Conversation

Kelsey Piper
@KelseyTuoc
I think a lot of people look into the case that AI will definitely kill us, find it's not airtight, and end up believing something like"AI is 10% likely to kill us" but not treating that like it's itself an enormously big deal that would if true need to be society's top priority.
48
511
London is pedestrian
@londoncrockett
Replying to
@KelseyTuoc
I think most people looking into AI threat find it's not 10%, but 0.00000?% & many other things are more likely to be bad, so they focus on that. You discover your kid might die from X, but the prob is low and ?, then you find out your kid is on meth, which do you focus on?
2
27
Kelsey Piper
@KelseyTuoc
Replying to
@londoncrockett
right if you think it's 0.00000?% then obviously do something else with your life, I'm talking about the half of surveyed ML researchers who say they think it's 10% and the large share of people I know who say similar things.
6
52
Kelsey Piper
@KelseyTuoc
Replying to
@CriticalAI
and
@londoncrockett
The survey was sent to 4271 authors who recently published in NeurIPS and ICML, of which 738 agreed to participate and sent in responses. (Most of the questions were not about extreme risk scenarios but about other aspects of AI's impact on society.)
3
1