evolution--"multiple solutions" between already highly evolved, hence already channeled organisms--Dennett gives no reason at all. Perhaps there are incredibly strong "physical channel" constraints ("magnets" or "attractors" in Animal Space); or, stronger than that, since we are talking beyond even animals, constraints on possible "Living Organism Space" (and so not necessarily even RNA-DNA based life). But Dawkins and Dennett are not prepared to buy such constraints. (To his credit, Dawkins says "I have an open mind about this controversy although I lean in one direction.") Convergent evolutionary examples won't wash here, precisely because the immense conservatism of evolution means that we are already talking about a single, highly restricted trajectory.
In fact, all the evidence thus far shoots the other way: all our experience with simulated evolution--from Dawkins's "biomorph" programs where he offered prizes to those who could figure out ways to actively select for interesting organism shapes, all the way to the sorry experiences with "artificial life" that Berlinski notes--demonstrates how hard it is to get anywhere without doing artificial selection or building in the solutions we want. Rather, just as Gould says, if we picture bacteria at the "left wall" and evolution spinning out creatures to the right, then we are on a special drunkard's walk far,
far out on the very, very long spindly end of a very special right-hand tail. You can't have your tape and rewind it too.
I began by criticizing doctrinal approaches to evolution and have since been emphasizing open questions, and that is how it should be: No one knows the whole truth about evolution, and my review of Dawkins did not pretend to such knowledge. But surely by now we can do better than serving up warmed-over Fisherian gradualism. Indeed, it is ironic that the very champions of the purely gene-centered view, who know very well that genetic variation provides the jet fuel that evolutionary change burns, continue to adamantly reject even the smallest deviation from Fisher's canon. Inbreeding is, after all, the royal road to extinction.
1 Arend Sidow, "Gen(om)e Duplications in the Evolution of Early Vertebrates," in Current Opinion in Genetics and Development 6, no. 6 (December 1996): 715-22.
2 Stephen Jay Gould, Full House: The Spread of Excellence from Plato to Darwin (New York: Harmony House,1996), p. 175.