AI
Prospect Theory in Large Language Models: A Meta-Analysis of Loss Aversion, Diminishing Sensitivity, and Probability Weighting
Publication: cs.AI updates on arXiv.org
Author: Adi Tarabeih
Source published: 27/10/2026, 04:00:00 UTC
Briefing created: 22/11/2026, 16:46:03 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 28/10/2026, 02:33:24 UTC.
Briefings
A meta-analysis of 102 fitted estimates from 11 studies on 32 LLMs found that loss aversion is modest (pooled lambda 1.26) and diminishing sensitivity depends on measurement method, while probability weighting shows human-like inverse S shapes. Findings are limited to 11 studies and 32 models, with uncertainty around specific model families and fitting methods.
Key points
- The pooled lambda value for loss aversion in LLMs is 1.26 with a 95% confidence interval of 1.06 to 1.46, indicating losses weigh less than gains.
- Probability weighting displays a human-like inverse S shape with gamma values of 0.81 (Tversky-Kahneman) and 0.84 (Tanaka method) on fewer than 10 studies.
Context
This research analyzes how Large Language Models (LLMs) compare to humans in decision-making under risk, specifically testing loss aversion, diminishing sensitivity, and probability weighting. The study aggregates 102 fitted estimates from 11 independent studies to determine if LLMs exhibit the same psychological biases as humans.
Limitations
Evidence is limited to 11 studies and 32 language models, restricting generalizability. Specific findings on loss aversion and diminishing sensitivity vary by measurement method, and probability weighting results are sparse across studies.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 22/11/2026, 16:46:03 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeienpa7rzb52dnfawuvxkpre56p3afxvoga5quzngyz7sid46m2kbm