There are things John Foote can see in a horse that most people can’t.
A renowned bloodstock agent for decades and the trusted go-to equine eye for trainer Tony Gollan, Price Bloodstock and a host of owners and trainers in Asia, Foote is collaborating with researchers behind a University of Hong Kong-funded study aimed at enhancing the understanding and welfare of racehorses through the integration of artificial intelligence, data science, visual computing, and bioinformatics.
The question is, can the equine intuition of Foote and others be machine-learned?
Alex So, who is undertaking the study alongside Professors Loretta Choi and Tak-Wah Lam at the university’s Equine Analytics and Bioformatics Lab, believes it can be achieved.
“With AI, it's not about ‘can a computer see something or not’, it's about who's building the computer. I think that's the main issue that people don't understand (with AI),” So told The Straight at Karaka ahead of Sunday’s New Zealand Bloodstock National Yearling Sale.
“We use maths in order to speak a certain language and that language represents what we see with our own eyes.
“If we have someone who does not have the skillset or experience, someone who is not John Foote or Merrick Staunton, you're not going to build a very good model.”
While one outcome of the project is to be able to use AI to help source quality racehorses - a goal that could be two to three years away - the initial instance is to help identify conformational faults in horses.
“Currently, our machine learning model has quite good accuracy at identifying joints and spots of the skeleton so that it's able to look at an image and say, ‘OK, this is the head, this is the eye, this is the fetlocks’,” So said.
“We're still training the model and wanting it to be better. The next step is videos.”
So suggested that what they are doing is not dissimilar to that of a physiotherapist who works with humans.
“Our end goal is to really be able to look at a horse, analyse its muscles, skeletal structures, bone density, the way it walks and see how much force is being put on each leg, how much force it can take, how much force it should be giving,” So said.
So, Professors Choi and Lam, who specialise in medical visualisation, geometric computing and computer graphics and DNA mapping respectively, aren’t the first to use AI to assess racehorses.
There are other people who have invested in technology to assist in yearling selection, such as Briton Tom Wilson’s Racing Squared, but So suggested it is different to what his study is aiming to achieve.
“He uses an AI generation algorithm called DeepLabCut and, if I'm not mistaken, it was developed at Harvard, but DeepLabCut was developed as a general use case so that that model teaches it to analyse the skeletal structures of multiple animals, all four-legged animals,” So explained.
“But in order to get better accuracy, and we really want to be pioneers and make sure our technology is applied in a good way, not just applied in any way, we're starting from scratch filming the horses, measuring their skeletal structures from every angle possible to get the most accurate model.”
Foote believes advancing technology can assist in identifying quality horses but that there will still be a mix of art and science behind selecting yearlings.
“I’ve been teaching Alex about conformation and that type of thing, and how horses walk and behave themselves, and just general conformation so that he can put that into his computer,” Foote said.
“Technology can help with anything, but there's so many things that go into (selecting a racehorse).”
Twenty-six-year-old So’s passion for racing and horses was sparked through his family’s involvement with the sport in Hong Kong.
So’s father Chris, a foreign exchange trader, has raced horses in Hong Kong for years and, after a “spell”, he has reengaged with the industry alongside his son in recent years under the So Bloodstock banner.
A friend for more than two decades, Foote was called upon to help buy horses on their behalf as well as provide practical insights into the AI project.
“Our end goal is to really be able to look at a horse, analyse its muscles, skeletal structures, bone density, the way it walks and see how much force is being put on each leg, how much force it can take, how much force it should be giving” - Alex So
David Hall, Cody Mo, Tony Cruz and David Eustace train horses for the Sos in Hong Kong while the Sos’ growing racing interests have extended to fillies who are being trained in Australia by Patrick Payne and Cliff Brown in Victoria and Joe Pride in Sydney.
The latter trains Lupa Capitolina, an unraced Almanzor two-year-old half-sister to Group 1-winning stablemate Ceolwulf.
Eustace, who was an integral member of the Ciaron Maher stable as co-trainer during a period in which the business invested heavily in sports science, technology and data, is supportive of the project.
“We want to align ourselves with people who have similar thinking to us and who are very forward thinking and believe in this idea and this approach,” So said.
So, who quit a job in finance to start his own AI business and to embark on the project, wants to expand the study’s dataset by partnering with stud farms and stables throughout Australasia and he intends on being at upcoming yearling sales in Australia both as a buyer and to further his research.
