[makoConstruct]

This is also something Yudkowsky has something interesting to say about

https://arbital.com/p/logical_dt/

Summary: LDT is a decision theory that is able to recognise that the decisions of all agents who implement LDT will be correlated. When it makes a decision it knows that it's making the decision for its entire decision-theoretic equivalence class. If it ever decided to knowingly do a thing that is "individually rational" but collectively disastrous, in cases where the personal cost of the disaster outweighs the personal benefits of defecting, then it would be malfunctioning.

I put quotes around "individually rational" because, in light of LDT, CDT is not really rational. It performs worse. It doesn't win.

And, personally, I don't think the old model of decisionmaking was closer to what humans intuitively consider to be sane, effective decisionmaking, I think we had to be taught to be ruthless parasites who can't keep promises, and I hope we can unlearn that now.