This academic paper by Jazmia Henry from the University of Oxford investigates the concept of fairness in Artificial Intelligence (AI) through the lens of dictator games, a type of utilitarian game. The study focuses on understanding how AI models play these games, mimicking human interaction and decision-making processes. Over 101 rounds of the dictator game, the study concludes that AI exhibits a strong sense of fairness, which is dependent on its perception of the other player’s trustworthiness. Moreover, the framing of the game significantly influences how much the AI is willing to give a recipient when it is the trustee. The paper suggests that AI, like humans, may experience inequality aversion, affecting its decision-making process.
Publication date: 9 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.05786
