I've thought about the prisoner's dilemma often. I once saw a documentary where Richard Dawkins talked about the programs people made in the 1980's to explore strategies. The gist of the experiment is that they assigned values to the permutations of defect/cooperate in one on one trials and the programs each had their own algorithms for deciding what to do with the other program with access to memory of previous trials.The conclusion of that study was that "tit for tat" tends to work best and that "nice" programs outperformed nasty ones.

The problem is that the simulation assumes that you know both who is the actor and whether they defected or cooperated in realtime. The real world is much messier and psychopaths thrive in that ambiguity with deceptive tactics. If you don't know that I'm screwing you how could you possibly resist me? That's fundamentally what fiat is about.

The lack of accounting for this factor seems like the massive simplification that leads to incomplete/false conclusions. It would be interesting if you could tweak those computer simulations so that you inject random noise that obfuscates both "who" is acting and "what" they did especially in the case of a defect. Imagine simulating a "long con" where you don't even know you got screwed 100 trials ago but now you found out you're down big.

It would be interesting to see how transparency/move fidelity would affect which program wins. I guess I have a hackathon project now lol.