This is the horse ‘PunterGPT’ has tipped to win the Golden Eagle

An artificial intelligence model built by experts on ChatGPT – dubbed ‘PunterGPT’ – has tipped the gelding Tom Kitten to win the $10 million Golden Eagle on Saturday.

PunterGPT was created last year before the Melbourne Cup and predicted winner Without a Fight to finish in the top three.

Nash Rawiller riding Tom Kitten wins in the Up And Coming Stakes last year. Credit: Getty Images

The AI punting tool is the handiwork of Michael Kollo, chief executive of generative AI company Evolved Reasoning, and John Conomos, the head of global quantitative research at Macquarie Group.

Kollo said PunterGPT took a different approach to other more data-driven predictions as it processes words, not numbers, to find overarching trends.

“Our job was to feed it a cross-section of different articles written by different professional punters, with the view that in among all of that expertise, there is a truth,” he said.

“A human would be talking to a bunch of experts on a topic, and trying to work out what’s common or what’s not, or you’d be trying to work out what’s different ... I think PunterGPT was trying to do the same kind of thing a human would be doing, but just with a lot more articles and a lot more subtlety.”

After its success in last year’s Melbourne Cup, Kollo rebooted PunterGPT for this year’s Golden Eagle – and Tom Kitten, starting from the favourable barrier 2, was the name that emerged as its top pick.

PunterGPT’s predictions ahead of the 2024 Golden Eagle.Credit: Michael Kollo

English raider Lake Forest and bay gelding Ostraka made up the first three, but both horses have largely flown under the radar of punters, sitting in the middle of the pack in betting odds.

The ability of AI to process high volumes of data and text faster than humans means it can analyse various factors such as expert predictions, a horse’s previous performance, track conditions, weather, jockey, barrier draw, the horse’s staying power and other data efficiently.

It’s these processing and pattern recognition skills that make AI well-placed to predict race outcomes. In fact, the horseracing industry already employs AI to track performance, data analysis and predictive modelling for future races.

“The probability of predicting winners, even if you have a great model and lots of information, is only marginally better than guessing,” Kollo said. “I think the way that people make money is simply by betting on lots and lots of races, and by doing that, you actually exploit effectively a couple of percentage advantage here and there.”

Michael Kollo and John Conomos have developed an artificial intillegence punting tool.Credit: Macquarie Group

Behavioural finance expert Andrew Grant, an associate professor at the University of Sydney, said that whilePunterGPT was an experiment that recreational punters seeking to improve their odds may find useful, predictions made by AI need to be considered in a wider context.

“You can’t sort of validate models very easily on the basis of, ‘it worked one year and that’s it’,” he said. “Realistically when we’re evaluating forecasts, we need to have a lot of forecasts and a lot of outcomes to really be able to say whether a model is good or not.”

The accuracy of AI racing predictions largely depends on the quality of the data being fed into machine-learning models, according to Professor Richi Nayak, an artificial intelligence and data mining expert at the Queensland University of Technology.

“We always say a machine learning proverb is garbage in, garbage out,” she said. “The better data we have, the better inferencing that we can do.”

Grant also noted issues around data availability and accessibility because bookmakers and analysts have extensive racing data that an average punter typically doesn’t have access to.

“It’s not a simple model that you might come up with by saying ‘this horse has won the last three starts, and therefore it’s going to win this race’. The sophisticated punters will have a lot more variables that they’re keeping track of to put into their models that are not easy for you to pick up just on the form guide.”

There is “always” a degree of uncertainty on race day too, Grant said. “They’re still not robots on a machine. It’s not like a roulette wheel or rolling a die.”

“There are things that can happen in a race itself that can affect the outcome, such as horses running into each other, or something like that ... which wouldn’t necessarily be obvious upfront.”

Sports news, results and expert commentary. Sign up for our Sport newsletter.