Betting On Horses with No-Code AI

Akkio's platform was able to build a money-making model with relatively little data

5 min read
Nine neck-and-neck horses gallop over a dirt surface in front of crowded stands

Horses break from the starting gate at the Saratoga Race Course in Saratoga Springs, New York.

Horsephotos/Getty Images

Artificial Intelligence is altering online gambling, shifting the odds away from handicappers and card counters and point spread calculators in favor of data scientists who can code. But that has left the non-coders of the world out of the money.

So, when the founders of the no-code AI platform Akkio offered to let a curious novice use their automated machine-learning system to bet on the races, it seemed like a chance to even the score.

Getting rich from AI-aided online gambling has been a dream of many software engineers ever since IBM's Deep Blue beat the world champion chess player Gary Kasparov in 1997. Since then game after game, from lofty Go to gritty Texas Hold'em poker have been mastered by almighty AI. As a result, the Internet is full of tipsters selling AI-generated betting strategies while online gambling companies play a daily game of whack-a-mole to keep so-called botters off their sites.

But using AI to gamble is not easy. It requires programming skills for data analysis and for scraping data from websites. It requires setting up servers, preparing dashboards and working with databases. However, the advent of no-code platforms, which allow users to manipulate a code base with graphical interfaces, are suddenly making AI available to non-coders. Akkio is one.

"All new technology needs to follow a progression from scientist to engineer to subject matter expert as the user," says Akkio co-founder Jon Reilly. Of course, Akkio wasn't built for gambling. It was made for businesses that want to use AI but can't afford expensive data scientists—or data scientists who don't want to waste time building models. Sales teams use it to rank leads according to how likely they are to result in a sale, for example. But Reilly agreed that the exercise would be useful to validate Akkio's AutoML (automated machine learning) platform.

"Horse race betting is an example of how a platform like Akkio can turn around a custom ML model and produce prediction outputs in 10 minutes," he said. "The applications in businesses are practically endless."

"I'm sort of like the plumber that comes in and makes sure all the data is flowing"

All Akkio needed was data to run through its prediction engine. Enter Chris Rossi. He's the horse betting expert who helped build a thoroughbred data system, TimeformUS, that was eventually bought by the horse racing information conglomerate, DRF—otherwise known as the Daily Racing Form. He now acts as a consultant to people in the horse-racing world, including what he described as teams of quantitative analysts who use machine learning to game the races.

His role is mostly data cleaning. "I'm sort of like the plumber that comes in and makes sure all the data is flowing," he said.

Rossi knows a lot about horse racing, and about the shadowy teams that bet billions a year on the races, reaping huge rewards—some of it from volume rebates on losing bets by the tracks who encourage the practice. You see the more people bet, the more tracks earn. "Horse racing gambling is basically the suckers against the quants," said Rossi. "And the quants are kicking the [crap] out of the suckers."

Rossi agreed that the track is a good testing ground for AI models. "Horse racing is great for something like this," he said. "It's cut and dry; the money doesn't lie." He also believes that there is a vast and growing market for no-code AutoML. "This is what people want," he said. "They don't want to sit in the weeds and try to learn Python."

Rossi sent Akkio's Reilly 700 rows of training data consisting of the history of horses scheduled to run 10 races a few days later at Saratoga Race Course. The data consisted of:

  • The horse's ID
  • The dates of the horse's previous races
  • The track where those races took place
  • The distance of each race
  • Whether the race was run on a turf or dirt surface
  • The horse's finishing position in each race
  • The purse, or money that the horses were racing for
  • The implied probability, a proxy for the betting odds given on the horse at the start of the race
  • The trainer's ID
  • The jockeys ID
  • The number of the times the horse has raced
  • The age of the horse

Reilly added a "race ID" column to help the model learn how different horses do against each other and added a column for the probability that each horse would finish first, based on past performance. The goal was to run that data through Akkio to predict the implied probability for each horse and its finishing position.

"If you're identifying post time favorites at a good rate, the wins will come"

Rossi said he expected the data to yield 2 winners out of 10 races. "You can get value out of a surprisingly small amount of data," he said. Once the data was loaded using Akkio's drag and drop web interface, Akkio did some feature engineering, automatically deciding which columns in the data are going to be predictive. Next, it searches for the best algorithm to solve the problem, builds a model and trains the model on the data. You can choose the length of time that you want the model to train, from 10 seconds to 30 minutes. Of course, the longer you train the model, the more accurate its predictions will be. Once Akkio built the model, it createsd a URL for the model for sharing or embedding in a webpage.

Then race day arrived. Akkio's predicted implied probability successfully mirrored the race time odds, itself an impressive feat. "If you're identifying post time favorites at a good rate, the wins will come," Rossi said. The predicted finishing position turned out to be not very useful overall. But going by the highest probability of finishing first, Akkio successfully identified the winner in six of the 10 races. Because in one race the horse with the highest probability didn't run, the success rate was really six out of nine.

No bets were placed on these races, but had bets to win been placed on the horses that Akkio gave the highest probability of finishing first, the return on investment would have been 140 percent, more than doubling any money wagered.

The exercise was repeated a week later, this time with money on the line. Rossi sent Reilly a similar spreadsheet for 8 upcoming races at Del Mar Thoroughbred Club in California. Using the online betting service, TVG, bets were placed, putting $10 to win on each horse that Akkio gave the highest probability of finishing first.

The first pick, the favorite to win, stumbled out of the gate, rallied strong but finished second. In the second race, Akkio's pick led the pack but fell behind in the home stretch to finish 5th. In the third race, Akkio's horse finished fourth. The pick in the fourth race was scratched. Finally, in the fifth race, Akkio's horse won, but that was the sole win for the predictions that day.

Still, Akkio's implied probability again closely mirrored the actual post time odds of the horses, and the horses it gave the highest probability of finishing first were in the lead for most of the time in each race, twice losing by a nose. And, the one winner paid $27.20 for every $2 bet, so the $10 wager returned $136, nearly double the $70 wagered overall (the track refunded the $10 bet on the horse that was scratched). "Like Louis Armstrong said about music, 'there's only two kinds, good music or bad music," Rossi said. "This is good music." He suggests increasing the model's parameters. "Every column that you add from here forward will just make it better and better," he said.

Of course, picking seven winners out of 16 races may not be statistically significant—that would require achieving positive returns over hundreds of bets above chance level—but a 43.75 percent success rate is about ten points better than the betting public. And the horse-betting experiment will continue to see how Akkio does over an extended period of time.

The Conversation (1)
Brian Bixby 15 Sep, 2021
M

Well, was wondering what I was going to do after I retire, sounds like ML might be worth looking into. :-)

Can This DIY Rocket Program Send an Astronaut to Space?

Copenhagen Suborbitals is crowdfunding its crewed rocket

15 min read
Vertical
Five people stand in front of two tall rockets. Some of the people are wearing space suits and holding helmets, others are holding welding equipment.

Copenhagen Suborbitals volunteers are building a crewed rocket on nights and weekends. The team includes [from left] Mads Stenfatt, Martin Hedegaard Petersen, Jørgen Skyt, Carsten Olsen, and Anna Olsen.

Mads Stenfatt
Red

It was one of the prettiest sights I have ever seen: our homemade rocket floating down from the sky, slowed by a white-and-orange parachute that I had worked on during many nights at the dining room table. The 6.7-meter-tall Nexø II rocket was powered by a bipropellant engine designed and constructed by the Copenhagen Suborbitals team. The engine mixed ethanol and liquid oxygen together to produce a thrust of 5 kilonewtons, and the rocket soared to a height of 6,500 meters. Even more important, it came back down in one piece.

That successful mission in August 2018 was a huge step toward our goal of sending an amateur astronaut to the edge of space aboard one of our DIY rockets. We're now building the Spica rocket to fulfill that mission, and we hope to launch a crewed rocket about 10 years from now.

Copenhagen Suborbitals is the world's only crowdsourced crewed spaceflight program, funded to the tune of almost US $100,000 per year by hundreds of generous donors around the world. Our project is staffed by a motley crew of volunteers who have a wide variety of day jobs. We have plenty of engineers, as well as people like me, a pricing manager with a skydiving hobby. I'm also one of three candidates for the astronaut position.

Keep Reading ↓ Show less