Political Intelligence for AI Safety: The AI Risk Attitudes Survey (AIRAS)

Cool project idea! Systematically mapping political viability of specific AI safety red lines would indeed be very valuable. The design principles (especially comprehension gating and tradeoff testing) and the white paper do look like a serious attempt that is ready to be looked at and red-teamed by experts in survey design. Also congrts on the dashboard; it does a great job illustrating what actionable output could look like.

A great next step would be to do quick validation of the survey with a few real human respondents. Even a very small pilot (a few dozens of respondents) would dramatically strengthen this. Another good step would be to strengthen the analysis of related literature on risk perception and public opinion methodology. I'd be particularly interested in seeing a deeper investigation of the difference betwween stated preferences and actual behavior or even political mobilization.

I hope you pursue this project further and that it will eventually yield some very useful data!

This project provides good infrastructure for testing public willingness to bear real costs for security from AI systems. Generic polling showing support for "AI regulation" helps policymakers understand the direction of public preferences but not their strength — you can consistently observe publics that support more spending in every individual department while simultaneously wishing to cut overall government spending, or who favour more immigration from every specific group presented to them while wanting lower immigration overall - presumably because frequently when asked the public considers whether something is first order good/bad without implications for the general budget constraint. The same dynamic could plausibly apply to AI governance: abstract support for "careful management" may not survive contact with specific costs, so verifying this is important to get a sense of the depth of support for safety measures.

The survey methodology is well-designed to help find this, with comprehension gates, relative priority and a large number of possible costs the public could be concerned about, meaning it should be able to provide a reasonable estimate of the gap between raw support and "hard support" that survives cost framing is a vital indicator for long-run political durability. The political viability scorecard is well formatted and gives a clear map for future policy design. I would be very interested to see how the public's level of commitment to various specific interventions in this framework compares to data not accounting for this.