No, lending that purpose is not a mistake, it's a tool to aid in understanding. The force of evolution is not conscious, does not think, and is governed by pretty basic rules, but what arises from those simple rules is as close as we can get to intelligent design (not the bogus religious kind) outside of a conscious designer. This similarities are so fundamental, in fact, that the chaotic process of evolution studied and refined into a set of algorithms has actually proven useful in the field of artificial intelligence. Machine learning agents can use Neuroevolution of augmenting topologies to train (essentially learn) because evolution can do the same thing our brains can by reacting to negative and positive stimulus and adjusting a strategy accordingly. The difference of course is that evolution needs a generation to make any adjustments and just tries random bullshit with no clear direction, so our brains are much more efficient and effective.

Note: NEAT works in tandem with normal training procedures and typically replaces the person who would otherwise be attempting to intelligently design the neural network architecture. You can train the weights directly using evolution (I've tried it before), but for the aforementioned reasons, it's slow as hell and doesn't work very well.