AI summaryⓘ
The authors study how large language models (LLMs) make risky decisions using the St. Petersburg game, where humans usually choose modest bets despite theoretically infinite expected winnings. While many LLMs give similar modest bids, the authors found that this surface similarity hides different underlying decision processes compared to humans. When the game rules were changed, LLMs shifted towards more mathematically rational choices rather than behaving like humans. Attempts to prompt models to think from a human perspective or apply instruction tuning improved some behaviors but did not fully align the decision mechanisms. The authors suggest evaluating LLM decision-making should look beyond just similar answers and focus on whether models reason in human-like ways.
Large Language ModelsSt. Petersburg gameRisk decision-makingExpected payoffBehavioral alignmentInstruction tuningHuman decision mechanismsOutcome similarityComputational rationalityPrompting
Authors
Chensong Huang, Changyu Chen, Chenwei Lin, Hanjia Lyu, Xian Xu, Jiebo Luo
Abstract
LLMs can appear cautious in risk decision-making tasks, yet cautious-looking outputs do not necessarily indicate alignment with human decision-making mechanisms. We investigate this distinction using the St. Petersburg game as a controlled testbed, a classical paradox in which the expected payoff is infinite, yet humans typically report low, finite willingness to pay. We evaluate 28 LLMs with a structured prompt suite that includes the original game; controlled decision variants that perturb truncation, repeated play, numeric endowment, and occupational identity; a human-perspective prompt that asks models to reason as human decision makers; and paired comparisons between base models and their instruction-tuned counterparts. In the original game, most models generate finite bids, creating the appearance of human-like risk behavior. However, this outcome-level resemblance masks substantial mechanism-level differences. The controlled variants reveal that rather than maintaining human-like behavior seen in the original game, models often shift to conditionally and computationally rational behavior. Human-cue prompting and instruction tuning often lower bids and reduce some visible pathologies, but most mechanism-level response patterns remain largely unchanged. These findings show that behavioral alignment in risk decision-making can be surface-level: LLMs may produce human-like risk decisions without exhibiting human-consistent mechanisms. High-stakes evaluations of LLM decision-making should therefore move beyond outcome similarity and examine whether the alignment is supported by mechanism-level consistency.