AI as Decision-Maker: Ethics and Risk Preferences of LLMs

TL;DR

This paper investigates how large language models exhibit diverse risk preferences, how alignment tuning affects their risk attitudes, and the implications for safety and economic decision-making.

Contribution

It provides an empirical framework to measure and analyze risk preferences in LLMs and explores the tradeoff between ethical alignment and risk-taking.

Findings

Alignment increases risk aversion by 10%

Caution persists across prompts and impacts forecasts

Tradeoff between safety and economic risk-taking

Abstract

Large Language Models (LLMs) exhibit surprisingly diverse risk preferences when acting as AI decision makers, a crucial characteristic whose origins remain poorly understood despite their expanding economic roles. We analyze 50 LLMs using behavioral tasks, finding stable but diverse risk profiles. Alignment tuning for harmlessness, helpfulness, and honesty significantly increases risk aversion, causally increasing risk aversion confirmed via comparative difference analysis: a ten percent ethics increase cuts risk appetite two to eight percent. This induced caution persists against prompts and affects economic forecasts. Alignment enhances safety but may also suppress valuable risk taking, revealing a tradeoff risking suboptimal economic outcomes. With AI models becoming more powerful and influential in economic decisions while alignment grows increasingly critical, our empirical framework serves as an adaptable and enduring benchmark to track risk preferences and monitor this crucial tension between ethical alignment and economically valuable risk-taking.