For policymakers to use forecasting in their work, they need accurate forecasts, but—perhaps equally important—the forecasts need to be about decision-relevant questions. However, knowing which questions will be the most valuable to forecast on can be difficult. How can policymakers identify the short-term events that are most relevant to important long-term outcomes? 

In “Conditional Trees: A Method for Generating Informative Questions about Complex Topics,” we describe a method for finding high-value forecasting questions that would be informative for predictions about AI risk. Using specialized interviews with domain experts and highly-skilled forecasters, we generated 75 new AI forecasting questions with resolution dates ranging from 2030-2070, and tested how informative they are: that is, how much would knowing the answer to a given forecasting question about the next few decades inform a participant’s forecast of the risk of AI causing human extinction by 2100? 

We found that questions generated using this method were more informative than the top AI questions on popular forecasting platforms. We think this result is useful for people trying to use forecasting for policy and other planning purposes. Higher value of information (VOI) questions are likely more useful as cruxes for future decisions, so these results suggest that investing resources in finding high VOI questions may result in questions that are more useful than those generated by existing platforms.