(When) Is Truth-telling Favored in AI Debate?
This work addresses the challenge of ensuring truthful AI outputs in safety-critical applications, though it is incremental by building on prior debate proposals.
The paper tackles the problem of evaluating AI-proposed solutions when human judges are unreliable by introducing a mathematical framework for AI debates, focusing on feature debates to analyze truth-tracking and incentives like confusion or stalling.
For some problems, humans may not be able to accurately judge the goodness of AI-proposed solutions. Irving et al. (2018) propose that in such cases, we may use a debate between two AI systems to amplify the problem-solving capabilities of a human judge. We introduce a mathematical framework that can model debates of this type and propose that the quality of debate designs should be measured by the accuracy of the most persuasive answer. We describe a simple instance of the debate framework called feature debate and analyze the degree to which such debates track the truth. We argue that despite being very simple, feature debates nonetheless capture many aspects of practical debates such as the incentives to confuse the judge or stall to prevent losing. We then outline how these models should be generalized to analyze a wider range of debate phenomena.