A pilot study using Polymarket data reveals that the value of human-AI collaboration depends on specific human traits rather than model benchmarks alone. Hybrid performance was found to be trimodal, with most users either deferring to or rubber-stamping the AI, while a minority achieved complementary reasoning.
- Most participants matched the model's accuracy or performed worse by using it to validate prior guesses.
- A minority engaged in genuine complementary reasoning, reaching accuracy that matched or exceeded the market itself.
- Collaborative traits such as perspective-taking, intellectual humility, and curiosity distinguished those who succeeded.
- Raw cognitive ability did not predict success; the results are statistically robust but preliminary.
The findings motivate a pre-registered replication to further validate how specific human capital factors enable effective hybrid intelligence.