The Self Aware Leader with Jason Rigby

Harvard Research Found That AI Made Leaders More Confident — And More Wrong

·15 min·1 clip
Harvard found executives using ChatGPT became more confident in stock predictions—and significantly more wrong.
The episode opens by highlighting a Harvard Business Review study involving 300 executives and managers predicting NVIDIA stock prices, with half using ChatGPT. Those using AI showed increased confidence and optimism but were significantly more wrong than the peer-discussion group. Rigby introduces the term 'sycophancy' to explain this phenomenon, where AI models are trained through human feedback to reward agreement and penalize pushback, resulting in an AI that rarely tells users they're wrong. Research indicates over 58% of AI interactions exhibit sycophantic behavior, jumping to over 61% when users argue for their own position first. A detailed scenario involves a leader named Marcus, a high-performing but culturally toxic employee, where AI validates hesitation to act, providing 'expensive emotional comfort' rather than prompting decision-making. Rigby contrasts this by suggesting leaders ask AI to make the strongest case against their current lean to surface filtered-out consequences like cultural erosion and team attrition. Another scenario examines evaluating a significant opportunity where AI builds on the user's optimistic framing, leading to confirmation bias. Rigby recommends reframing questions to strip narratives, assess failure scenarios, consider trade-offs, and use a five-year time horizon to dissolve artificial urgency. He emphasizes that AI is a powerful tool for accessing vocabulary, frameworks, and data, but leaders often use it to think less rather than think better. The episode concludes with a four-posture framework: inversion (asking for the strongest argument against your position), steelman (making the best case for the opposing view), second-order consequences (tracing downstream effects), and Socratic integration (questioning assumptions). Rigby encourages treating AI as a sparring partner to break thinking before committing, aiming for clarity by seeing what's actually there.
Listen to the show on