Artificial intelligence tools like ChatGPT are trained on trillions of points of information. They are also capable of processing statistics and information quickly so, given the NFL's never-ending stats supply, I thought I’d see if AI could use that to predict the winner of Super Bowl LVIII.
As the 2023 NFL season comes to its ultimate showdown on Super Bowl Sunday, the San Francisco 49ers will face off against the Kansas City Chiefs in a repeat of the 2020 Super Bowl.
This clash of the Titans brings with it not only Taylor Swift and her army of fans on the side of the Chiefs, but also a vast amount of data that could have the name of the eventual victor hidden within the numbers.
In a purely for fun and not at all serious experiment I decided to feed as much data as I could find into the three leading chatbots to see what they came up with.
ChatGPT from OpenAI, Claude from Anthropic and Google’s Bard with Gemini Pro were my predictors in this experiment.
Finding appropriate data points
The first task was to work out the most important points of data in predicting an eventual winner, including previous performances. I started by asking each of the AIs which information they think would be most beneficial — remembering they are trained on most of the web including sport reports.
Between them they suggested regular season win records, point differentials, strength of schedule, playoff experience, injuries, rankings, QB performance, weather, coaching records and recent trends.
So with the help of Microsoft Copilot and its access to Bing Search, I compiled those statistics and fed them back into my predictors.
We then added red zone efficiency, third down conversion rates, turnover differential, sacks allowed, field goal percentage, yards per punt and other factors such as pass rush and coverage rates. I also fed in recent news stories on each team.
| Chatbot | Winner | Score | MVP |
|---|---|---|---|
| ChatGPT | Kansas City Chiefs | 30-27 | Patrick Mahomes |
| Bard | Kansas City Chiefs | 28-24 | Nick Bosa |
| Claude | Kansas City Chiefs | 31-27 | Patrick Mahomes |
Bard, ChatGPT and Claude each predicted a victory for the Kansas City Chiefs over the San Francisco 49ers — but that it will be a close game.
It seems the fact the parent company of all three chatbots are based in the San Francisco Bay area didn’t have any real influence on the predictions made for the final result as none went for their home town.
After they made their predictions I asked them to make a call on the final score. They each protested but finally gave in to my pestering.
ChatGPT predicted a score of 30 to 27 in favor of the Chiefs, Bard gave the Chiefs 28 to 24 over the 49ers and Claude predicted a final score of 31 to 27, again to the Chiefs.
Any predictions for the actual game?
Claude explained: “The Kansas City Chiefs will edge out the San Francisco 49ers 31-27 in a tightly contested Super Bowl matchup defined by the stellar play of Patrick Mahomes and the Chiefs' superior defense overcoming the 49ers' strong running game and resilient effort.
Bard suggests both teams will struggle offensively early in the game, ending the first half tied at 10 points a piece. The turning point will be a relentless pass rush from the Chiefs in the third quarter, forcing a crucial interception.
Finally, ChatGPT predicts that pivotal turnovers forced by the Chiefs' defense, critical third-down conversions by Patrick Mahomes under pressure, and decisive special teams plays will set the tone of the game
Claude and ChatGPT gave Mahomes MVP but Bard, ever the contrarian, suggested the Most Valuable Player award might go to 49ers defensive lineman Nick Bosa.
What AI can’t predict is whether Taylor Swift will be at the game or whether we will see any big moments involving her and Travis Kelce. It also wouldn't speculate on the rumored retirement of Kelce.
More from Tom's Guide
- Apple GPT: latest news, rumored release date and predicted features
- Bing with ChatGPT is now Copilot — what it means for you
- Google unveils Gemini AI for Bard chatbot — and it could beat ChatGPT
