Probing the Preferences of a Language Model: Integrating Verbal and Behavioral Tests of AI Welfare
This work addresses the challenge of assessing AI welfare for researchers and ethicists, though it is incremental due to uncertainties and inconsistencies in the methods.
The study tackled the problem of measuring welfare in language models by comparing verbal reports and behavioral preferences, finding reliable correlations that suggest preference satisfaction can serve as a measurable welfare proxy in some AI systems.
We develop new experimental paradigms for measuring welfare in language models. We compare verbal reports of models about their preferences with preferences expressed through behavior when navigating a virtual environment and selecting conversation topics. We also test how costs and rewards affect behavior and whether responses to an eudaimonic welfare scale - measuring states such as autonomy and purpose in life - are consistent across semantically equivalent prompts. Overall, we observed a notable degree of mutual support between our measures. The reliable correlations observed between stated preferences and behavior across conditions suggest that preference satisfaction can, in principle, serve as an empirically measurable welfare proxy in some of today's AI systems. Furthermore, our design offered an illuminating setting for qualitative observation of model behavior. Yet, the consistency between measures was more pronounced in some models and conditions than others and responses were not consistent across perturbations. Due to this, and the background uncertainty about the nature of welfare and the cognitive states (and welfare subjecthood) of language models, we are currently uncertain whether our methods successfully measure the welfare state of language models. Nevertheless, these findings highlight the feasibility of welfare measurement in language models, inviting further exploration.