The Failures of Mathematical Anti-Evolutionism
Download 0.99 Mb. Pdf ko'rish
|
The Failures of Mathematical Anti-Evolutionism (Jason Rosenhouse) (z-lib.org)
x
x figure 6.1 Two simple fitness landscapes. (Top) A single hill with one maximum point. Starting from the x, a hill-climbing algorithm will quickly find the max. (Bottom) Starting from the x, a hill-climbing algorithm will only get you to the local max at the top of the left-most hill, but since it cannot go downhill it will never find the global max point at the top of the middle hill. then you can be sure there are other landscapes where it performs poorly. More precisely, the average performance of any algorithm over all possible landscapes is no better than blind search. Conse- quently, researchers must tailor their choice of search algorithm to the problem at hand because there is no all-purpose algorithm to which they can appeal. Expressed differently, if we think of the targets of the search as having a high information content, then we can say the researcher must use prior information about the problem to get access to the information in the target. You need information to get information. We are almost ready to explain why anti-evolutionists believe there is a point of attack in these observations. It will be helpful, though, to return once more to our distinction between track one and track two mathematics. Up to this point I have offered a track one understanding of what the NFL theorems assert. However, if you 198 6 information and combinatorial search read Wolpert and Macready’s paper, you will a find a decidedly track two presentation. The theorems are expressed with copious amounts of jargon and notation, and they will be unreadable to anyone without significant mathematical training. If your intent is merely to under- stand the main ideas underlying the theorems, then it is fine to remain at a track one level. If instead you presume to use the theorems as the basis for an argument against the fundamental soundness of a success- ful scientific theory, then you really must engage at a track two level. With that in mind, let us have a look at what the main NFL theorem really says. In keeping with our previous discussions, our point is to emphasize the precision that goes into expressing a proper mathematical theorem. It will not be necessary to parse every symbol, and you are welcome just to skim the following paragraph. Let us define some notation. We let denote the space to be searched. We let f denote the fitness function, we let Y denote the range of values that f can take on, and we let denote the set of all fitness functions. Let α i and α j be two algorithms that search for an optimal point in by searching one point at a time. We imagine that each algorithm has carried out m steps, and that this has produced an ordered m-tuple of measured values of f. We denote this m-tuple by d Y m . Finally, let P(d Y m | f,m,Y,α) denote the conditional probability of obtaining the sample d Y m , given f, m, Y, and α. Then we have the following result: 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