Solving safety of superhuman intelligence, in the sense of not killing everyone, could take at least about 30 years (half of the 60+ years AI took to reach today's capabilities).
Unresolvable The public record could not settle it.
Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?
Said for an organisation they lead or work for; it counts on both records.
Reasoning
Hedged estimate of how long solving superhuman AI safety could take; no horizon and untestable.
Cite this claim
Eliezer Yudkowsky, “Solving safety of superhuman intelligence, in the sense of not killing everyone, could take at least about 30 years (half of the 60+ years AI took to reach today's capabilities).” (said 7 April 2023). Verdict: unresolvable. Reliability API, claim c_9dc7b5430830, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_9dc7b5430830
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More prediction claims by Eliezer Yudkowsky
-
Resolved false
“The current crop of people will not try it.”
-
Resolved true
“Alignment will not be solved in a few years.”
-
Resolved true
“even if GPT-5 doesn’t end the world, which I agree is like more than 50% of where my probability mass lies”
GPT-5 will not end the world, with more than 50% of Yudkowsky's probability on that outcome.
-
Mostly true
“Most AI researchers will not want to make unFriendly AIs.”
-
Mixed
“Beating GPT-4 on GI just forces OpenAI to release 4.5.”
Verdicts are made automatically; how verdicts are made.