Artificial intelligence (AI) will greatly broaden its role in sports betting as soon as Super Bowl 58, according to Microsoft Director of Sports Jon Flynn.
Casual bettors using language models like ChatGPT and sharps enhancing their algorithms with AI will affect wagering like never before, Flynn told the Gaming Today Playbook podcast.
It was all a matter of time. But Flynn believes the time comes on Feb. 11.
“We’re going to see the rise of the casual gambler that is informed only by AI,” Flynn said. “I think one of the litmus tests is, I think the Super Bowl this year, we’re going to see a rise in the casual bettor who’s using AI, specifically ChatGPT from OpenAI or Bard from Google or Gemini or Llama from the Meta open source version into informing who and how they bet. I think that’s going to be a great step for us to look at.”
Microsoft has an interest and expertise in these developments because its Azure Cloud technology can be used to store and process information needed to build or bet markets with greater speed and insight.
But language models, which have shown their fallibility in a year of high use and visibility, won’t just regurgitate winners on command. But they can lead a bettor down a path to an informed decision with solid data and a series of educated questions.
“If it gives you a response that you’re not satisfied with, ask it again,” Flynn said. “But it’s going to give you some sort of an answer. It’s not programmed to give you no answer in its current sort of open AI form, if we’re sticking with ChatGPT.
But there’s also a way that you can, through using the Azure platform, you can create your own Azure open AI instance where you can upload data. And it could be Excel spreadsheets, it could be text blocks, it could be a model that you have, and then you can query that data itself. And that may be game data. That may be historical game data against these two teams.”
Then the human has to make decisions.
“And then how are we going to bet, based upon what I know here? That’s going to give you more targeted information about this,” Flynn continued. “Because remember, the large language model, if you just log on to OpenAI, it’s giving you data on everything because that’s the billions of parameters that it’s been trained upon. That’s probably a little too broad.
If you narrow that down and you can instruct it to concentrate on your sports and concentrate sports for the last X amount of years, or the last X amount of seasons, and this positional player and the statistical analysis of that defender against this opposite, you can do all of that. But if you put it down into your own instance, what you can then do is you can then build upon the answers … ask it to add onto what it already knew, that you are satisfied with, and then add on. So it’ll then eventually build a whole picture.”
“I always caution everyone, if you ask any large language model — depending on whose it is — a question, you take that output and you run with it, you’re probably doing yourself the disservice because it may be right … but is it right enough?”
GT: Will humans be fast enough to process and exploit all this information for in-game betting?
JF: I’m going to lean back to high-frequency trading. A lot of those trades weren’t made by humans. A lot of those trades were made by algorithmic execution engines that had a threshold built into them.
If these parameters are hit, make that trade. If these parameters are hit, sell, buy, whatever the case may be. Hold.
So if you look at the game of ping pong, competitive table tennis, it is one of the fastest games on Earth. The way they’re hitting the ball, you can’t even see the ball. You just hear it. You could bet how many hits each player will have per point. You can bet on anything that you want, but there’s no way you can see that and count in real time what’s going on.
So you can then create an application to bet for you on your behalf with those parameters that you want. Because number one, it may be too fast. You may not be there. You may be living.
I don’t want to sit in front of a screen and do betting all day. I may be out with my kids hiking or maybe on an airplane, maybe on a cruise. Who knows?
For me, I’ve got a sense of confidence in either the application I’ve bought or the application I’ve built to go and bet on my behalf. And I’ll put in thresholds and I put in guards that I don’t lose the house. I don’t lose that while I’m away on a cruise with the kids.
And then another paradigm we’re going to see, I predict, is that we’re going to be asked if we want to make this bet. So we’re going to have the app, and I’m going to get something from FanDuel. They’re going to go, “It is the fourth quarter. It is a potential buzzer-beater here. Do you want to put $20 on the line?”
Boom. All the backend stuff is taken care of for you. Because that bet needs to be registered now. It’s an opt-in, right? But that’s going to come to you.
Check out the rest of the interview in the Gaming Today Playbook podcast.