Experience Stern | Faculty & Research

Overview: From choosing acquisitions to prioritizing R&D projects, high-stakes judgments are often seen as uniquely human. New research, “The Strategic Foresight of LLMs: Evidence from a Fully Prospective Venture Tournament,” from strategic foresight expert and NYU Stern Professor Aticus Peterson, along with co-authors Felipe Csaszar (University of Michigan Ross) and Daniel Wilde (Indiana University Bloomington Kelley), challenges that assumption. Using a real-time prediction experiment, the team finds that frontier AI models significantly outperform experienced humans at forecasting which new ventures will succeed. The results suggest that AI decision-making may already surpass human judgment in core strategic tasks.

Why Study This Now: Business leaders are beginning to ask whether AI can support or even replace elements of strategic decision-making. This study is timely because it provides rare evidence of how AI performs under genuine uncertainty.

What the Researchers Found: The researchers ran a live prediction tournament using Kickstarter ventures launched after every AI model’s training cutoff. A unique aspect of this study is that predictions were made before outcomes were known, eliminating the possibility of data leakage. Key findings include:

• The top AI model correctly predicted which of two ventures would outperform the other nearly four out of five times, compared to about three out of five for the best human.
• Performance gaps compound quickly when screening many opportunities, giving AI a large advantage in early-stage evaluation.
• Neither wisdom-of-the-crowd averaging nor human/AI hybrid teams outperformed the best standalone AI model.
• Statistically combining human and AI predictions did not improve outcomes when AI accuracy already exceeded human baselines.

What Does This Change: The findings challenge the prevailing belief that strategic foresight and executive judgment are inherently human domains. They also raise concerns about an “augmentation trap,” where adding human oversight to AI decisions may reduce performance rather than improve it. For investors, corporate development teams, and innovation committees, the results suggest managers should evaluate whether AI can improve strategic forecasting and where human input adds value.

Key Insight: “AI is not just automating routine tasks,” explains Peterson. “In some strategic contexts, it already outperforms human judgment, forcing managers to rethink when human intuition adds value and when it may hold decision-making back.”