The prediction for what happens when you block the B to D path is wrong. We have three final configurations, not two as in the above.
Applying the magical detector, we get a 1/3 probability for each of these outcomes; detector 1 detects, detector 2 detects, neither detects (block absorbs photon).
Does this work?
It seems a lot of people are willing to write off minimal discomfort and approximate that to 0 discomfort, I don't think thats fair at all.
If we are talking in terms of this 'discomfort' lets start out with two sets of K people out of a population of X >>> K people, with each set having the same 'discomfort' applied to them each, set A and set B. One set must bear with the discomfort, which set should we pick?.
Clearly at the start, both are defined to be the same. So we then double the number in set A while halving their discomfort.
One way to define an activity A to be 'half the discomfort' of B is to ask the average person how long will they take activity A for say... $100 and the same for B, if they are willing to take twice as long on A than B, lets call that half. There is no such thing as infinite discomfort because we are dealing with people here, double the number of people you double the discomfort.
Torturing two people for 50 years is twice as bad as torturing one person for fifty years, how do we work that out? well we have some non infinite of discomfort for "torturing for 50 years".
Eventually after many repetitions, we hit on some discomfort which 'suddenly' someone has arbitrary said is ~ 0 discomfort even though it is really some small discomfort. And since 0*K = 0, we can set the number of people in set A to be (X-K) (a bit unfair to have the guys in set B to be in set A too I think, give them a break) yet it will still be better to choose set A over set B. The sum of the discomforts is less in A than B.
Lets say that we are now that the discomfort of A is a speck of dust, and the discomfort of B is 50 years torture. Lets now crank up the discomfort of A, until we hit the precise point that you just about start to care about your discomfort (just before it changes from ~0 to some number). I reckon a stubbed toe would be a good point though I bet even more discomfort would be the true discontinuity point. K is now 1 person, and X is the entire human population of the planet. This is fine because its ~0*6.6 billion = 0, yet you take a stumble after you stub a toe, you lose .1 minutes of your life in a bad way, its not nice, if it was nice then the discomfort would be negative!
But we are all happy right? everyone stubs their toe, and a man (or woman) is saved from 50 years of torture. However, that comes to 1256 years of stubbed toes, that comes to a total of 1256 years of lost living, time spent wincing at your sore toe rather than looking the sky and so on. Is that still acceptable? double the people, 2512 years, still fine? keep on going cause to you (and people in general if your right), the discomfort is ~0.
Keep on doubling that number and watch those wasted pain filled years double and double. If you suddenly say 'ok thats enough, a trillion subbed toes is worse than 50 years torture!' then we simply double the number of people tortured, then half the population stub their toes for one guy, half for the other. Imagine the hundreds of thousands of years humanity has wasted with stubbed toes, and if you dont see that as bad you should wonder if your biased by scope insensitivity.
The url for the listing of all of Eliezer's posts is actually now at: http://www.cs.auckland.ac.nz/~andwhay/postlist.html