Humans vs. AI: A Live Tennis Prediction Experiment
2026-04-19
Can Human Intuition Beat the Model?
Every tennis fan has opinions about who's going to win. Sometimes those instincts are right. You notice things models may miss — a player who looks flat in warmups, a rivalry dynamic that brings out someone's best tennis, home-crowd energy, or a recent slump that the numbers haven't fully captured yet.
But over time, are those instincts actually better than a prediction model built on thousands of matches, surface data, player form, rankings, and market signals?
We decided to find out.
How the Experiment Works
We're running a public experiment called Humans vs. AI.
- Our model generates predictions for selected upcoming tennis matches
- Before each match starts, anyone can submit a prediction for the winner
- Once the match ends, we compare: were you right? Was the model?
- We track everything on the same set of matches — your record, the crowd consensus, and the model's record
This is not a hypothetical backtest or a cherry-picked highlight reel. It's a live, transparent comparison running in public.
How to Participate
On the web Sign in to the dashboard and submit your prediction for any featured match. You can update it any time before the match starts.
On Telegram Join our channel and vote in the daily match polls. If your Telegram account is linked to your web account, your predictions stay unified across both.
What We Track
On the experiment page, you can follow:
- Model accuracy — how often the model predicts the correct winner
- Crowd consensus accuracy — how often the majority vote gets it right
- Average participant accuracy — how all users perform in aggregate
- Your personal record — your own wins, losses, and accuracy over time
We're especially interested in:
- Does the crowd outperform the model over a large sample?
- Are there match types where human judgment has a consistent edge — upsets, rivalries, surface switches?
- When humans disagree with the model, who tends to be right more often?
- Does the crowd add signal, or just noise?
Why This Matters
Most prediction models operate as black boxes. You're told to trust the output without any way to verify whether it's actually better than your own judgment.
We think that's the wrong approach.
If you consistently outperform the model, you should know that. If the model consistently outperforms you, that's useful information too. And if the crowd consensus beats the model in specific situations, that tells us something important about where the model can improve.
Public accountability is more valuable than black-box confidence.
The Honest Truth
The model will not get every match right. Neither will you.
Tennis is volatile. Injuries, momentum swings, bad days, and retirements are part of the sport. The question is not who is perfect. It is who is more right, more often, over time.
Over a small sample, anything can happen. Over hundreds of matches, patterns emerge. That is what makes this worth running in public — and why we are committed to publishing the results regardless of how they land.
The model does not care who wins. We are curious whether you can.
Join the Experiment
It is free, it takes a few seconds per match, and it gives you a real way to test your instincts against a model trained on 8,600+ tennis matches.
Humans vs. AI. Let's find out.