According to a new paper on arxiv.org, researchers found that GPT-4.1 reproduces key behavioral patterns from classical economic theory when assigned different socioeconomic personas in a controlled gambling experiment.
The study placed GPT-4.1 in a simulated slot-machine environment with three personas: Rich, Middle-income, and Poor. According to the research, across 50 independent iterations per condition and 6,950 recorded decisions, the Poor persona played a mean of 37.4 rounds per session (SD=15.5) compared to just 1.1 rounds for the Rich persona (SD=0.31), a highly significant difference (Kruskal-Wallis H=393.5, p<2.2e-16).
The paper states that the model reproduced “key behavioral signatures predicted by Kahneman and Tversky’s Prospect Theory without being instructed to do so.” Risk scores showed large effect sizes between personas (Cohen’s d=4.15 for Poor vs Rich).
According to the researchers, emotional labels appeared to function as “post-hoc annotations rather than decision drivers” (chi-square=3205.4, Cramer’s V=0.39), and belief-updating across rounds was negligible (Spearman rho=0.032 for Poor persona, p=0.016).
The findings, the paper states, carry “implications for LLM agent design, interpretability research, and the broader question of whether classical cognitive economic biases are implicitly encoded in large-scale pretrained language models.”