LLM Decision-Making

This field examines how large language models make choices, use tools, ask for information, and act alone or with other agents, including in settings such as finance, healthcare, law, and strategic games. Researchers draw on behavioral economics and cognitive science, decision theory, reinforcement learning, and multi-agent methods, using controlled choice experiments and benchmarks that test constraints, calibration, fairness, and long-horizon performance. A central challenge is that correct answers or stated reasoning do not guarantee sound actions: choices can shift with framing, option order, context, and how evidence or costs are presented. Current work explores more reliable decision layers, explicit uncertainty and deferral, evidence-grounded rules, and tests that distinguish good outcomes from stable, human-aligned decision processes.