The thing is, computers can make virtual mistakes with no consequences on reality; and then whichever approach works better in the simulation is what the robots will do for real. For some things, computers can already run simulations faster than real-time; it's only a matter of time before most of the relevant things can be simulated faster than a human brain can.
You are entirely missing my point. My point is a very high IQ one, so don't feel dejected.
For example, your statement assumes that there is a way to measure globally what is "better". My entire point is that there is no such metric, and that life is entropic meaning the possibilities (i.e. orthogonal probabilities and diversity) are always expanding, thus no one can know what is better
for everyone now, future, or in retrospect. Any one can from their arm chair claim that the present would be better had the past been such and such different, but this is incorrect because we can't change just one thing as other things are impacted in unpredictable ways (this is one reason top-down governance is such a failure, even if there were no corruption). Life is more localized variables than can be communicated to a single a metric in real time. If in fact we could transmit all the variables to a omniscient metric in real-time, then the past and present would collapse into a single point in time and the universe would not exist. Such a God would be quite lonely.
Even if you could simulate all the variables, then you would not have time to run a simulation and then go back and rerun it in real life to your omniscient advantage, because the simulation becomes a component of the real life thus impacting things in ways you can not predict. Don't forget Coase's Theorem and competition.
This is way over most of your heads. I think I will need to carefully write a book on this. And I don't have time right now.
If things happen different from the simulation, you adapt the simulation to the new data. You don't need to model the whole universe though; some simplified models can be right often enough that for most practical purposes you don't need to model everything. My point is just that creativity isn't limited to humans, and machines can be better at it; machines can fail better than humans and harvest just the good consequences of failing. Current machines can already beat humans in chess; and we keep making them more and more powerful; if the trend continues for long enough, one day will have machines that will be able to anticipate our moves better than we can anticipate theirs on most things that matter. Humans are more predictable than we like to think we are.
I'm not talking about current economics, i'm talking about the technological singularity (which may be delayed significantly depending on a shitload of factors, including major changes in the economic scenario).