AI Agents Aid Bayesian Network Construction from Expert Opinion.

Building Bayesian Belief Networks (BBNs) for decision-making under uncertainty typically requires either extensive expert judgment or large datasets, both of which can be challenging to acquire. This new methodology proposes using Large Language Models (LLMs) to bridge this gap. It employs a panel of AI agents, each adopting specific personas and contexts, to estimate probabilities for BBNs. The framework then applies a trimmed-mean rule to filter out noise from the agents' responses, enhancing the reliability of the probability estimates. Illustrated with a model of customer intention to consult a doctor in an alternative healthcare system, the approach revealed that subjective norms had a stronger causal impact than self-efficacy. This suggests that improving both confidence and community norms simultaneously is the most effective strategy.