Every indexed document

they’re grounded in a voluminous source of evidence—namely

although the future-life preferences are

but it doesn’t seem to prevent us from designing

and let F be the problem of maximizing the expected sum of

corresponding to how well the machine has

we’ll be able to create provably beneficial AI.

such that no matter how perfectly a system solves F

so that we’re sure to be happy with how they behave? On the

then “smarter than gorillas” is also meaningless

but it is utterly meaningless for current AI

“so ‘smarter than humans’ is a meaningless

or indeed many other games still to be invented. Put another

the goal of checkmate; the same chessboard and pieces can be used

given its experience of the world. This is

“Contrary to the more fantastic predictions for AI in the popular press

however. We are already devoting huge scientific and technical

there is no race of superhuman robots on the horizon or probably even possible.”

it might arrive considerably sooner than expected. One variation

if we were to detect a large asteroid predicted to collide with

we’ll run out of gas before we get there!” The claim