Jonathan_Graehl comments on The Absent-Minded Driver - Less Wrong

27 Post author: Wei_Dai 16 September 2009 12:51AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (139)

You are viewing a single comment's thread.

Comment author: Jonathan_Graehl 16 September 2009 02:15:35AM *  0 points [-]

From the paper:

At START, he is given the option to be woken either at both intersections, or only at X. In the first option he is absent-minded: when waking up, he does not know at which intersection he is. We call the second option ‘‘clear-headedness.’’ As in the previous discussion, the question at X is not operative (what to do) but only whether it makes sense to be ‘‘sorry.’’ If he chose clear-headedness, his expectation upon reaching X is 1/4. If he had chosen absent-mindedness, then when reaching X he would have attributed probability 2/3 to being at X and 1/3 to being at Y.

I disagree. If I opt to be woken at X, then when I'm woken at intersection X, I know I'm at X. Even if the idea is that I'll immediately go back to sleep, forget everything, and wake again at my payoff/destination, I still knew when I was woken at an intersection that it was X.

Unless I chose to be a) woken at X and b) have no memory of having chosen that, in which case it's fair to say that "I'm at some intersection" (but they suppose I remember the road map I was given indicating intersectons/payoffs X,Y). I guess that's what must have been meant.

Comment author: gwern 03 January 2010 03:44:49PM 0 points [-]

I think it must've.

If you remember your choice, then there's no reason to ever choose 'wake me up only at X'. If you wake up only at X, then you will either take action to turn off and score 0 points, or you will go back to sleep and continue to C and score 1 point.

0 or 1 is pretty bad, since you could then do better just by randomly picking from A, B, or C (as (1/3)*1 + (1/3)*4 + (1/3)*0 = 1.6...). Eliminating one branch entirely eliminates the problem.