The Failures of Mathematical Anti-Evolutionism
Download 0.99 Mb. Pdf ko'rish
|
The Failures of Mathematical Anti-Evolutionism (Jason Rosenhouse) (z-lib.org)
Theorem 6.1 For any pair of algorithms α
1 and α 2 , we have that f ∈ P(d Y m | f,m,α 1 ) = f ∈ P(d Y m | f,m,α 2 ) . A serious anti-evolutionary argument based on the NFL theo- rems would need to engage with all of those letters. Now let us have a look at how the anti-evolutionists actually use this theorem. William Dembski introduced the NFL theorems into ID discourse in his 2002 book No Free Lunch. Roughly, his idea is this: The evolutionary process can be seen as a search of genotype space undertaken using the algorithm of passing random genetic 6.8 the no free lunch theorems 199 variations through the sieve of natural selection. Evolutionists claim that this algorithm is effective at searching the space, in the sense that it outperforms blind search by finding functional structures and complex adaptations. The No Free Lunch theorems imply that this can only occur if the algorithm has been tailored in some way to the fitness landscape. If evolution can find high-information targets in genotype space, it is argued, it can only be because information was already built into the search algorithm. Expressed differently, we have seen that when scientists con- front specific search problems, they use information about the fitness landscape to tailor their algorithm to the problem. They know the landscape has a certain general shape, and they choose an algorithm known to be effective on landscapes of that shape. Dembski sees this knowledge as information built into the algorithm by the scientists. Analogically, if we hypothesize that Darwinian processes represent successful search strategies, then in some way information has been built into them from the start. Recall that in Dembski’s formalism, complex biological struc- tures represent rare instances of CSI (“complex, specified informa- tion,” as explained in Section 5.6). If Darwinian mechanisms rou- tinely find such structures in reasonable amounts of time, then they are outperforming blind search. He writes: It follows that any success an evolutionary algorithm has in outputting specified complexity must ultimately be referred to the fitness function that the evolutionary algorithm employs in conducting its search. The No Free Lunch theorems dash any hope of generating specified complexity via evolutionary algorithms. (Dembski 2002, 196) He later elaborates: The No Free Lunch theorems are essentially bookkeeping results. They keep track of how well evolutionary algorithms do at optimizing fitness functions over a phase space. The fundamental 200 6 information and combinatorial search claim of these theorems is that when averaged across fitness functions, evolutionary algorithms cannot outperform blind search. The significance of these theorems is that if an evolutionary algorithm actually proves successful at locating a complex specified target, the algorithm has to exploit a carefully chosen fitness function. This means that any complex specified information in the target had first to reside in the fitness function. (Dembski 2002, 212) At this point, however, we should pause to wonder if the No Free Lunch theorems really apply straightforwardly to evolution. There is certainly a sense in which evolution by natural selection can be said to comprise an algorithm for searching genotype space, but is it a search in the precise, technical sense laid out by the theorems? Can the evolutionary process be formalized as an optimization problem so that all the parameters of the theorem, all those letters we saw in the theorem’s formal statement, are assigned proper values? There is some urgency to this question, since there are at least two obvious ways in which evolution differs from the searches envi- sioned by the NFL theorems. The first is that the theorems assume that the fitness landscape remains unchanged throughout the search process. This is plainly not the case for evolution. What constitutes fitness in one environment might be entirely different from what constitutes fitness in a different environment. It is sometimes said that biological fitness landscapes are made of rubber, and animals alter a landscape’s shape by moving around on it. The second point of difference is that evolution is not a straight- forward optimization process. Philosopher Sahotra Sarkar makes this point very well. After discussing a number of failed attempts to find a parameter that evolution can be said to maximize, he writes, The conclusion from all these negative results is that, in general, evolution by natural selection cannot plausibly be viewed as a mathematical optimization process leading to the maximization of any parameter that can be interpreted as a fitness. 6.8 the no free lunch theorems 201 Consequently, the NFL theorems for optimization are irrelevant to biological evolution. Dembski’s excitement over these theorems reveals little more than a lack of awareness of mathematical evolutionary theory. Download 0.99 Mb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling