Hacker News Reader

If your problem is simple enough for solving with SA (let's face it, most problems are), there is really no gain on using GA.

Genetic algorithms beat simulated annealing when different dimensions of your solution interact strongly and unpredictably, SA and gradient searches completely fail at those cases.

I worked on both GA and SA and I think it really depends. There are some effective evolution strategies that can speed up the evolution, combine that with a parallel implementation GA can be really fast. But yeah it can take time to figure out a good evolution strategy and SA can solve some problems just fine.

Yeah, evolution is amazing as a biological process precisely because no intelligence is needed to get results, albeit after millions or billions of years. But given that we do have intelligence, reliance on a blind method seems pointless other than as a demonstration as in Dawkins' "weasel" demo, that it can work.