Most of us – even those who don’t really follow technology news – will have heard of the big advances in AI over the last couple of years. ChatGPT is getting most of the attention, but there are rivals, including Google Gemini, Meta’s Llama, Perplexity AI, Anthropic, and several more. There is a lot of hype around these products, some of it justified, some of it unwarranted. Yet, some of that hype is starting to die down just a bit as we get to grips with AI and its potential ceiling.
That said, you may have seen online adverts suggesting that you use AI prediction markets, including trading stocks and, of course, sports betting. It’s a bit of a Wild West situation out there, so we would suggest that you be cautious – very cautious. There are benefits to using machine learning to make sports predictions. Yet, there is still no magic wand, and AI has some specific limitations that arguably don’t get talked about enough.
AI is wholly dependent on the data it has access to
The first thing to note is that AI is data-dependent. That’s an obvious point, yes, but it’s important to stress that it can’t reason beyond the parameters of that data. You could feed the latest NBA odds and historical stats, but it is not able to provide the spark of human intuition that is required to look beyond the data. In short, it’s useful for analyzing large data sets – much more than we humans are – but it’s not an oracle that can act outside the boundaries of what it doesn’t have access to.
Secondly, it’s important to point out that AI has limited access to the live web. The newest version of ChatGPT’s underlying LLM model (GPT-4o) has some access to real-time news, but many important sports data providers block the web crawlers that retrieve that information. In addition, some AI providers have signed exclusive deals with media publications to access data. This is beneficial. However, exclusivity suggests exclusions, and those access deals mean some AI bots will “know” things that others do not.
Thirdly, AI is often guilty of a process called “overfitting”. This is not limited to sports but all kinds of data analysis. Overfitting is, in essence, the process of learning historical data sets too well and not being incorporate changes to the patterns. A good way of thinking about it is if you were a college student taking an exam; you got the answers to last year’s tests and learned them off by heart without understanding the broader concepts. But when the exam comes, it’s a new set of questions.
In sports, for example, an AI that learns from historical NBA data might predict that a team like the Golden State Warriors will always rely heavily on three-point shooting because that’s been their style for years. However, if the team suddenly changes its playbook or key players are unavailable, the AI might still predict based on the old pattern rather than adjusting to the new reality. This can be tempered by giving the AI new data to work with, but as we have suggested earlier, that data is not always readily available.
AI is not capable of thinking outside the box
Finally, we should remind you that AI does not “think”; it “predicts”. Yes, we know we are trying to make an argument about predictions, but we aren’t talking about that type of prediction. It predicts in the sense of word association. Indeed, the LLMs (large language models), which are the engine room of the AI, are sometimes referred to as next-word predictors. As such, they don’t reason out their answers; they predict a bunch of words in relation to each other.
The discussion of the LLMs is complicated, and we don’t have the space to go into a full explanation (the next word predictor is somewhat facile, too). Yet, it does serve to underline our main point that AI cannot operate outside of its parameters, at least not yet.
In the end, AI is useful for sports predictions. And, in all likelihood, it’s going to become even more useful in the future. However, it should also be thought of as a number-cruncher rather than some sort of sentient tool. By all means, use it to analyze data that you believe is useful for your sports predictions, but be skeptical about its ability to see beyond the information you give it.
