The Limitations in Using AI for Sports Predictions

 

Sports are unpredictable by their very nature. Sure, there are good teams, and bad teams, and some great teams and terrible teams. Yet, there is always the opportunity for a shock. In college football, basketball, baseball, and other sports, we have seen plenty of those down the years.

Right now, however, there is a growing consensus that technology, notably AI, can be used for sports predictions. This argument is correct, up to a point. The success of ChatGPT and related AI programs has given us tools for analysis that far outstrip what we can do as mere humans, and that is starting to filter into a market for predictions for sports betting and fantasy sports. 

AI Can’t Predict the Unpredictable

Yet, you should remain skeptical about how effective AI and other algorithm-based tools are at predicting sports. First and foremost, as we mentioned, there are moments in sports that cannot be predicted on data models. These range from shock results, like Appalachian State defeating Michigan in 2007, to unbelievable turning points within games, like the New England Patriots’ 25-point comeback at Super Bowl LI.

Secondly, and perhaps more importantly, there are limitations to what AI tools can look at when it comes to sports. Yes, an AI can read millions of data points, and then use that data to come up with a stats-based prediction. Yet, human analysis can be superior when it comes to unquantifiable data points. For instance, consider a basketball player whose points-scoring record does not match up with their performances. AI will predict based on the data, i.e., the points scored and other metrics, whereas we can see the, let’s say, human angle. This is called unquantifiable data.

An important point to consider is that the effectiveness of AI can be overexaggerated through misleading statistics. For example, some models claim to be able to predict with accuracy ranging from 70-87%. That sounds great, but if you have ever gone through a weekly college football schedule and tried to predict the winner of each match, you might come out with similar results. The difficulty is when you put those predictions to use, i.e., by trying to beat the spread or in combined parlay bets.

Everyone Will Be Using AI Soon

Without sounding like we are contradicting ourselves, we believe all of us will soon be using AI for sports predictions in the near future. The technology is simply too powerful to ignore. And it is accelerating. OpenAI recently announced that it would be releasing GPTs, which are going to be specialized and customizable versions of ChatGPT focused on a specific area. These will train on specific subjects, say, college basketball, and get better over time. So, it makes sense that people will use them as companions for their sports bets or fantasy leagues.

However, even if AI becomes superior at sports predictions to an expert human, there is still an issue: sportsbooks will have to model their odds setting against the capabilities of AI. By that, we mean betting markets and odds are not simply plucked out of thin air; they are developed through actuary and algorithms to ensure there is a margin for profit. In the simplest terms: If everyone is betting on Team A because the AI predicts that the team will win, the odds for Team A will drop to reflect that. If there was a foolproof system to always win at betting, then there would be no sportsbooks to take bets.

In the end, it’s important to remember that AI and other data analysis tools are just that – tools. As such, they should be used as an extension of your own analysis and intuition. They are not a complete replacement for it.