If each TDT agent cares about its number of descendants, then they will instrumentally care about having a greater proportion of TDT agents to mutually cooperate with, and will therefore defect when one of the three players is a CDT agent, like the clique bot.
They only behave as you describe if they consider only the immediate, but not long term, consequences.
If each TDT agent cares about its number of descendants, then they will instrumentally care about having a greater proportion of TDT agents to mutually cooperate with, and will therefore defect when one of the three players is a CDT agent, like the clique bot.
They only behave as you describe if they consider only the immediate, but not long term, consequences.
Seems correct to me. Judging agents with a different utility function than they use to decide will lead to them looking stupid.
I mentioned this to JGWeissman privately at the time, but I want to confirm now that this was exactly the solution I intended.