Does AI Make People More Open or Reinforce Bias?-AI Recommendation and Confirmation Bias

Abstract

AI algorithms excel at processing information and providing personalized recommendations, yet their inherent opacity obscures information sources, potentially amplifying confirmation bias-the tendency to favor information that confirms existing views. In this paper, we build an analytical model to study how AI affects users with confirmation bias in both information consumption (click, read, and transmit etc.) and subsequent decision-making (in the context of investment decisions). We find that AI could reinforce or mitigate user confirmation bias, depending on the type of decisions and whether signals align with users' preexisting views. Specifically, when signals originate from correlated sources, AI reinforces confirmation bias in information consumption decisions, making users more likely to consume when signals align with their preexisting views and ignore when signals contradict them. The impact of AI on investment decisions is more nuanced: once consumed, users rely more heavily on platform-provided signals than their preexisting views, potentially mitigating or reinforcing confirmation bias depending on the context. We also investigate how platforms' strategic actions, such as information filtration or accuracy enhancement strategies, impact users' decision-making processes under the influences of AI.

Keywords: AI Algorithm, Confirmation Bias, Motivated belief, Information Filtration, Accuracy Enhancement, Platform Strategy