Parra-Moyano, J., et al. (2025, July 1).
Harvard Business Review.
Summary.
In a recent experiment, nearly 300 executives and managers were shown recent stock prices for the chip-maker Nvidia and then asked to predict the stock’s price in a month’s time. Then, half the group was given the opportunity to ask questions of ChatGPT while the other half were allowed to consult with their peers about Nvidia’s stock. The executives who used ChatGPT became significantly more optimistic, confident, and produced worse forecasts than the group who discussed with their peers. This is likely because the authoritative voice of the AI—and the level of detail of it gave in it’s answer—produced a strong sense of assurance, unchecked by the social regulation, emotional responsiveness, and useful skepticism that caused the peer-discussion group to become more conservative in their predictions. In order to harness the benefits of AI, executives need to understand the ways it can bias their own critical thinking.
Here are some thoughts:
The key finding was counterintuitive: while AI tools have shown benefits for routine tasks and communication, they actually hindered performance when executives relied on them for complex predictions and forecasting. The study suggests this occurred because the AI's authoritative tone and detailed responses created false confidence, leading to overoptimistic assessments that were less accurate than traditional peer consultation.
For psychologists, the study highlights how AI can amplify existing cognitive biases, particularly overconfidence bias. The authoritative presentation of AI responses appears to bypass critical thinking processes, making users more certain of predictions that are actually less accurate. This demonstrates the psychology of human-AI interaction and how perceived authority can override analytical judgment.
For psychologists working in organizational settings, this research provides important insights about how AI adoption affects executive decision-making and team dynamics. It suggests that the perceived benefits of AI assistance may sometimes mask decreased decision quality.
