Quote from: sporket on April 30, 2014, 06:11:39 PM

Quote from: p0peji on April 30, 2014, 05:55:58 PM

...
1. Prisoners dilemma
2. The problem with this statement is that you never know what is the best interest of the whole.

Yup. Not even sure about the intended meaning of "best interests of the whole."

Prisoners' dilemma is not realistic because that logic only works in one-time games. Society is a repeated game, and someone who defects most of the time is going to lose. If the majority defects most of the time, collusion will appear in which the cooperating minority cooperates inside the group and becomes better. Tit-for-tat (cooperate first) is a surprisingly excellent strategy. Very few non-colluding strategies can do better. The only collusion strategy that does way better is master-slaves. Trust me, 99% of the cases people invoke prisoners' dilemma they use it wrong. If anything, it proves how "unenlightened self-interest" (economic rationality) is dumb and suboptimal compared to "enlightened self-interest" (economic super-rationality).

It is similar to the implications of Godel's theorems. A Turing machine cannot transcend its condition, but mathematicians are not Turing machine. Guess what, neither machine learning is Turing-limited because it is computation in the limit. Don't be the Turing machine in society.

Regarding 2, yes, that is a problem. I do not believe it is a fatal problem, because there exist Pareto superior options almost always. Just like (artifical example) VCG auctions make everybody be truthful (barring collusion).