Could dressage judging become unsubjective?

[Trot_on]

I've just been reading Pammy Hutton's opinion piece on dressage judging, and it got me thinking about how, with the use of modern tech and AI that judging could become unsubjective... I don't think it ever should be 100% as there needs to be room for artistic interpretation and emotion in judging but it could help with supporting correctness of movement etc.

A friend of mine has a vet who uses gait analysis technology which measures what the eye cannot see with gait inconsistencies, and it'd be interesting if say, something was used alongside human judging that gave a % for things like gait inconsistency, BTV, angle of curb etc which, when put together with the human judges note on harmony, rider ability, artictic license could support a fairer and more consistent dressage scoring system in the future.

Interested to hear other's thoughts on such a thing...

 

[stangs]

Using AI makes no sense if you want to make the judging more objective/horse friendly. All it would achieve is making the current bias towards BTV + flashy-moving horses even stronger.

I like your idea of using gait analysis technology (potentially to calculate the frequency of conflict behaviours as well?) but it would need a lot of development and testing to be considered even remotely reliable, and you’d have to develop a new style of marking to make sure the judges are incorporating the information into the score in a consistent way.

 

[equinerebel]

There's been discussion of using this technology in figure skating, which also needs to find a balance between fair scoring (country bias in this case), and artistic interpretation.

 

[JFTDWS]

Using AI makes no sense if you want to make the judging more objective/horse friendly. All it would achieve is making the current bias towards BTV + flashy-moving horses even stronger.

I like your idea of using gait analysis technology (potentially to calculate the frequency of conflict behaviours as well?) but it would need a lot of development and testing to be considered even remotely reliable, and you’d have to develop a new style of marking to make sure the judges are incorporating the information into the score in a consistent way.

There's no reason for this to be the case - AI will do what it's told to do, within limitations, and if it's programmed and trained to reward the head on, or in front of, the vertical, and penalise flashy but incorrect movement it will. More reliably than the average human - if it's programmed with that intent, and done well.

I'm not convinced current AI tech is capable of doing it well yet, but that's a different issue.

 

[Alibear]

Just chuckling at AI doing what’s it’s told and the problems it has as it creates hallucinogenic data.

Getting something coded/trained on a closed data set but that could still cope with the vast variations in horse shapes , gaits and ways of going and the need to judge relative to those. You’d need a very wealthy investor to get it built and also get/film the correct videos to train it on. It would be a very interesting project but I’m not confident it’s feasible.

 

[JFTDWS]

Well most of the problems with AI in general is that it's recursive - having been trained on real world data (not always reliable, but that's a dataset problem not a theoretical AI issue), it adds to its own dataset, and it trains on its own products, which is when it gets really bonkers. Obviously that wouldn't apply to a hypothetical well designed dressage judging program.

And, to be fair, I did qualify that statement - if it's done well. Most AI that we're seeing right now is not done well (or ethically, but that's another matter).

 

[Burnttoast]

Given there is a lameness app out there being used by vets already presumably it wouldn't be beyond the wit of man to add things like position of poll, verticality of face etc to that kind of technology and position cameras appropriately.