Water Cooler Trivia’s chief trivia officer, Eli Robinson, shared Monday that he had fed the following question to the much-ballyhooed A.I. chatbot, ChatGPT: “What name is shared by director Stone, politician and military officer Cromwell, and Last Week Tonight funnyman John?”
ChatGPT’s response — “John” — was incorrect, whereas 87% of all humans who attempted to answer this question correctly replied with “Oliver.”
ChatGPT’s answer was dumb, but the bot’s no dummy. In the same web posting, Robinson noted that AI entities like ChatGPT typically answer 95% of Water Cooler Trivia’s questions correctly nowadays, whereas that percentage was just 73% in 2021.
To paraphrase Vince Vaughn’s character in Swingers, our little baby bot’s “all growns up,” to the point where, on Wednesday, some 1,100 tech leaders including Elon Musk and Steve Wozniak sent an open letter to artificial intelligence labs, urging them to pause development of advanced A.I. systems that could present “profound risks to society and humanity.”
“These things are shaping our world,” Gary Marcus, an entrepreneur and academic who has long complained of flaws in A.I. systems, told The New York Times. “We have a perfect storm of corporate irresponsibility, widespread adoption, lack of regulation, and a huge number of unknowns.”
Similar fears confront the sports betting industry.
“A.I. and machine learning are going to become more prevalent, and regulators are going to have to step in to govern what’s going on,” Dr. Kasra Ghaharian, a postdoctoral research fellow at UNLV’s International Gaming Institute, told Sports Handle. “Because of the complicated regulatory landscape, gambling technology has lagged behind other industries. It’s going to be important to bridge that gap between the data science community and regulators.”
But with trepidation comes opportunity — for bookmakers and bettors alike.
Bringing the Derby to the book
Jarrod Barnes played college football at Louisville and Ohio State, earning a degree (B.S. in exercise science, M.S. in sports management) from each school. While in Louisville, he worked at Churchill Downs — home, of course, to the Kentucky Derby.
“I’d escort high-net worth individuals to their seats and help them navigate the venue,” he said of his Churchill job, which he held when he was 19. “One of the things I always enjoyed the most was the expectation of people when they came to the Downs. You knew people were there to have a good time. What I valued the most was the smile on people’s faces. It was almost like a red carpet event to be there — really the whole summer, but obviously the Derby was the biggest.”
Now a professor at New York University’s Robert Tisch Institute for Global Sports, Barnes thinks mobile sportsbooks could do a far better job emulating the brick-and-mortar Churchill Downs experience online — and that A.I. and machine learning can help them get there.
“People bet a lot of money because they’re having a great experience,” said Barnes, who’s also pursuing a doctorate in learning design at the University of Illinois. “If you go to some sportsbooks online, it’s not necessarily intuitive. Amazon Next Gen Stats, they’ve made the experience of Thursday Night Football interesting through machine learning. Even if I know nothing about football and know that Patrick Mahomes has a 25 percent change of throwing to the right side of the field, I might place a prop bet.”
When asked if he thought U.S. sportsbooks could do a better job of harnessing such buzzy technology, Ghaharian responded, “I think it’s kind of unknown at the moment, to be honest with you. From my perspective, I think we hear a lot about A.I. and machine learning. It’s quite a sexy topic, but I don’t really know to what extent operators are using the technology.
“I have seen third-party data companies advertising their data science chops — some companies based out of the UK and other places in Europe that are very much focused on machine learning and AI for responsible gambling and marketing personalization. AutoML — automatic machine learning — takes machine learning a step further where you’re automatically creating models. I think that’s an opportunity for the gambling industry for personalization, recommendation systems, consumer protection.”
Bettors already building bots?
Professional sports bettors and syndicates that use computer modeling and algorithms to identify weak lines to exploit have been around for awhile now. But what if A.I. allows more casual bettors to do the same? And what if the sportsbooks — microbetting platforms aside — don’t keep pace?
“In sports betting, a couple of things are really interesting,” Barnes said. “There can be barriers to placing bets, whether it’s comfortability or knowledge. So one of the first things I think about when it comes to ChatGPT is actually lowering that barrier for entry to placing that bet, because I have a tool or partner to talk about this. The challenge is the data that ChatGPT has been trained on is limited to a 2021 data set. If I place a prop bet on the Super Bowl, I’m not going to get a really good answer, because the data set has only been trained up to 2021.”
But Barnes thinks A.I. technology is advancing at such a brisk pace that 2021 will soon seem like 1921.
“There are folks who’ve attempted to hack together their own personal ChatBot outside of ChatGPT to give them tips for sports betting. It’s highly complex, but it’s possible,” he said. “The ChatGPT technology is only six months old and it’s only going to get better over time. On the consumer side, [there might be] more avid bettors who’d be willing to hack together their own data sets and train the A.I. model on their own data. It could be from this past week’s NBA games, from prop betting to actual outcomes.
“You think about the 45 to 50 percent of fans who are not bettors and the influence of a personal chatbot that could conversationally prompt them to bet. That’s the world I think we’ll live in. We’re seeing the early hacks of very technologically savvy people trying to build that, but that’s where I think we’ll get.”
Ghaharian thinks it’s a bit more complicated than that, but he assumes certain brainiac bettors are already ahead of the pack.
“I don’t think that [recommending certain wagers] would be a function of ChatGPT on its own. Layering a natural language processing engine on top of a recommendation engine, you could maybe do that,” he said. “That would be really cool. Could you layer it on top of the recommendation engine so you’re not just delivering the customer a product, but some real text? Yeah. I think those are two separate things that could be built on top of each other. I’m sure there are people already doing that.”