Improving judgments of existential risk: Better forecasts, questions, explanations, policies

Abstract

Forecasting tournaments are misaligned with the goal of producing actionable forecasts of existential risks, an extreme-stakes domain with slow accuracy feedback and elusive proxies for long-run outcomes. We show how to improve alignment by measuring facets of human judgment that play central roles in policy debates but have long been dismissed as unmeasurable. The key is supplementing traditional objective accuracy metrics with reciprocal intersubjective metrics, such as skill at predicting other forecasters’ judgments, on topics that resist objective scoring, such as long-run scenarios, probative questions, insightfulness of explanations, and impactfulness of risk-mitigation options. Forecasting tournaments could then support a broad research program on refining probabilistic judgment. In practical terms, we focus on running four types of tournaments which improve predictions of short-term events, conditional trees of events portending long-run existential risks, persuasive rationales for probabilistic judgments, and comparative impacts of risk-mitigation policies. After introducing the four types of tournaments, we discuss essential steps for scaling up the endeavor, such as securing talent with appropriate expertise and viewpoints, motivating diligent participation, and addressing potential information hazards that make the knowledge gained in the tournaments valuable to bad actors. We discuss criteria for selecting or designing proper scoring rules that normalize and bound individual forecasts and close with a discussion of how tournaments can help in the challenging but crucial task of translating research findings on existential risk into policy actions. – AI-generated abstract.