I Put a Bunch of AI Agents in Charge of a Fantasy Soccer Team. They’re Not Better Than Humans—Yet
Summary
- An experiment deploying AI agents to manage a Premier League fantasy team showed improved performance with specialized multi-agent harnesses, though initial suggestions were flawed.
- Recent academic testing on prediction markets like Kalshi and Polymarket revealed that frontier models can outperform the average person in forecasting tasks.
- Benchmarks such as Prophet Arena and Prediction Arena are being used by researchers to measure and track the forecasting capabilities of AI models.
- While AI prediction skills are rapidly improving, human superforecasters still maintain an advantage in deep real-world understanding, though that edge is narrowing.
Why it matters
As frontier models and multi-agent systems improve at forecasting and processing real-world data, industries reliant on prediction and trading face potential disruption, shifting human roles toward supervision and orchestration.