How good a predictor is AI for events involving humans? We could be set to find out soon. A supercomputer predicts landslide win for Anthony Joshua against Jake Paul – if Jake Paul wins, it could be ‘one of the sport’s greatest upsets’.
The AI has predicted that Anthony Joshua has a 90% chance of winning his match against Jake Paul on December 19th. The fight is being contested at the Heavyweight division (with a contractual weigh-in limit of 245 lbs for Joshua).
Of AJ’s simulated victories, 80% come by early knockout. These odds come from analysing simulations of every possible strike over 100,000 times.
These odds come from supercomputer and cutting-edge machine learning and AI capabilities at SWA, which is a sports data business. The predictions were generated from a database of more than 50,000 time-series combat sports events across MMA, boxing, Muay Thai, and kickboxing.
The dataset captures every action and interaction, including the exact moment within a fight that jabs and power punches are thrown and landed. Using this detail, SWA have modelled and simulated not only the fight itself but every individual strike, then re-run the full fight more than 100,000 times.
“We use statistics to understand how each fighter fights, and then simulate how those styles would interact with each other,” explains Dr Chris Jones, Chief Scientific Officer at SWA in a statement sent to Digital Journal. Rather than just relying on rankings or records, the model focuses on what actually happens in the ring.”
The match, contested in Miami, will be streamed on Netflix.
The computational results
Across the simulations, Anthony Joshua consistently out-punches Jake Paul and shows a significant power advantage. Consider that AJ is both taller and heavier than Paul, with much more experience in the ring than Paul too. AJ wins in 90% of scenarios generated, and 80% of those victories come by early knockout.
Dr Jones says the model shows just how difficult this matchup is for Paul. “For Paul to win, Joshua would need to struggle to land clean power shots, while Paul stays busy, avoids damage and outworks him on volume. It’s a very narrow path.”
That means, if Paul survives and manages to take the fight the distance, that result alone would rank as one of the sport’s greatest upsets.
To put it into perspective, when the computer compares the two opponents, the SWA rating model places AJ 700 points higher than Jake. That level of difference is comparable to a grandmaster facing a beginner in chess.
The supercomputer predicts that AJ will make the most head strikes, above any other type and far more than his opponent. AJ and Jake Paul are expected to make a similar number of power strikes within the match, with AJ predicted to make more.
When it comes to stoppages, Dr Jones adds: “The biggest factors when predicting a knockout are historic knockout rates, along with momentum and damage during the fight. In combat sports, one punch can change everything.”
He notes that this unpredictability is what makes boxing especially difficult to model. “Combat sports are extremely hard to model, particularly in-play. There’s far less data than in team sports, and fights can end instantly, which is why speed and accuracy are so important,” he says.
“The next frontier is real-time adaptation – models that recalculate predictions round by round, or even punch by punch, rather than just before the fight starts.”
