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.</p>\n","updatedAt":"2026-06-04T16:00:20.771Z","author":{"_id":"64c939307dba66c3a7e4d215","avatarUrl":"/avatars/b4c7f43b47db93ca5d7aa30e3d9ef80e.svg","fullname":"BruceLyu","name":"brucelyu","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9157504439353943},"editors":["brucelyu"],"editorAvatarUrls":["/avatars/b4c7f43b47db93ca5d7aa30e3d9ef80e.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.04978","authors":[{"_id":"6a21a08e3490a593e87b10d9","name":"Chensong Huang","hidden":false},{"_id":"6a21a08e3490a593e87b10da","name":"Changyu Chen","hidden":false},{"_id":"6a21a08e3490a593e87b10db","name":"Chenwei Lin","hidden":false},{"_id":"6a21a08e3490a593e87b10dc","name":"Hanjia Lyu","hidden":false},{"_id":"6a21a08e3490a593e87b10dd","name":"Xian Xu","hidden":false},{"_id":"6a21a08e3490a593e87b10de","name":"Jiebo Luo","hidden":false}],"publishedAt":"2026-06-03T00:00:00.000Z","submittedOnDailyAt":"2026-06-04T00:00:00.000Z","title":"Probing Outcome-Level Resemblance and Mechanism-Level Alignment in LLM Risk Decisions: Evidence from the St. Petersburg Game","submittedOnDailyBy":{"_id":"64c939307dba66c3a7e4d215","avatarUrl":"/avatars/b4c7f43b47db93ca5d7aa30e3d9ef80e.svg","isPro":false,"fullname":"BruceLyu","user":"brucelyu","type":"user","name":"brucelyu"},"summary":"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.","upvotes":1,"discussionId":"6a21a08e3490a593e87b10df","ai_summary":"Large language models exhibit surface-level human-like risk decisions in the St. Petersburg game without consistent human-like decision-making mechanisms, highlighting the need for deeper analysis beyond outcome similarity in high-stakes evaluations.","ai_keywords":["large language models","risk decision-making","St. Petersburg game","expected payoff","human decision-making mechanisms","structured prompt suite","controlled decision variants","instruction-tuning","mechanism-level consistency","behavioral alignment"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64c939307dba66c3a7e4d215","avatarUrl":"/avatars/b4c7f43b47db93ca5d7aa30e3d9ef80e.svg","isPro":false,"fullname":"BruceLyu","user":"brucelyu","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.04978.md"}">
Probing Outcome-Level Resemblance and Mechanism-Level Alignment in LLM Risk Decisions: Evidence from the St. Petersburg Game
Abstract
Large language models exhibit surface-level human-like risk decisions in the St. Petersburg game without consistent human-like decision-making mechanisms, highlighting the need for deeper analysis beyond outcome similarity in high-stakes evaluations.
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.
Community
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.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2606.04978 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.04978 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2606.04978 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.