Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
76 70 82 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
60 42 78
All types combined: 74 (likely 68 to 80, high confidence), from 79 claims: 71 factual claims, 8 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Eliezer Yudkowsky's claims
21 claims
“We (MIRI) tried and failed to find a coherent formula for an agent that would let itself be shut down (without that agent actively trying to get shut down).”
Eliezer Yudkowsky, author 10 June 2022 AGI Ruin: A List of Lethalities (MIRI)
MIRI tried and failed to find a coherent formula for an agent that would let itself be shut down without actively trying to get shut down.
Resolved true The record shows it happened or was so.
The evidence is the person's or their organisation's own records.
Said for an organisation they lead or work for; it counts on both records.
What would count as true or false
Judged as of 2022-06-10. True: MIRI's published work shows that it researched the problem of an agent that would allow itself to be shut down without being incentivized to seek shutdown (corrigibility and the shutdown problem, for example Soares, Fallenstein, Yudkowsky and Armstrong, 'Corrigibility', 2015), and that it reported no satisfactory, coherent solution as of the claim date. Mostly true: MIRI did work on it and reported the problem as unsolved, but with a minor difference in detail, for example the work was partly done by collaborators or framed as utility indifference. Mixed: MIRI worked on related problems, but it is unclear whether it attempted this specific formulation. Mostly false: MIRI did only marginal work on this problem, or published a proposed solution it regarded as adequate. False: MIRI never worked on the shutdown problem, or had published a solution it endorsed as coherent.
Evidence
- Corrigibility - Machine Intelligence Research Institute Machine Intelligence Research Institute (Soares, Fallenstein, Yudkowsky, Armstrong; AAAI Workshops 2015) · January 2015 Own record
“In short, a corrigible solution to the shutdown problem does not yet exist, and there is some question about exactly which behaviors should be incentivized.”
- Corrigibility - Machine Intelligence Research Institute Machine Intelligence Research Institute (Soares, Fallenstein, Yudkowsky, Armstrong; AAAI Workshops 2015) · January 2015 Own record
“While some proposals are interesting, none have yet been demonstrated to satisfy all of our intuitive desiderata, leaving this simple problem in corrigibility wide-open.”
- 2015 in review - Machine Intelligence Research Institute Machine Intelligence Research Institute · 29 July 2016 Own record
“Corrigibility was a mid-level priority for us in 2015, and we spent some effort trying to build better models of corrigible agents. In spite of this, we didn’t achieve any big breakthroughs.”
- Disentangling Corrigibility: 2015-2021 — AI Alignment Forum AI Alignment Forum (post by Koen Holtman) · 16 February 2021
“Agent designs that provably meet more of them have since been developed, for example here.”
Reasoning
MIRI's 2015 'Corrigibility' paper, co-authored by Yudkowsky, addresses the shutdown problem [E1]. It says that "a corrigible solution to the shutdown problem does not yet exist" (E1) and that no proposal met all the desiderata, "leaving this simple problem in corrigibility wide-open" (E2). MIRI's 2015 review says it put effort into models of corrigible agents but "didn't achieve any big breakthroughs" (E3). E4 says designs meeting more of the desiderata were later developed, which does not show that MIRI found or endorsed a coherent solution, so the record meets the criteria for true.
Cite this claim
Eliezer Yudkowsky, “MIRI tried and failed to find a coherent formula for an agent that would let itself be shut down without actively trying to get shut down.” (said 10 June 2022). Verdict: resolved true. Reliability API, claim c_7f208fa7a9e8, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_7f208fa7a9e8
Claims without a verdict
Is a verdict here wrong, or do you speak for Eliezer Yudkowsky? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.