forecastingresearch.substack.com/p/ai-risk-policy-forecasts
1 correction found
US-only policies are forecast to significantly reduce AI risk, but are still seen as less effective than bilateral policy
The underlying report does not show all US-only policies reducing risk. It says US-only pre-release authorization reduced risk, but a US-only capability slowdown did not reduce risk relative to the unconditional baseline.
Full reasoning
This heading overgeneralizes the results for US-only policies.
In Wave 12, the two US-only policy variants discussed were:
- US-only pre-release authorization, and
- US-only capability slowdown.
The official LEAP report says these did not behave the same way. It reports that:
- US-only pre-release authorization was associated with a reduction in catastrophic risk.
- US-only capability slowdown was not associated with a risk reduction relative to baseline.
The report also gives the expert example: the median expert baseline was 2%, and under a US-only AI capability slowdown it remained 2%. So it is inaccurate to summarize the US-only policies as a group as being forecast to "significantly reduce AI risk."
1 source
- Longitudinal Expert AI Panel — Wave 12: Safety Policies
The bilateral version of both of these policies was associated with a meaningful reduction in catastrophic risk compared to the unconditional baseline. US-only pre-release authorization was also associated with a reduction in risk, though a smaller one than the international version. A US-only slowdown, meanwhile, was not associated with a risk reduction relative to baseline. ... experts gave a baseline 2% chance ... This risk remained at 2% if a US-only AI capability slowdown was implemented.