Evolutionary Algorithms: Are We There Yet? — December 17th, 2010 by Ann Gauger
In the recent past, several papers have been published that claim to demonstrate that biological evolution can readily produce new genetic information, using as their evidence the ability of various evolutionary algorithms to find a specific target. This is a rather large claim.
It has thus fallen to others in the scientific or engineering community to evaluate these published claims. How well do these algorithms model biology? How exactly was the work done? Do the results make sense? Are there unexamined variables that might affect the interpretation of results? Are there hidden sources of bias? Are the conclusions justified or do they go beyond the scope of what has been shown?
A new paper by Montañez et al. [1], just published in the journal BIO-Complexity, answers some of these questions for the evolutionary algorithm ev [2], one of the computer programs proposed to simulate biological evolution. As perhaps should be no surprise, the authors found that ev uses sources of active information (meaning information added to the search to improve its chances of success compared to a blind search) to help it find its target. Indeed, the algorithm is predisposed toward success because information about the search is built into its very structure.
These same authors have previously reported on the hidden sources of information that allowed another evolutionary algorithm, AVIDA [3-5], to find its target. Once again, active information introduced by the structure of the algorithm was what allowed it to be successful.
These results confirm that there is no free lunch for evolutionary algorithms. Active information is needed to guide any search that does better than a random walk.
[2] doi:10.1093/nar/28.14.2794