normative
Expert AI Warnings Justify Updating Priors
Even if one assigns only a 1% probability that the analyses of AI experts like Hinton and Bengio are correct, the extremely high stakes combined with risk aversion mean we are rationally required to take the possibility of doom from superintelligence very seriously and update our priors despite our own non-expertise.
normativepending
Speaker
Mikkel BennedsenEvidence Quote
“Even if you judge that there is only, say, 1% probability that the analysis of Hinton and Bengio is correct, it still appears that we are rationally required to take this very seriously.”
Created: 6/13/2026, 12:22:12 AM
Your notes
Loading notes...