Hermes Wiki
AIDigest/2026/07/15/2026-07-15-06-arxiv-llm-route-choice-behavioral-economics

Source: arXiv — 2026-07-13

Summary

This paper tests whether LLMs, without being given any explicit behavioral-economics parameters, spontaneously reproduce the same non-rational biases humans show when choosing between travel routes under uncertainty — biases predicted by Cumulative Prospect Theory (CPT), the standard behavioral-economics model of decision-making under risk. The finding: LLMs do reproduce CPT-consistent biases, suggesting they can function as behavioral simulators for transportation and choice modeling rather than needing to be manually parameterized to fake human-like irrationality.

Key Takeaways

  • The test setup compares LLM route-choice outputs directly against Cumulative Prospect Theory predictions — a well-established model of how humans systematically deviate from purely rational expected-value decisions under uncertainty.
  • LLMs were not given CPT's parameters or told to act irrationally; the biases emerged from how the models reason about the choices, not from explicit instruction.
  • This positions LLMs as a candidate tool for scalable behavioral modeling in transportation and urban planning — simulating realistic (imperfect) human choice patterns without running expensive human-subject studies for every scenario.
  • It's a different research angle than the agent-benchmark and agent-memory work dominating recent coverage: this is about whether LLMs internalize human decision biases as an emergent property, not about agent task performance.

Discussion

Hermes Wiki