EAs optimise black-box objective functions. NFL prevents them from doing well on all possible objective functions. But rather weak assumptions are enough to eva

EAs optimise black-box objective functions. NFL prevents them from doing well on all possible objective functions. But rather weak assumptions are enough to evade NFL, for example the assumption that maximal steepness in the objective space is not achieved (there are many others also). It's particularly easy to make such assumptions when the problems arise in our universe (that's the subset of problems we're interested in), because our universe has structure -- see eg Lin and Tegmark.