My comment on that was here. To quote it in its entirity:
Um: yuck. For one thing, why is the agent assumed to be summing infinite sequences in its utility function in the first place?
My reaction these days is much the same. We can forget about most things after the universal heat death. So: this is a kind of fantasy mathematics. Best to ignore it—and focus instead on things that have some chance of being useful.
You’re objecting to his 2009 paper, which is about a sum over infinite time. This 2007 paper is timeless.
My response to his 2009 paper is here. At present I think that it a) assumes what it is trying to prove, and b) assumes there is no time discounting, and so could be seen as an argument for time discounting. But I haven’t looked at it closely enough to be confident of (a), given how accurate Peter typically is.
The future is big enough that I don’t think that cutting off the infinite tail helps. Look at the gigantic quantities of utility involved in hypothetical but practically grounded futures involving AI-assisted mankind reaching merely the rest of our galaxy, and according to which the most important thing for everyone to do right now is either work on FAI or donate to the SIAI. Is that a reasonable calculation, or a Pascal’s Mugging?
As a rule of thumb, whenever something blows up infinitely at an infinite limit, one should consider that it may blow up practically at a practically reachable point.
I take unreasonableness to be part of what it means, to judge something to be a Pascal’s Mugging, but if you want to argue that it’s both reasonable and a Pascal’s Mugging, go right ahead.
Look at the gigantic quantities of utility involved in hypothetical but practically grounded futures involving AI-assisted mankind reaching merely the rest of our galaxy, and according to which the most important thing for everyone to do right now is either work on FAI or donate to the SIAI. Is that a reasonable calculation, or a Pascal’s Mugging?
Personally, I rate that as not being a reasonable calculation—though I do think that machine intelligence is an important field, which people should consider working on, if they have an aptitude for it.
The future is big enough that I don’t think that cutting off the infinite tail helps.
It sure helps with the particular problem of adding up an infinite number of finite things giving an infinite value. That’s what the problem under discussion here boils down to.
My comment on that was here. To quote it in its entirity:
My reaction these days is much the same. We can forget about most things after the universal heat death. So: this is a kind of fantasy mathematics. Best to ignore it—and focus instead on things that have some chance of being useful.
You’re objecting to his 2009 paper, which is about a sum over infinite time. This 2007 paper is timeless.
My response to his 2009 paper is here. At present I think that it a) assumes what it is trying to prove, and b) assumes there is no time discounting, and so could be seen as an argument for time discounting. But I haven’t looked at it closely enough to be confident of (a), given how accurate Peter typically is.
The future is big enough that I don’t think that cutting off the infinite tail helps. Look at the gigantic quantities of utility involved in hypothetical but practically grounded futures involving AI-assisted mankind reaching merely the rest of our galaxy, and according to which the most important thing for everyone to do right now is either work on FAI or donate to the SIAI. Is that a reasonable calculation, or a Pascal’s Mugging?
As a rule of thumb, whenever something blows up infinitely at an infinite limit, one should consider that it may blow up practically at a practically reachable point.
Is that an or, or an exclusive or?
I take unreasonableness to be part of what it means, to judge something to be a Pascal’s Mugging, but if you want to argue that it’s both reasonable and a Pascal’s Mugging, go right ahead.
Personally, I rate that as not being a reasonable calculation—though I do think that machine intelligence is an important field, which people should consider working on, if they have an aptitude for it.
It sure helps with the particular problem of adding up an infinite number of finite things giving an infinite value. That’s what the problem under discussion here boils down to.