Embedded Files
Selected Publications:
Selected Publications:
Risky-choice Framing Sensitivity in LLMs and Impact on Human Decisions" (w/ Haritima Chauhan ). 2026. Computers in Human Behavior Reports.
Abstract
We study whether frontier large language models (LLMs) exhibit prospect-theoretic framing patterns in risky choice and how their risk posture compares with – and shapes – that of humans
in business-relevant contexts. In two preregistered between-subjects discrete-choice experiments, a panel of 17 LLMs and 1,029 human participants evaluated four numerically equivalent gain–loss framing scenarios, chose between a certain option and a probabilistic one, and provided a brief justification. Most frontier reasoning and chat models favored certainty in 87–100% of trials across frames and contexts, substantially attenuating the classic gain–loss reversal, while lighter-weight models exhibited greater human-like framing sensitivity and, in some cases, extreme risk-seeking. Human participants reproduced the classic framing effect with modest variation across scenarios. When shown a “committee of advisors” recommendation aggregated from the first experiment, however, human choices shifted sharply toward certainty in both frames – by 8.3 to 15.5 percentage points in the gain frame and 25.1 to 37.3 in the loss frame, controlling for risk preferences and other factors. Topic modeling of justifications shows that unadvised humans drew on moral and responsibility-based reasoning essentially absent from LLM output; under AI advice, that vocabulary receded and certainty- and probability-based language expanded – humans began to sound like the LLMs. Deploying an LLM as a decision aid is thus not a neutral upgrade: it imports a model-specific distribution of risk recommendations and delegates not only the choice but the framing of the decision itself, with direct implications for AI governance.
Keywords:
Generative AI, Large language models, Framing effects, Risky decision-making, AI governance, Prospect theory
Bravado and Beguile: When the Beautiful Bet Boldly (attractiveness and risk preferences). Journal for Risk and Uncertainty. Vol. 72 (4).
Abstract
This paper explores the relationship between facial attractiveness and risk-taking preferences using three standard elicitation methods, each modeled on its own scale. Drawing on incentivized samples of students and professionals recruited directly from LinkedIn (plumbers, lawyers, and MBA-holding business professionals) along with a standard, continuous attractiveness measure, I find that beauty is associated with greater risk tolerance across generalized, financial, and social risk domains, most prominently among lawyers and MBAs. A validated proxy for confidence is consistent with partial mediation in the generalized and financial domains, with strength varying by occupation, while a direct association between attractiveness and risk tolerance persists throughout. Beauty is particularly correlated with cooperation in the socially risky Stag Hunt setting, a link that operates largely outside the confidence channel. These associations have implications for occupational selection, fair talent allocation, digital recruiting, and other managerial settings.
Keywords Beauty · Risk preferences · Cooperation · Attractiveness · Confidence · Occupational selection
Language AI Analysis of Trust Game Perceptions: Endogenous Topic Discovery and Behavioral Prediction" (w/ Haritima Chauhan and J. Braxton Gately). 2026. Forthcoming, Discover Artificial Intelligence.
The Trust Game (and its many variants) is a widely used experimental paradigm for studying trust and prosocial behavior, yet its interpretation remains debated. We demonstrate endogenous topic discovery and labeling via language AI as a scalable complement to traditional methods, uncovering semantic themes not commonly associated with the game. Using 1,124 subjects' descriptions of Trust Game interactions from two studies, we show that social concerns emerge as key predictors of prosocial behavior, alongside trust-related language, even when the interaction is framed as a loan repayment. The pipeline converts unstructured responses into predictions of individual decisions: we train a cross-validated XGBoost–Random Forest ensemble that classifies prosocial and selfish behavior from topic features alone well above chance (AUC 0.73). Explicit language categorized as risk, gambling, investment, or game-theoretic themes was less informative in models predicting passing than language categorized as trust, teamwork, or involving moral self-image. These findings both validate and challenge conventional assumptions, highlighting the importance of social and moral language in decision-making.
Sibling Kinship, Norms, Language AI, and Prosociality (w/ Haritima Chauhan and Manda Tiwari) 2026. Judgment and Decision Making https://doi.org/10.1017/jdm.2026.10032
This manuscript examines how growing up with a sibling relates to prosociality and how knowledge of a partner’s sibling background may serve as a behavioral cue. In a series of experimental games, we found that individuals with siblings were significantly more likely to cooperate in stag hunt and contribute more in public goods and dictator games than only children on average. In two treatments where a sibling status cue is exogenously revealed, only-child pairs exhibited reduced prosociality. Only children exhibit different empirical expectations of behavior compared to those with siblings, while generally sharing the same normative beliefs. Text analysis of subjects’ perspectives on the games corroborates these patterns. We conclude that only children exhibit more context-dependent prosociality, with behavior more closely aligned with empirical expectations than normative beliefs, a pattern not observed in those with siblings.
Keywords: cooperation, kinship, prosociality, siblings, norms, altruism, language AI, cues
Chatbot or Humanaut? How the Source of Advice Impacts Prosocial Behavior (w/ Haritima Chauhan). 2026, Journal of Behavioral and Experimental Economics.
This paper explores how the source of advice -- human or generative AI (genAI) -- relates to behavior in three classic bargaining games commonly used to assess prosociality and cooperative welfare gains. Utilizing a novel experiment, we show that the source of advice matters. While both sources of advice increased prosociality, players preferred human advice over that from genAI and were more willing to pay for it. Prosocial behavior was more prevalent when players received human advice -- advice increased the probability of adopting the Pareto-optimal strategy by 14% in the stag hunt and boosted contributions of 19% to the public goods game and 8% in dictator. Leveraging language AI advances, we demonstrate that the advice corpora differ significantly. Humans were more objective, specific, intuitive, and norm-oriented; genAI offered guided reasoning and targeted concepts of risk and strategy. Entities adopting genAI technologies should balance AI agency with human oversight, mindful of behavioral salience and moral credibility.Revisiting Erat and Gneezy’s White Lies Paradigm (w/ Haritima Chauhan). 2024, Journal of Economic Psychology.
When Pretty Hurts: Beauty Premia and Penalties in eSports Contracts (w/ Haritima Chauhan and Steven Kistler). 2024, Journal of Economic Behavior and Organization.
Gender Penalties and Solidarity -- Teaching Evaluation Differentials in and out of STEM (w/ Andrew Hussey). 2023, Economics Letters.
Show No Quarter: Combating Plausible Lies with Ex Ante Honesty Oaths (w/ Haritima Chauhan) 2023, Journal of Economic Science Association.
Initiating Free-flow Communication in Trust Games. (w/ Haritima Chauhan) 2023, Frontiers in Behavioral Economics.
You Can’t Hide Your Lying Eyes: Honesty Oaths and Misrepresentation (w/ Fenndy Liu and Haritima Chauhan). 2022, Journal of Behavioral and Experimental Economics, Vol. 98. https://doi.org/10.1016/j.socec.2022.101880. Cited in List, J. A. (2026), Experimental Economics: Theory and Practice (University of Chicago Press), Chapter 18 (IRB and instruments for data collection).
Abstract: Lying about race or personal characteristics for a job or in college admissions is common and has recently become a high-profile issue. In this paper, we explore the decision to misrepresent oneself and determine how honesty oaths impact personal characteristic reporting. To do this, we execute an experiment on Amazon MTurk, using a self-reporting task involving human eye color. We find that honesty oaths elicit more truthful behavior – primarily reducing implausible lies (maximal outcome lies). As a result, we spent 27.6% less on bonuses than we would have without oath-taking. There is some evidence that if one believes lying is common, they are more likely to lie as well. We conclude that oaths decrease extreme misrepresentation and expectations of group behavior significantly impact the decision to deceive.- Beauty Premiums Among Academics (w/ Andrew Hussey, Alex Nikolsko-Rzhevskyy, and David Taylor, 2020). Economics of Education Review, Vol. 78(3).
This paper examines the effects of instructors’ attractiveness on student evaluations of their teaching. We build on previous studies by holding both observed and unobserved characteristics of the instructor and classes constant. Our identification strategy exploits the fact that many instructors, in addition to traditional teaching in the classroom, also teach in the online environment, where attractiveness is either unknown or less salient. We utilize multiple attractiveness measures, including facial symmetry software, subjective evaluations, and a novel, proxy methodology that resembles a “Keynesian Beauty Contest.” We identify a substantial beauty premium in face-to-face classes for women but not for men. While gender on its own does not impact teaching evaluation scores, female instructors rated as more attractive receive higher instructional ratings. This result holds across several beauty measures, given a multitude of controls and while controlling for unobserved instructor characteristics and skills. Notably, the positive relationship between beauty and teaching effectiveness is not found in the online environment, suggesting the observed premium may be due to discrimination.
Stay at Home Orders, Loneliness, and Collaborative Behavior (w/ Marine Foray and Andrew Hussey). 2021, Economics and Human Biology, Vol. 43(1).
Linguistic Signaling, Emojis, and Skin Tone in Trust Games PLoS ONE (2020), 15(6): e0233277.
This paper reports the results of an experiment involving text-messaging and emojis in laboratory trust games executed on mobile devices. Decomposing chat logs, I find that trust increases dramatically with the introduction of emojis to one-shot games, while reciprocation increases only modestly. Skin tones embedded in emojis impact sharing and resulting gains – to the benefit of some and detriment to others. Both light and dark skin players trust less on receipt of a dark skin tone emoji – suggestive of statistical discrimination. In this way, computer-mediated communication leads to reduced gains for dark-skinned persons. These results highlight the complex social judgment that motivates trust in an anonymous counterpart.
Google Sites
Report abuse