Since the launch of ChatGPT, there have been countless articles and conversations explaining how generative AI will transform our lives. But what if it can be used to predict the future?
Betting is a huge industry and football bets make up a significant amount of revenue. The English Premier League (PL), for example, generated over $70 million of bets in the 2021/22 season – more than any other league. What if AI can be used to help us make better wagers?
Cipher Sports’ AI claims it is worth betting $4.40 that Everton will beat Brighton in this weekend’s PL fixtures, and $1.44 that Manchester City will beat Bournemouth.
Cipher claims that either just one or neither team will score at the Villa vs Nottingham Forest game (you could place at least a $2.28 bet on that), and there will be no more than 2.5 goals in the Bournemouth vs Manchester City game ($2.82).
So, how does the AI know any better?
Ultimately, artificial intelligence takes large volumes of data, and with past patterns and outcomes, it will churn out a prediction as accurately as possible – It can forecast the weather, it can tell a factory machine is likely to break soon, even a chatbot’s response to a query is technically a prediction of the answer.
Truly, betting can only be a game of likelihood and chance, and bringing in computer technology has only helped make it more accurate: “People have been doing that for, you know, probably 30-40 years in some form or another,” says Darryl Woodford, CTO of Australian sports betting company, Cipher Sports, “and getting more advanced as computer technology gets more advanced.”
So how is this better than Paul the Octopus, who was used to predict the outcome of international football matches during the 2010 World Cup?
Betting with the help of computer intelligence has advanced from spreadsheet models to coding models, “and generally, what we’ve seen is everyone’s taking a very similar approach that they’re trying to simulate the game as accurately as possible and predict how likely different outcomes are and extract results from there.”
But, with the help of AI, over the last few years a lot more computer power has been thrown at it, says Woodford, “than ever before, and because you can run these models a lot faster, you can put a lot more data into them.”
Cipher Sports launched in 2021, with a predominant focus on the US market although it covers sports globally – It provides referrals to sportsbooks such as BetMGM, DraftKings, and PointsBet, and sells services to media companies and sportsbooks too.
Covering a range of sports including football, American football, baseball, basketball, horse racing, and more, Cipher Sports predicts the performance of individual players, to predicting match outcomes.
“My grandfather was a jockey in the UK, he did races in the southwest of England, so it’s kind of in my blood a little bit,” says Woodford.
Where to put your money
Earlier this month, the firm took to predicting, what is now US’s most-watched programme ever with almost 124 million views, American Football’s Super Bowl between San Francisco 49ers and Kansas City Chiefs by jersey number.
Uniquely, the betting firm used its in-house touchdown projection AI, based on its model’s simulations combined with what the firm knows about the rosters of each team to attempt to “identify an edge against the books” for the first touchdown scorer.
It predicted that the first touchdown had a 22.6% likelihood to be player #23, Christian McCaffrey, followed by player #1, Isiah Pacheco (11%), and player #4 Rashee Rice (8.9%).
While Woodford says the results went well, and “it showed decent results in the day,” the numbers were not too dissimilar to most other betting companies who also had McCaffrey as the most likely to score the first touchdown (he did).
“One thing I always say is the games like the Super Bowl are generally the thing you want to bet on the least because they are the most efficient markets out there you’ve got.”
“Everybody wants to predict it, everybody wants to bet on it, the markets get tighter and tighter,” says Woodford, “you’re just competing against everybody else who wants to put a large volume of money into those.”
Instead, his advice to find success is in the lower league and college football games, for instance, and other lesser-followed sports games. While they are not as big, they still gain significant interest, and there are a lot of statistics and data available for Cipher AI to take and make predictions out of.
“The general kind of approach to statistics in the US just means so much of that data is available,” says Woodford, whether that be football, basketball, or data.
“We cover 20,000 odd sporting events a year across all of those different leagues, and I think one of the things that sets us apart is that we treat college basketball games the same as we would treat the Super Bowl in terms of how much attention and computational power we put towards it,” enthuses Woodford.
“I think being able to provide predictions across such a broad range of games really gives you an advantage because the majority of places you’ll find betting opportunities is smaller happenings.”
Challenges
Of course, computers cannot predict everything, and while there are large data loads out there for it to be given, there is intuition and gaming knowledge that still demands human intervention.
For instance, back in 2021, the Rugby League changed its ‘Six Again’ rule that meant ‘ruck’ infringements from penalties instead awarded the attacking side six more tackles.
“It basically changed the whole speed of the game,” says Woodford, and if a computer receives data from 2018 to 2024, but there were big rule changes in 2021, the computer will not take that into account or understand.
“Or, it won’t have the domain knowledge to say, okay, these run metre statistics, for example, is going to be massively inflated now because there are more plays in the game, so there are more opportunities for a player to run.”
It is instances like these where humans and their knowledge are crucial to understanding the statistics it’s giving, “identify what statistics is causing that problem, and change it in the processing stage, or just pull it out completely and say okay we can’t use that as a predictive statistic.”
In the future, the growth of generative AI also weighs on betting firms. One, is because it is on the sports betting companies to build more accurate models to compete with the ever-growing intelligence of accessible AI such as ChatGPT. Two, because those betters who can get hyper-accurate results themselves may be a threat to the company’s profits.
“So, trying to say, okay, do we actually want to take bets from this customer at all or do we better protect our profits by limiting those customer’s ability to bet,” says Woodford.
“I know people that it has happened in the UK, Australia, and the US. If you are good at what you do, the makers will eventually cut your limits and will not be able to take bets.
“It’s not really a fair fight in that.”
