Difficulties with Evaluating a Deception Detector for AIs

TL;DR

Developing reliable AI deception detectors is challenging due to the lack of confidently labeled examples and the complexity of evaluating their effectiveness, requiring new approaches and further research.

Contribution

The paper identifies key obstacles in evaluating AI deception detectors, analyzes existing methods, and discusses potential but insufficient workarounds, highlighting the need for further research.

Findings

Current lack of confidently labeled deceptive and honest examples

Existing empirical methods face significant obstacles

Proposed workarounds are valuable but insufficient alone

Abstract

Building reliable deception detectors for AI systems -- methods that could predict when an AI system is being strategically deceptive without necessarily requiring behavioural evidence -- would be valuable in mitigating risks from advanced AI systems. But evaluating the reliability and efficacy of a proposed deception detector requires examples that we can confidently label as either deceptive or honest. We argue that we currently lack the necessary examples and further identify several concrete obstacles in collecting them. We provide evidence from conceptual arguments, analysis of existing empirical works, and analysis of novel illustrative case studies. We also discuss the potential of several proposed empirical workarounds to these problems and argue that while they seem valuable, they also seem insufficient alone. Progress on deception detection likely requires further consideration of these problems.