The Volunteer’s Dilemma (http://en.wikipedia.org/wiki/Volunteer’s_dilemma) is in essence the Prisoner’s Dilemma with more players—which means that defection is an even more dominant strategy. The problem is that the decision whether to do unpleasant tasks becomes a Volunteer’s Dilemma with multiple future selves as my competition
Ah, interesting; I was recently trying to think about procrastination in terms of defecting against your future self in an iterated Prisoner’s Dilemma, but this makes more sense.
Of course the standard causal reasoning would say to defect (actually, if you have a sufficiently low estimate of the probability that anyone else will cooperate, I guess it would say to cooperate), but TDT-like reasoning should allow acausal coordination to get an average of one cooperator (or however many are needed). At a Bay Area LW meetup last year, I was in a discussion (well, mostly just listening to a discussion) about the Luring Lottery. Someone, I think it was Nick Tarleton, concluded that in a Luring Lottery of n TDT agents each trying to maximize its own winnings, assuming each one knows how many other players there are and that the others are also TDT agents, each would pseudorandomly choose with probability 1/n to submit one entry, or to submit zero otherwise. (I may be misremembering the details, but it was along those lines.) A similar strategy would probably apply in a Volunteer’s Dilemma among TDT agents.
So perhaps we could reduce procrastination by taking TDT sufficiently seriously! If there’s one situation that merits the assumption that multiple agents are strongly correlated (and/or can very reliably predict each other), it’s where all the agents are slightly delayed versions of yourself. (That was the original context in which I was thinking about this; real-world results pending.)
Ah, interesting; I was recently trying to think about procrastination in terms of defecting against your future self in an iterated Prisoner’s Dilemma, but this makes more sense.
Of course the standard causal reasoning would say to defect (actually, if you have a sufficiently low estimate of the probability that anyone else will cooperate, I guess it would say to cooperate), but TDT-like reasoning should allow acausal coordination to get an average of one cooperator (or however many are needed). At a Bay Area LW meetup last year, I was in a discussion (well, mostly just listening to a discussion) about the Luring Lottery. Someone, I think it was Nick Tarleton, concluded that in a Luring Lottery of n TDT agents each trying to maximize its own winnings, assuming each one knows how many other players there are and that the others are also TDT agents, each would pseudorandomly choose with probability 1/n to submit one entry, or to submit zero otherwise. (I may be misremembering the details, but it was along those lines.) A similar strategy would probably apply in a Volunteer’s Dilemma among TDT agents.
So perhaps we could reduce procrastination by taking TDT sufficiently seriously! If there’s one situation that merits the assumption that multiple agents are strongly correlated (and/or can very reliably predict each other), it’s where all the agents are slightly delayed versions of yourself. (That was the original context in which I was thinking about this; real-world results pending.)