Human AI Construction of Bayesian Networks for Operational Decision Support -- A Virtual Survey Approach

Short summary

This paper proposes using LLM-based AI agents with specific personas to estimate probabilities for Bayesian Belief Networks, bridging expert opinion and data-driven learning. A trimmed-mean rule removes noise from agent responses, and a six-step framework is demonstrated on a healthcare customer-intention model. The approach reveals that subjective norms have stronger causal impact than self-efficacy on consultation intention, challenging initial assumptions.

• •LLM agents with personas estimate probabilities for Bayesian Belief Networks, bridging expert and data-driven approaches
• •Trimmed-mean rule filters noisy agent responses in a six-step BBN construction framework
• •Applied to healthcare: subjective norms outweigh self-efficacy in driving doctor-consultation intention

Generated with AI, which can make mistakes.