AI falls on the final furlong in predicting Kentucky Derby winner
The Kentucky Derby, one of the three races which make up the Triple Crown in US horse racing, was won over the weekend by Always Dreaming, the 9-2 favourite which came home by 2 ¾ lengths in cool and damp conditions. Yet for one company which placed its collective AI efforts on predicting the runners and riders correctly, Always Dreaming did not feature in their winners’ enclosure – despite getting the result spot on last year.
Unanimous AI, founded in 2015, offers what it calls ‘swarm AI’ technology which “amplifies human intelligence, empowering groups to harness their collective knowledge, wisdom and intuition by forming real-time AI systems,” in the company’s own words.
In other words, Unanimous AI aims to build a ‘super expert’ harnessing human capabilities with machine data usage. Partnering with TwinSpires, a horse racing betting firm, the aim was for not just a combination of individual race picks from the experts, but building an super expert and having them ‘think together’ as a real-time swarm moderated by AI algorithms.
Last year, the company made headlines by correctly predicting the superfecta – the top four horses in exact order – at odds of 542 to 1. This time round, the company’s picks were Classic Empire, which finished 4th, McCraken (8th), and Irish War Cry (10th). Always Dreaming was ranked fourth by its system, while Lookin at Lee, a 32-1 outsider, confounded the experts by finishing second – with the odds of the superfecta as a result being a whopping 76,000 to 1.
Prior to this year’s race, the company, led by Louis Rosenberg, was confident yet cautious about its chances. “While predicting sports always involve a large element of chance, Unanimous AI taps the intelligence of groups and evokes the best possible prediction based on the available information,” he said. “We have seen this work in a wide range of fields, from forecasting movie box office to predicting the price of Bitcoin. We are excited to see how these handicappers do against one of the most unpredictable of events.”
This time around, the post-mortem was more circumspect, although the company noted the odds were more significantly against them this time around, as well as adding it had still placed more horses than the average individual expert.
“The swarming process amplified the intelligence of the experts, boosting the average performance from 1.6 horses correct up to 2.0 horses correct. That means the experts would have been better off, as a group, going with the swarm than going with their own individual picks,” the company noted. “But without 32-1 Lookin at Lee in anyone’s forecast, the players’ pool missed out on the massive superfecta.”
You can read the full post from Unanimous AI here.