Posted by Richard B. Hoppe on September 30, 2005 01:35 PM

Writing for the Discovery Institute, Casey Luskin has dissed evolutionary research performed using the Avida research platform. (Luskin is a new “program officer” for the DI.) As I wrote last year, computer models employing evolutionary mechanisms are a thorn (or maybe a dagger?) in the side of ID creationists. The models allow testing evolutionary hypotheses that in “real” life would take decades to accomplish or are impractical to run in wet lab or field. They also allow close control of relevant variables – mutation rates, kinds of mutations, the topography of the fitness landscape, and a number of others, enabling parametric studies of the effects of those variables on evolutionary dynamics. A number of publications using Avida (see also here) have established that it is a valuable complement to wet lab and field studies in doing research on evolutionary processes.

In his testimony at the Dover trial on September 28, Rob Pennock described a study that has particularly irritated ID creationsts, The evolutionary origin of complex features, published in Nature in 2003. In that paper Lenski, Ofria, Pennock and Adami showed that there are circumstances under which structures that meet Behe’s operational criterion for irreducible complexity (IC) – loss of function due to knockout – can evolve by random mutations and selection. Since IC is the core negative argument of ID – IC structures and processes allegedly cannot evolve by incremental “Darwinian” processes – the demonstration that they can evolve by Darwinian processes knocks out IC as a marker of intelligent design. And since IC is a special case of Dembski’s Specified Complexity, it also weakens Dembski’s core argument.

Various ID creationists have criticized the Lenski, et al., study on a variety of specious grounds, and I’ve discussed those critiques in several places, including an extended discussion here. Luskin’s critique is shallower and less informed than some I’ve read. I’ll hit a few low points in his critique.

Luskin wrote

Pennock and his other co-authors claim the paper “demonstrate[s] the validity of the hypothesis, first articulated by Darwin and supported today by comparative and experimental evidence, that complex features generally evolve by modifying existing structures and functions” (internal citations removed). Today in court, Pennock discussed the paper today asserting that it was a “direct refutation” of irreducible complexity and a “general test” of Darwinian theory.

I do not have access to Pennock’s testimony at the moment, but that’s about what I’d have said. Co-option and modification of existing structures is a ubiquitous phenomenon in evolution at levels ranging from molecular mechanisms to high-level structures like wings. And in the Lenski, et al., study, sure enough, those same phenomena were observed occurring under the Darwinian mechanisms – reproduction, heritable variation, competition, and mutation.

Luskin spent a good deal of space exploring the conjecture that Pennock’s co-authorship of the Lenski, et al., paper was a conspiracy to get an expert on Behe’s irreducible complexity involved without directly citing Behe. He wrote

I can think of no reason why a philosopher, who otherwise never authors technical papers in scientific journals, whose career specializes in rebutting ID, should be a co-author a [sic] technical research paper in a top technical science journal on the evolutionary origin of biological complexity, a claim which ID challenges, unless that paper somehow required some expertise on ID. Indeed, this paper now appears strategically arranged: is it mere coincidence that this paper appeared as a primary exhibit in the first trial against teaching ID? The reality is that Avida study, in which Pennock was third author, has much to do with strategically rebutting ID.

Looks more to me like it empirically rebuts irreducible complexity.

In his conspiracy theorizing Luskin neglected to mention that in addition to his appointment in philosophy, Pennock is also a Member of the Digital Life Laboratory at Michigan State, along with two of the other authors, Charles Ofria and Richard Lenski. Chris Adami is also associated with the Devolab as a collaborator. The work published in the Lenski, et al., paper is well within Pennock’s professional purview: it’s by four colleagues associated with the same lab. It wouldn’t surprise me at all to find that Pennock suggested the study’s main outlines to his co-authors, since IDists use the notion of irreducible complexity as their primary weapon in their culture war against evolutionary theory and Pennock is interested in that effort. Knowing something about Avida, I have no problem imagining that Pennock saw the Avida platform, a main tool in the Devolab, as an excellent tool to do some research on the question of the evolvability of IC structures.

That they didn’t mention intelligent design isn’t amazing. As far back as Darwin the question of how those kinds of structures could evolve has been raised, so Behe contributed no new issue to address. Lenski, et al., anchored their paper directly to Darwin in the first paragraph. Since the ID creationists have not published anything in the professional literature of biology to which to refer in the context of the Lenski, et al., paper, it seems strange to complain that they didn’t reference ID. If ID had some actual professional literature to cite one might sympathize, but it doesn’t. Luskin’s conspiracy theory is more than a little incongruous coming from the socio-political movement that didn’t bat an eyelash when the ID-sympathetic editor of an obscure taxonomic journal slid around the publishing society’s editorial guidelines to get Meyer’s Hopeless Monster published. Do I detect projection here?

Then Luskin repeated a common ID creationist criticism by writing

Pennock asserted on the witness stand that this study accurately modeled biological reality. Well, if biological reality was pre-programmed by an intelligence to evolve certain simple logic functions, then he’s right. Avida programmers knew that EQU was easily evolvable from the proper combination of only 5 primitive logic operations before the simulation even began. This is called “evolution by intelligent design,” because the environment seems literally pre-programmed to evolve the desired phenotype.

In fact, Avida programmers had no idea whether digital critters capable of performing input-output mappings corresponding to EQU could evolve in Avida. That human programmers could write an Avida instruction string that performed the EQU mapping is irrelevant to the question of whether it could evolve via Darwinian mechanisms. Programmers writing code is ID’s position, not evolution’s. (Incidentally, Luskin misrepresents what actually evolved in the Lenski, et al., experiments. “EQU” didn’t evolve any more than flight evolved in animals like birds and bats. Morphological structures that enable flight evolved; flight didn’t. Similarly, assembly language programs that performed the input-output mapping corresponding to EQU evolved, not EQU itself. That’s not a trivial distinction. A given function can be performed by a number of different structures.)

Further, while human programmers could write an Avidian critter to perform the input-output mapping corresponding to EQU using 5 “primitive” logic functions, the 23 lineages that evolved in the main condition in the Lenski, et al., study did so in 23 different ways. The 23 lineages that evolved to perform the mapping were all different, and in addition to EQU they performed 17 different combinations of the “primitive” functions, ranging from four to eight. None of them evolved the ‘EQU’ program the human programmers wrote. That phenomenon extends to other aspects of the Avida critters. For example, running Avida with no fitness landscape, so selection is on replication efficiency alone, evolves critters that perform self-replication in fewer instructions than any human-written program. Recall Leslie Orgel’s Second Law: Evolution is smarter than [programmers] are.

Luskin then wrote a particularly error-filled paragraph.

Pennock seemed impressed that the digital organisms “invented” many creative ways of performing EQU. But the flaw of the simulation lies therein: EQU was destined to evolve because the addition of each logic function greatly improved the fitness of the digital organism. Pre-programmed into the Avida environment were functionally advantageous digital mutations which were guaranteed to keep the digital organisms alive and doing interesting things. Thus, if one assumes that anything more than extremely minor cases of irreducible complexity never exist, then Pennock’s program show evolution can work.

There are four main problems with Luskin’s representation in those four sentences. First, “functionally advantageous mutations” were not “pre-programmed” into the Avida environment. Random mutations occurred, some of which were deleterious (in the sense of decreasing reproductive fitness) or even lethal, some were neutral, and some were advantageous. Gee. That’s just what a slew of biological research teaches us: mutations come in three basic flavors.

Second, Luskin claimed that those mutations were “… guaranteed to keep the digital organisms alive …”. That’s flatly false. Tens of thousands of digital organisms die in the course of an Avida run under the conditions Lenski, et al., used. Some die because they fail to replicate due to lethal mutations in their replication code, and some die because they’re killed – over-written – by a reproducing neighbor regardless of their advantageous mutations. Thousands of species emerge, flourish for a while, and then go extinct, and thousands of lineages go extinct. There are no guarantees at all.

Third, it is not necessary that “…the addition of each logic function greatly improve[s] the fitness …”. While the fitness landscape defined by the values assigned to the various logic functions in the Lenski, et al., study was fairly steep from simple to complex logic functions, a number of Avida runs I have done with a flatter topography produce lineages that also perform the most complex logic functions. It just takes longer because the dynamics of lineages evolving on flatter landscapes are slower. So long as there is at least some net selective advantage for performing a more complex function, more complex functions can evolve.

Finally, do we now have a distinction between micro-IC and macro-IC? Luskin’s reference to “extremely minor cases of irreducible complexity” suggests that we have to make that distinction, but where the boundary might be is not clear. Do I hear echoes of “microevolution is fine, but not macroevolution”?

I’ll note here that what one might reasonably believe to be cases of irreducible complexity, like a three-legged stool which cannot function if any of the three legs or the seat is ‘knocked out’, are no longer IC. William Dembski has recently added two new operational tests for ICness. In addition to Behe’s original knockout operational criterion, now Dembski tells us we must also determine (1) that a simpler structure cannot perform the same function, and (2) after a successful knockout one must show that no adaptation and/or rearrangement of the remaining parts can perform the original function. As Dembski tells us, that means that a three-legged stool is not IC since a solid block of wood can keep one’s butt off the ground. I have argued elsewhere that Dembski’s additional operational criteria mean the Death of Irreducible Complexity (see also here and Mark Perakh’s post here). On Dembski’s new and improved definition, not even Behe’s mousetrap is IC.

Finally, Luskin claimed that a control condition in the Lenski, et al., paper that employed a fitness landscape that was flat across logic functions except for EQU showed that

… when there is no selective advantage until you get the final function, the final function never evolved! This accurate modeling of irreducible complexity, for there is no functional advantage until you get some to some minimal level of complexity—that’s IC in a nutshell. Yet the study found that the evolution of such a structure was impossible.

Of course they claim this is what they “expected,” but without using the words “irreducible complexity,” they just demonstrated that high levels of irreducible complexity are unevolvable.

I’ll be darned. Luskin is back to If it can evolve incrementally by indirect routes involving intermediates that are themselves functional, it ain’t IC. If it can’t evolve, it is IC, and by the way, IC shows that it can’t evolve. In other words, we’re back to the “We don’t know how it could have, so it couldn’t have and therefore ID” style of argument. And there’s that “high levels of irreducible complexity” phrase – we’ve got to deal with microIC and macroIC again, too. Stuff that’s got just a little bit of IC can evolve, but stuff that’s got a whole lot of IC can’t. When did irreducible complexity become a scalar variable? Luskin must be resurrecting Behe’s even more question-begging “evolutionary” definition of irreducible complexity. I guess it’s handy to have a series of alternative definitions of a core concept ready to hand.

In a way I feel sorry for Luskin. It can’t be easy writing about genuine research when you have no clue what it did and what it means. On the other hand, he has plenty of role models for that behavior at the Discovery Institute, and I have no doubt that he’ll learn fast.