As artificial intelligence (AI) capabilities advance, machine agents are increasingly integrated into financial markets, making it crucial to understand their influence on market dynamics. This study investigates how the presence and disclosure of machine agents influence return extrapolation, a key behavioral bias in financial decision-making, and subsequently explores which human–machine interaction (HCI) design paradigms may mitigate this bias.
Human–machine hybrid Prediction Markets are used to extract and integrate beliefs from both human and machine agents. This work, through two empirical studies, aims to address two research questions: (1) To what extent does the introduction of machine agents and their presence disclosure affect the return extrapolation of the entire trader population, including both human and machine traders? (2) Which design paradigms of HCI may further mitigate return extrapolation?
First, the introduction of machine agents significantly reduces return extrapolation. Second, disclosing the presence of machine agents weakens their effectiveness in reducing return extrapolation. Third, implementing a competitive goal structure between human and machine agents further reduces return extrapolation. Fourth, this competitive structure proves most effective when machine agents share profits based on human performance.
The findings offer valuable insights into the role of AI in financial markets and provide guidance for the design and governance of financial trading and prediction platforms.
