Predict-AI-bility of how humans balance self-interest with the interest of others

• DOI: 10.48550/arxiv.2307.12776
• Publication Date: 2023-01-01
• Authors: Valerio Capraro; Roberto Di Paolo; Veronica Pizziol
• Keywords: FOS: Computer and information sciences; General Economics (econ.GN); Computer Science - Artificial Intelligence; Social and Personality Psychology; Engineering Psychology; Social and Behavioral Sciences; FOS: Economics and business; Computer Science - Computers and Society; Artificial Intelligence (cs.AI); Computer Science - Computer Science and Game Theory; Computers and Society (cs.CY); Quantitative Methods; Psychology, other; Economics - General Economics; Computer Science and Game Theory (cs.GT)

Generative artificial intelligence (AI) holds enormous potential to revolutionize decision-making processes, from everyday high-stake scenarios. By leveraging generative AI, humans can benefit data-driven insights and predictions, enhancing their ability make informed decisions that consider a wide array of factors outcomes. However, as many carry social implications, for AI be reliable assistant it is crucial able capture the balance between self-interest interest others. We investigate three most advanced chatbots predict dictator game across 108 experiments with human participants 12 countries. find only GPT-4 (not Bard nor Bing) correctly captures qualitative behavioral patterns, identifying major classes behavior: self-interested, inequity-averse, fully altruistic. Nonetheless, consistently underestimates inequity-aversion, while overestimating altruistic behavior. This bias has significant implications developers users, overly optimistic expectations about altruism may lead disappointment, frustration, suboptimal in public policy or business contexts, even conflict.