AI Experts Just Scored Safety Ideas — Here's What They Found
Their surprising scores reveal what really works to keep AI safe.
Scott Alexander posted a survey of utility weights for AI safety accomplishments, discussed in Towards the Autograntor. The survey asked people to assign point values to AI safety milestones, anchored so that 1 million points equals a pause with China à la AI Futures Project’s Plan A. Austin and Caroline said they didn’t spend much time on it and had low confidence
- Manifund's survey used a point system where 1 million points equals a major AI pause with China.
- Experts agreed on some safety wins, like effective altruism and small grants, but clashed on open-sourcing AI and export controls.
- Public concern about AI risk hasn't yet led to big policy changes, but a crisis could trigger action.
Why It Matters
Understanding expert disagreements helps you judge AI safety debates and policy decisions that affect everyone.