Presentation
Source Bias in AI Decision-Making: A Simulation of Large Language Models’ Use of Human vs AI-Labeled Advice
SessionPoster Session 2
DescriptionAs AI systems become trusted agents in financial decision-making, understanding their behavior is essential. This study explores whether ChatGPT, acting as a simulated investor, makes different decisions based on whether advice is labeled as coming from a human or another AI. In our controlled simulation, identical investment recommendations were presented with varying source labels—either “Human Expert” or “AI System.” We then tracked how often ChatGPT followed each type of advice and analyzed the resulting portfolio performance. Our goal was to detect any source-based bias and determine whether such bias affects investment outcomes. Does ChatGPT favor one source over the other? And if so, does that lead to suboptimal decisions? Findings from this study offer important implications for how we design and trust AI systems that are expected to reason objectively. The results also raise broader questions about how AI models interpret social cues like authority, expertise, or self-reference embedded in natural language.
Event Type
Poster
TimeWednesday, October 15th5:30pm - 6:30pm CDT
LocationRiverside East
Similar Presentations