Machine Learning in dressage judging

[Demerara_Stables]

I, for one, welcome our AI overlords into dressage judging.

https://www.chronofhorse.com/article/annual-dressage-stakeholder-meeting-highlights-technology-assisted-judging-rebuilding-trust/

“research physicist David Stickland launched a startup company, Global Equestrian Technology, which is developing a software that measures horses’ movement using modern video analysis tools. It can precisely track skeletal points on a horse’s body and use them to measure things from whether the horse executed the correct number of one-tempi changes in a test, to how far it moved forward in its 12 best steps of piaffe, to more subtle things like the angle of its face relative to its chest or how much it over-tracked in the walk. The program can be trained to correlate these so-called “observables” with numerical scores typically assigned by judges.”

[NJRider]

I agree. I doubt the 68 plus year old S judges will however. Their priority is clinging on to their roles.

[SillyHorse]

I have mixed feelings about this. It would be good to remove bias from judging, though.

[initiate1987]

It doesn’t sound like human judges are meant to be fully removed with this technology. David wants the judges to be able to focus on the quality rather than the quantitative stuff like how many steps did the horse rein back, how far did the piaffe travel, etc. It could evolve further into assigning scores, but at the moment, it sounds like that’s not the intent.

[Libby2563]

Interesting. Sounds like they’ll be focusing on more mechanical/quantitative aspects first, like number of steps/tempis. But if quality is ever to come into it, someone would have to instruct the machine on what’s good and what’s not as good. How might breed and other biases play into that?

Also although the intentions are good, I wonder whether judges agree that this would be a step in reducing “the complexity of judging.” As a parallel, electronic scoring/scribing should be simpler but often isn’t viewed that way due to technological glitches or the learning curve. Now judges will have to make their own assessments plus pay attention to a computer beeping that someone has possibly done too-few tempis, and weigh their trust in the program against what they perceived. I foresee a lot of hesitation.

Is there a future in which drones assist by providing overhead video to evaluate accuracy and bend/straightness?

[luvmyhackney]

I think it would be cool to let human judges still give a score, but rate it against the electronic scoring. Then variances can be reviewed, etc.

[pluvinel]

@Demerara_Stables

…“research physicist David Stickland launched a startup company, Global Equestrian Technology, which is developing a software that measures horses’ movement using modern video analysis tools.

Stickland is a physicist at CERN…he seems to hang around FEI types and created a company called Global Dressage Analytics back oh, about 15 years ago.

[atlatl]

Not to be too much of a nerd here, but what is described is neither artificial intelligence nor machine learning IMO. As described, it sounds more like an analytical program that is implementing a series of rules as defined by whomever is developing such “a software.”

In AI and machine learning, the system develops its own rules based on training data provided. The classic example of faulty training data is related to distinguishing wolves from dogs. Long story short, all the training pictures of wolves had snow in the background so the system concluded that snow was a defining characteristic of being a wolf.

Now, if they do integrate quality which is more subjective than the objective things mentioned, that would be getting into AI. Of course, there’s the whole training data problem all over again.

What I wonder about is whether anyone is going to invest in this startup. Seems a pretty small and niche market.

[atlatl]

@pluvinel

@Demerara_Stables

…“research physicist David Stickland launched a startup company, Global Equestrian Technology, which is developing a software that measures horses’ movement using modern video analysis tools.

Stickland is a physicist at CERN…he seems to hang around FEI types and created a company called Global Dressage Analytics back oh, about 15 years ago.

They are still around; I only know because I did an internet search.

Looks like a handy tool for those that like to analyze their scores and trends. I always just used a spreadsheet, but hey if someone wants to spend 20-50 euros a month that’s fine too. Their last news post is from 2016…

[Quincy]

Inserting “AI” into whatever is being sold is the new trendy thing. It has its use, but it’s not the incredible new thing like it’s being marketed as, if the service/product is even using AI (as an aside, AI has been around for a while).

[Demerara_Stables]

Agreed re: comments on AI. This isn’t really AI (I was being sarcastic but I forget that doesn’t come across well on the internet ), and people calling things AI is just trendy.

However, I do really like the idea of using video analysis to objectively score movements and movement quality (specifically the mention of overstep at the walk), as well as start to have data around welfare topics like BTV.

I think there are way too many horses that spend too much time behind the vertical, but until we have data to show how that affects them and how much time they actually do spend behind the vertical, we can’t address the problem.

[Aussie_2020]

The only judge I want to ride for.

[lorilu]

That depends on what horses are used to “train” the AI. If they only use European warmbloods, all
Of us who ride non-traditional horses will suffer. Likewise if they only use horses at the FEI level.

[lorilu]

Were they the Nerd group who analyzed scores for bias?

[Nous]

I agree, the problem being not to produce what is essentially gaits analysis software but to provide software which is capable of quantifying the quality of the training that was put on the horse.

[pluvinel]

@lorilu

Were they the Nerd group who analyzed scores for bias?

The COTH Nerd Herd were a group of nerds who met on COTH. It included a couple of PhD’s, academics, engineers, operations people. They wrote a peer-reviewed paper analyzing 45,000 US scores that was published in the ASA (American Statistical Association) Journal of Quantitative Analysis in Sports.

[Snow_Mass]

“Members of the press were not granted access to the meeting, but Linda Keenan, secretary general of the International Dressage Trainer’s Club, and Klaus Roeser, secretary general of the International Dressage Riders Club, hosted a press conference afterward.
Rebuilding public trust in dressage sport is a top priority for all stakeholders at the meeting, they said.”

I don’t see how hosting a closed door meeting and keeping the press out contributes to rebuilding public trust. They have miles to go before they sleep…