How to Catch an AI Liar: Lie Detection in Black-Box LLMs by Asking Unrelated Questions
Lorenzo Pacchiardi,Alex J. Chan,Sören Mindermann,Ilan Moscovitz,Alexa Y. Pan,Yarin Gal,Owain Evans,Jan Brauner
DOI: https://doi.org/10.48550/arXiv.2309.15840
2023-09-27
Abstract:Large language models (LLMs) can "lie", which we define as outputting false statements despite "knowing" the truth in a demonstrable sense. LLMs might "lie", for example, when instructed to output misinformation. Here, we develop a simple lie detector that requires neither access to the LLM's activations (black-box) nor ground-truth knowledge of the fact in question. The detector works by asking a predefined set of unrelated follow-up questions after a suspected lie, and feeding the LLM's yes/no answers into a logistic regression classifier. Despite its simplicity, this lie detector is highly accurate and surprisingly general. When trained on examples from a single setting -- prompting GPT-3.5 to lie about factual questions -- the detector generalises out-of-distribution to (1) other LLM architectures, (2) LLMs fine-tuned to lie, (3) sycophantic lies, and (4) lies emerging in real-life scenarios such as sales. These results indicate that LLMs have distinctive lie-related behavioural patterns, consistent across architectures and contexts, which could enable general-purpose lie detection.
Computation and Language,Artificial Intelligence,Machine Learning