The Funder’s Meta-Problem
Sequential decision-making under uncertainty for Effective Altruism funders
2023-01-01
A simulation study of planning policies for an Effective Altruism funder maximizing welfare through intervention selection. Accumulated welfare differs substantially between naive policies — e.g., relying on a single study to form beliefs about effectiveness — and more advanced probabilistic policies that optimize re-evaluation timing. The gap is more pronounced in sequential decision-making and persists even against a hypothetical optimal policy, indicating considerable room for methodological improvement in how funders update beliefs.