Proposal for quantitative AI risk modeling
researchers propose shifting ai safety from qualitative evaluations to quantitative probabilistic modeling, citing the nuclear industry's transition after 1975.
drawing parallels to the 1975 w Åsh-1400 report that revolutionized nuclear safety, this proposal argues that current ai risk management relies too heavily on qualitative capability evaluations. the authors suggest that quantitative modeling forces explicit documentation of assumptions, enabling critique and iterative improvement while allowing regulators to set specific probability benchmarks. to demonstrate the viability of this approach, the team constructed nine probabilistic models specifically examining ai-enabled cyber attacks, hoping to invite criticism and further development of quantitative ai safety standards.