You're giving far too much weight to "luck" and far too little to extraordinary time-lengths. Computers struggle to handle a single genome, let alone mix and m

You're giving far too much weight to "luck" and far too little to extraordinary time-lengths. Computers struggle to handle a single genome, let alone mix and match multiple. And even when they do, it's a single organism. In nature, there are thousand's of organisms mixing and matching, thriving and dying all the time, spread over hundreds of millions of years. No computer's bandwidth could come close to that. But I have a deeper discomfort with your question. It assumes that the neural network / tensor-flow model we use with computers approximates a human's. I'm not terribly convinced of that. Sure, human neuron's have are individually approximated by neural network neurons. But concepts such as "instinct" are not captured well by that model. It's weird that a newborn baby has seen far fewer colors, and patterns than a well trained neural network, yet seems to know far more. It signifies we're missing something pretty central to the model.