The consequences of AI training on human decision-making

Authors: LS Treiman, CJ Ho, W Kool

Published: 2024

Publication: Proceedings of the National Academy of ..., 2024 - pnas.org

People alter their behavior when they know their actions will train AI, leading to unintentional habits and biased training data for AI systems.

Methods: Five studies were conducted using the ultimatum game; participants were tasked with deciding on monetary splits proposed by either humans or AI, with some informed their decisions would train the AI.

Key Findings: Behavioral changes in participants when training AI, persistence of these changes over time, and implications for AI training bias.

Limitations: The specific sample size and demographic diversity of the participants were not disclosed, which may limit the generalizability of the findings.

Institution: Massachusetts Institute of Technology, Yale University, Washington University in St. Louis

Research Area: AI Ethics, Behavioral Economics,Decision-Making in AI Systems

Discipline: Artificial Intelligence , Behavioral Science

Citations: 13

DOI: https://doi.org/10.1073/pnas.2408731121