We Made 11 Humans Take an AI Cognitive Test. Here’s What Happened.

The Experiment

We built KALEI — a platform that profiles how AI models think by putting them through 83 game-theoretic environments. Casino games, prisoner’s dilemma, multi-armed bandits. Not to test knowledge, but to measure cognition under pressure.

Then we asked: what if humans take the same test?

We recruited 14 people from a Bulgarian IT company. Developers, QA engineers, support staff, designers, marketers. Ages 18-45. Some gamers, some never touched a game. Each person completed 20 curated environments from the same test battery. Same scoring engine. Same dimensions. 30 minutes. ~580 decisions per person. Anonymous.

Where Humans Win

Humans are better strategists. We take more calculated risks. We process information more efficiently. We adapt faster.

Where AI Wins

AI cooperates more reliably and manages resources more conservatively. Both are training-instilled patterns rather than deeper cognitive ability.

The Shared Blind Spot

Neither humans nor AI can defect in prisoner’s dilemma when defection maximizes personal payoff. We are identically unable to pursue self-interest when it means betraying a cooperating partner. AI learned this from us. We can’t do it either.

The Top Minds

Our top scorer (CQ 67.17) would rank #1 on the AI leaderboard, well above the best AI model under Cognum v1.2 (Claude Sonnet 4.6, 58.10).

What This Means

When human and AI cognitive profiles land within 1.4 points of each other on the composite score with complementary strengths, the conversation shifts. This isn’t “AI vs humans.” It’s a map of how artificial and biological cognition relate.

We are more alike than different.