|
Author
|
Topic: Eric Anderson: Bits, Bytes and Biology: What Evolutionary Algorithms (Don’t) Teach...
|
Moderator
Administrator
Member # 1
|
posted 03. February 2005 13:39
Bits, Bytes and Biology: What Evolutionary Algorithms (Don’t) Teach Us About Biology
by Eric Anderson
Abstract: Far from constituting a devastating critique of irreducible complexity, the evolutionary algorithm, Avida, is a flawed effort that bears little relevance to the biological world. In their haste to affirm the Darwinian creation story, the Avida authors seem oblivious to, or conveniently ignore, the fact that they have incorporated as premises the very conclusions they are trying to reach. Such efforts are at best misleading, at worst deceptive. Ironically, the main piece of data obtained by the Avida researchers that is not based on circular evolutionary assumptions, upon closer inspection supports, rather than refutes, Behe’s notion of irreducible complexity.
To read the entire paper, click here.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 03. February 2005 21:00
Into the (non-existent) breach once more. I have neither the time nor inclination to do a thorough analysis of all of the paper's s misconceptions, so I will confine myself to a few remarks on some of the more obvious and foundational errors in the paper.
First, a point of partial agreement with Anderson: Anderson wrote quote:
The authors begin by declaring their allegiance to Darwinian evolution in the following terms: “Charles Darwin’s theory of evolution, including its intertwined hypotheses of descent with modification and adaptation by natural selection, is widely regarded as one of the greatest scientific achievements of all time.” (p. 3)
Compare the characterization ("allegiance") with the statement of (supposed) fact quoted. It is a fact that the theory of evolution is accepted as the best available account of biological diversity by nearly every scientist working in evolutionary biology, paleontology, and their allied disciplines. The phrase "greatest scientific achievement of all time" seems to me to be hyperbole. Einstein's annus mirabilis is to my mind a greater achievement.
Now to more substantive stuff.
Anderson wrote quote:
By executing certain functional operators, “organisms” are rewarded with additional energy, which allows them – under traditional Neo-Darwinian doctrine – to reproduce faster, which, notwithstanding contrary evidence, is for the Neo-Darwinist the great and ultimate goal of all creatures biological. (p. 3)
Anderson here misrepresents evolutionary theory. Relative reproductive success in lineages, not raw rate of reproduction, is the primary variable implicated in evolution: see the distinction between r-selection and K-selection. In some conditions, high reproductive rate is indeed important, in others, reproductive success with fewer offspring is important. But Anderson's phrase "to reproduce faster" is a misrepresentation.
Anderson wrote quote:
Avida was programmed so that a slight, successive cumulative pathway to the ultimate complex function existed. In other words, the researchers assumed that the ultimate complex feature was not irreducibly complex, and wrote their program in such a way as to guarantee that it would not be irreducibly complex, before they even ran the very first simulation. Thus, it is deeply troubling to find the authors suggesting that Avida demonstrates that complex systems are not irreducibly complex. (p. 4, italics original)
In fact they didn't "program" it that way. Avida has provision for altering the selective environment to test various configurations of selective environments. They ran dozens of experimental and control conditions in different selective environments to investigate the conditions under which irreducibly complex structures might evolve.
That last quotation contains a notion that is echoed soon thereafter. To repeat, Anderson wrote quote:
Avida was programmed so that a slight, successive cumulative pathway to the ultimate complex function existed. In other words, the researchers assumed that the ultimate complex feature was not irreducibly complex, and wrote their program in such a way as to guarantee that it would not be irreducibly complex, before they even ran the very first simulation. (p. 4, italics original)
That says as plainly as possible that in order to be called "irreducibly complex" we must know that there is no possible evolutionary pathway to it.
And two paragraphs later quote:
Evolutionary algorithms that assume a cumulative or non-cumulative pathway at the outset simply cannot, by definition, demonstrate whether the complex system is irreducibly complex. (p. 4, RBH bolding, italics original)
In each case, Anderson is defining IC to be a system to which there is no evolutionary pathway. But that is not Behe's original definition of IC nor is it Dembski's most recent revised definition of IC. Neither make any reference to whether there is or is not a cumulative evolutionary pathway. Anderson is using a definition of "irreducible complexity that does not match any offered by Behe or Dembski. In footnote 6 (p. 2) Anderson wrote quote:
As used in this essay “irreducible complexity” matches the irreducible complexity initially proposed by Behe and refined by Dembski, meaning that there is no cumulative pathway to such complexity. I have shown in my article “Irreducible Complexity Reduced: An Integrated Approach to the Complexity Space,” (see www.evolutiondebate.info) that this is not a logical requirement of the concept of irreducible complexity, and I term Behe’s and Dembski’s approach “per se irreducible complexity.” Avida targets this per se irreducible complexity, and it is sufficient for our present purposes.
What is Behe's original definition of irreducible complexity? It is quote:
A single system composed of several well-matched, interacting parts that contribute to the basic function of the system, wherein the removal of any one of the parts causes the system to effectively cease functioning. (Darwin's Black Box, 39)
Nothing in that definition about whether a "cumulative pathway" exists. And Dembski's most recent revision of "irreducible complexity" is this: quote:
A functional system is irreducibly complex if it contains a multipart subsystem (i.e., a set of two or more interrelated parts) that cannot be simplified without destroying the system’s basic function. I refer to this multipart subsystem as the system’s irreducible core. (p. 2, italics original)
Refining that definition, Dembski tells us quote:
For an irreducibly complex system, each of the parts of the irreducible core plays an indispensable role in achieving the system’s basic function. Thus, removing parts, even a single part, from the irreducible core results in complete loss of the system’s basic function.
...
To determine whether a system is irreducibly complex therefore employs two approaches: (1) An empirical analysis of the system that by removing parts (individually and in groups) and then by rearranging and adapting remaining parts determines whether the basic function can be recovered among those remaining parts. (2) A conceptual analysis of the system, and specifically of those parts whose removal renders the basic function unrecoverable, to demonstrate that no system with (substantially) fewer parts exhibits the basic function. Indispensable parts identified in step (1) and then confirmed in step (2) to admit no simplification belong to the irreducible core of an irreducibly complex system. (pp 4-5)
Neither Behe's nor Dembksi's definition (which I critiqued here and here) make any mention of the evolutionary history (or lack thereof) of the structure in question. Both are phrased in terms of the current appearance and operation of the system and, in the case of Dembski's revision, a set of potential alternatives that does not have to exist but merely must be subject to a "conceptual analysis." So Anderson's complaint that Lenski, et al., "assumed that the ultimate complex feature was not irreducibly complex" by testing selective environments with different combinations of selectable 'simpler' functions is false unless Anderson defines "irreducibly complex" as "unevolvable by any route except a giant jump from individual components to final assembly." But that of course begs the question. It also refutes Anderson's claim that evolutionary algorithms cannot be useful in determining whether IC structures can evolve. Anderson claims quote:
Evolutionary algorithms that assume a cumulative or non-cumulative pathway at the outset simply cannot, by definition, demonstrate whether the complex system is irreducibly complex. (p. 4, RBH bolding, italics original)
But that's true only if "irreducible complexity" is defined in terms of evolvability, which it is not.
Anderson's essay badly mistakes the question that was asked in the Lenski, et al., paper. It was (in effect) 'What are some conditions under which structures that meet Behe's original definition of irreducible complexity might evolve?' The assembly language programs that evolved in the Lenski, et al., study are clearly irreducibly complex in the operational definition of Behe's original formulation, a knockout criterion. They evolve in a selective context where some 'simpler' logic functions are reproductively advantageous, but no particular single 'simpler' function and no pair of 'simpler' functions is necessary in order for digital entities capable of performing an input-output mapping corresponding to EQU to evolve. Moreover, analyses of the actual evolutionary sequences show that there is no single pathway defined by the various selective environments they studied. Irreducibly complex assembly language programs evolved by varied and circuitous routes. There is an embarrassment of riches in the Lenski, et al., data.
By using a definition of "irreducible complexity" that posits that there must no evolutionary pathway to an IC structure, Anderson flatly begs the question of whether IC objects can evolve. He has placed ID in the box of having to demonstrate that there is no conceivable direct or indirect evolutionary pathway to a given structure in order to classify it as IC. If it adopts Anderson's view, ID must (once again) attempt to establish a universal negative. Of course, it simplifies life for IDists in one way, since one need no longer bother with notions like "well-matched parts" and "irreducible core" and "conceptual analysis." Just define IC as "unevolvable" and one can jettison all that unnecessary stuff.
RBH
Added in later edit: If you examine the complete line of descent of a lineage provided here, it is obvious that it does not represent evolution by "cumulative complexity" in Dembski's sense.
Later edit to fix a mis-statement.
[ 04. February 2005, 13:26: Message edited by: RBH ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 08. February 2005 04:04
Just for the heck of it I'll comment at more length on Anderson's so-called "evolutionary assumptions". Anderson's "assumptions" are italicized.
1. There is a cumulative pathway to complexity.
Nope. The study (Lenski, et al.) investigated conditions under which "complexity" could evolve. It ran control conditions for that purpose. See my extended remarks above.
2. Relatively few changes are required to get from the initial organism to the complex feature.
In what units of analysis? Scaled to the size of the original replicating code in the Ancestor (15 instructions/genes), a whole lot of individual mutations intervened between Ancestor and a string of code capable of performing EQU. The range was 51 to 721 mutational "steps". Scaled to operations (AND, OR, NOR, etc.), the various runs produced code capable of performing EQU in around eight 'steps' on average (XOR almost always appears later than EQU). Compare that with Nick Matzke's description of a possible pathway to the flagellum, which requires six roughly analogous 'stages.'
3. There are regular and closely spaced fitness plateaus in proceeding from one function to the next.
Anderson here displays a certain confusion regarding what fitness landscapes are. He writes quote:
For Darwin, the fitness landscape was essentially level, with “slight successive variations” leading slowly, almost imperceptibly, to new organisms over a flat plain.
That's just bizarre. Anderson confuses the topography of a fitness landscape with the dynamics of a population evolving on it. Were the fitness landscape a "flat plain," as Anderson claims Darwin assumed, evolution by natural selection would not happen; only neutral drift would occur.
A fitness landscape is a topographical representation induced by an evolutionary operator given a fitness function. For a given fitness function, each evolutionary operator (three such operators in the Avida program are substitution mutation, deletion mutation, and insertion mutation) induces a different fitness landscape. I have elsewhere given examples; I won't bother here unless it becomes a point of contention. The main assumption of evolutionary theory (in this respect) is that evolution mostly proceeds incrementally, where the increments are single steps on one or another fitness landscape. Hence a population is evolving on several landscapes simultaneously. In the case of the Lenski, et al., paper, the three kinds of mutations mentioned above -- three different evolutionary operators -- were enabled. Hence the Avidian populations were evolving simultaneously on three different fitness landscapes, none of them "flat plains."
4. Intermediate steps provide a functional advantage.
That partly true: the theory of evolution hypothesizes that complicated structures are built up out of simpler structures, often through cooption. And it suggests that those 'steps' are themselves either neutral or selectively advantageous. And by golly, that occurs in Avida when the simpler structures are present. Anderson is here complaining that evolutionary theory doesn't posit saltational leaps (tornados in junkyards)! In addition, examination of lineages that evolved to perform EQU shows that they did not follow a monotonically increasing fitness slope; there are declines in fitness during their evolutionary histories. The several "intermediates" were themselves selectively advantageous, of course, but the evolutionary trajectories of evolving Avidian lineages were not monotonic ascents of fitness slopes.
5. Each functional advantage is promptly rewarded.
In both senses in which Anderson defines that claim it is false. First, Anderson writes quote:
Even if there is a functional advantage in a theoretical sense, nature does not operate as an infallible computer program, doling out rewards at each turn. Although a particular change might constitute an objective functional advantage, it is still a significant matter to get the change incorporated into the population. The challenges of reproducing, the numerous vagaries and hazards of nature, all decrease the likelihood of any particular directional change.
In fact, an advantageous mutation is not automatically "promptly rewarded" in Avida. I've watched literally hundreds of Avida runs, dumping dozens of runs to disk for examination. Inspection shows that it is common for an advantageous function to appear in the population in a single offspring, then almost immediately be lost, stay lost a while, then reappear, perhaps to be lost again, then reappear again. That happens because a given Avida critter is always subject to being over-written by a reproducing neighbor -- being killed.
Then Anderson writes quote:
Furthermore, an advantageous change in one sense can be deleterious in another sense. The tremendous interconnectedness of an organism’s biological systems makes incorporating any particular change into the main whole a significant challenge.
Well, the assembly language code of Avidian genotypes is pretty interconnected and vulnerable to disruption, and disruption occurs very frequently. Even in the lineages that evolved to perform EQU there were numerous deleterious mutations. Many more lineages go extinct than survive, and a goodly number are killed off by lethal mutations to their replication code or to code that performs an advantageous function and they're squeezed into extinction by competition for the limited resources in the Avida world.
6. “The benefits increased exponentially with the approximate difficulty of each function.”
Explicating this Anderson writes quote:
In other words, each step closer to the complexity goal (wait a minute, what is that word “goal” doing in an alleged evolutionary mechanism!) is rewarded in a way that makes that step more advantageous vis-à-vis previous steps.
First, "goal" is Anderson's word: It doesn't appear in the Lenski, et al., paper. Second, the Lenski, et al., procedure did not reward steps that are closer to some "goal". It differentially rewarded the performance of logic functions, with more complicated functions earning larger rewards. That is, the several fitness landscapes were not flat with respect to complexity. But that's not a necessary condition. All that is required is that performing each of the several logic functions is independently advantageous. Under those conditions (not reported in Lenski, et al.), Avida critters evolve to perform all of the various logic functions, though it takes longer (in Updates) than if there is differential advantage to performing more complex functions. As I write this, an Avida run in another window on this machine is approaching Update #5,000, and various lineages in the population are performing NOT, NAND, OR_N, OR, and AND_N. Two critters are performing NOR. That's with equal reward for all functions. As I watched now, the two critters that were performing NOR were just killed off. As I said above, that's not infrequent.
In the Section titled "Avida's Results" Anderson wrote quote:
With the foregoing assumptions built into the Avida program, the researchers cheerfully announce that they have shown that a cumulative pathway to complex systems exists.
In fact, they "announce" no such thing. The word "cumulative" appears just twice in the Lenski, et al., paper, both times with reference to the vertical axis of a graph showing "the cumulative number of generations in which an individual differed from its parent by one or more mutations" (p. 140; in paper body and in Figure 2 caption). Anderson's attribution of that announcement to Lenski, et al., is simply false.
Finally, Anderson descends to the 'what good is half an eye' level of argument: quote:
It is worth asking again, what good is a cornea without a lens? What good are a cornea and a lens without the retina? In most complex biological systems it is unclear what functional advantage a few spare parts of the system might endow upon the organism. And it strains reason to suggest that the organism would be rewarded “exponentially” for exercising its forethought in accumulating each part until the final system eventually comes together.
That merely displays a profound misconception of how biological structures are hypothesized to evolve in real evolutionary theory, as distinguished from a caricature that has no basis in the biological literature.
Anderson concludes by asserting that quote:
Such efforts [as Avida] are at best an exercise in irrelevance, at worst, an exercise in self-deception. Similar efforts in the future might be appropriately submitted to an amateur computing journal, but bear little relevance to the origin and development of life on the Earth.
Given the multiple misrepresentations, misconceptions, and false statements in Anderson's essay, one is hard pressed to describe the sheer arrogance of that remark. One might recommend ICR as a suitable venue for his essay.
RBH
[ 08. February 2005, 07:59: Message edited by: RBH ]
IP: Logged
|
|
|
Printer-friendly view of this topic