Yeah, but that really doesn't answer the question as to whether in principle it is possible to model game theoretic situations to produce the optimal outcome for either all participants or a sole participant by a sufficiently complex calculus or even AI, in the future.

Regardless of that, it seems that the whole issue is marred by what counts as 'rational behavior' and if humans can ever be consistent in that behavior in a multitude of situations.

I don't think it's feasible to keep believing in the universality of economic rationality when there are plenty of scenarios which don't contain it. We can play a game that doesn't contain it if you like.

You are now called Toby, Toby has chronic fatigue syndrome. The rules of the game are as follows:
(1) A move is whether you decide to go to work on a given day.
(2) If you become too tired, you will have to spend some time out of work to recover. Becoming too tired is a function of the hours worked within a time period.
(3) If you don't work enough, you will be fired.
(4) You lose when you are fired or when you become very ill from working too much.

This is an incomplete information game in two senses, you don't know the rules fully - only enough to make moves, you don't know the probability distribution of outcomes, you don't know the utility function or expected loss of your moves.

Make a move, and I'll tell you whether its expected value with the hidden utility is positive, negative or 0.

To make it easier, if you give me a sequence of 10 moves, I'll tell you whether it's losing or not. EG: denote by W a day that you work and N a day that you don't, you could give me a string like WWWNWWWN, and I'd tell you whether it's losing or not.