Essentially the argument is problematic due to lack of omniscience. The typical solution to this is to adopt a stable yet flexible enough strategy (also known as a robust solution). It is almost never too simple a strategy if you start from a complex optimization problem like this.

(Compare expectation maximization to robust decision making. Or maximin to minimax to simple optimization approaches. None win.)

Currently there is no algorithm for solving general decision problems on a global scale. Even genetic algorithms (constrained innovation and permutation) can and do produce solutions too late to matter, especially when selective pressure is low. Human intelligence easily gets caught in local maxima too, not to mention its fallibility. (I'd call it lossy heuristic approach with memory. Heuristics are transmitted over generations in a strongly lossy way both in number and count. New ones take the place of the old ones with only relatively low bar of improvement to pass. Actor attempts to reconstruct the heuristic by observing behaviour of other actors given its capabilities.)