Long before anyone had heard of ChatGPT and Gemini, it was common enough to see sports articles and previews highlighting computer picks – often tagged as “supercomputer” picks – for various sports events, including the NFL season. Now, though, with the ubiquity of so many AI bots, there is always a flurry of those “AI predicts X” articles before the beginning of any competition or tournament.
Most fans know to take AI predictions with a pinch of salt, but there is also a conveyance of omnipotence as the bot will have crunched way more data than any human possibly could. We also know that individual games can throw up surprises, with underdogs capable of beating the NFL game spreads any given week. But laying out predictions for a season can be, in some senses, easier than picking individual game winners, as one expects the cream to rise to the top and luck to even out. But AI’s performance in predicting the NFL season thus far? Distinctly average, even poor.
The Chiefs prediction was badly off
A case in point is the prediction that the Kansas City Chiefs would do well, with some AI predictions settling on over 12 wins. On Sunday, the Chiefs were eliminated from contention for the Playoffs. They will be lucky to finish the season anywhere close to a winning record, never mind being one of the NFL’s dominant teams. Other predictions that were off included the Baltimore Ravens (predicted 11-6 but are stuck on .500 for now) and the Broncos (9-8; they could reach 15 wins).
We should say for balance that AI was on the money with some predictions, including one we have seen that outlined a poor season for the Washington Commanders. The Commanders were the surprise packages of the season last time around, but despite making the NFC Championship Game in January, AI correctly predicted the slide back to obscurity.
AI sports predictions are based on human ones
Still, by and large, AI predictions were just as prescient as human ones, which is to say off the mark in many cases. It is arguably interesting, too, to consider how much of an overlap there was between AI projections and human ones. Indeed, most AI predictions did not deviate too much from the sportsbook betting odds for the season. When you think about it, this is not unsurprising, given that human prediction articles and betting odds will form a major part of the data fed to the AI before it makes its predictions.
There should, however, be a lesson here, especially for those people who are tempted to take up AI betting subscriptions. There are limits to what AI can do about sports predictions beyond examining human-created data. We are not yet at the point where AI can watch the games and start formulating data from what it sees, at least not with consumer-facing chatbots. Of course, you could feed it massive amounts of data from proprietary specialists’ football platforms, but in most consumer-facing cases, the AI is going to scrape analysis from some sports analysis websites and deliver a consensus.
In the end, AI has performed pretty much to the same standard as human-led predictions for the NFL season. It’s been a fascinating season, with underdogs rising to the fore and preseason favorites struggling. It feels like it was impossible to predict, and that’s why both humans and AI have struggled. Perhaps that’s just what happens with sports, but it feels like we are a long way off from considering AI as some sort of oracle when it comes to telling us what is going to happen.
