AI EXISTENTIAL RISK PROBABILITIES ARE TOO UNRELIABLE TO INFORM POLICY

Summary from the TLDR AI newsletter

This essay critiques the use of AI existential risk probability estimates for policy making, arguing that they are too unreliable and lack a solid inductive or deductive basis, often resembling educated guesses rather than evidence-based forecasts. The authors challenge the legitimacy of these forecasts in guiding public policy, especially when linked to costly or freedom-limiting actions, and advocate for an evidence-based approach that acknowledges uncertainty in AI progression. They propose focusing on definable AI milestones for more meaningful policy decisions while cautioning against leveraging speculative existential risk probabilities in high-impact decisions.