[pyrois]

Sure, they are commensurate at some point. I'd pick one nanosecond of torture over the dust-motes, for instance. But I'm not certain that I would ever choose the 50 years of torture over the dust motes for any number of dust-moted people. That's because it's not continuous, so you can't do the sort of epsilon-delta proofs that Yudkowsky's argument depends on.

At least, in my opinion.

[ikeboy]

That means that for some amount of time, say X seconds, you would prefer X-1 seconds of torture for one person over specks, but prefer specks over X seconds of torture for one person.

But let's double each side; presumably you would make the same decision if asked again, right? So now you prefer X-1 seconds of torture done to each of two people, over twice as much specks.

Now, unless X is very low, you should prefer X for a single person over X-1 for two people. (If you disagree with this, please give a plausible value for X that makes it false.)

So you prefer X on a single person over double!specks, but prefer single!specks over X. This seems extremely unlikely. Or even if true, we should be able to make you pick torture for 50 years just by multiplying specks another couple of orders of magnitude.

Does this make his argument any clearer?