> The point is, that it is likely after a while humans will be extremely predictable by the model that machine learning will come up with and select competing models.

Doesn't that inherently create new criteria for fitness to be judged, conditions that humans could then exploit through marginal and expected creative digression?