Calibrating Conservatism for Scalable Oversight
This work addresses the problem of maintaining human oversight over autonomous AI systems that may exceed human capabilities, offering a practical method with statistical guarantees.
The paper introduces Calibrated Collective Oversight (CCO), a method for scalable oversight of AI agents that aggregates diverse scoring functions to penalize deviations from conservative behavior. In experiments on SWE-bench and MACHIAVELLI, CCO reduces ethical violations while preserving reward, with empirical violation rates matching specified targets.
Agentic AI systems capable of autonomous planning and extended environmental interaction pose a fundamental control problem: how can humans maintain meaningful oversight of systems that may exceed their own capabilities? Existing approaches to scalable oversight rely on complex assumptions, remain largely heuristic, or lack practical methods for sequential settings with statistical guarantees. We introduce Calibrated Collective Oversight (CCO), which aggregates diverse auxiliary scoring functions into a penalty measuring deviation from a conservative baseline. Inspired by Attainable Utility Preservation, CCO enables collective conservatism: actions face a penalty proportional to overseer concern, so high-utility actions are still selected when overseers find them unobjectionable and overridden only when concern accumulates. CCO calibrates this conservatism online using Conformal Decision Theory, ensuring that undesirable outcomes remain below a user-specified target threshold with finite-time bounds and no distributional assumptions. On a modified version of SWE-bench, weaker overseers successfully constrain an adversarially misaligned stronger agent; on MACHIAVELLI, CCO substantially reduces ethical violations while preserving reward. In both settings, empirical violation rates closely match the specified targets, as predicted by the theory.