'AI-powered judge' takes boxing closer to brave new world it appears to seek | Hacker News Reader

'AI-powered judge' takes boxing closer to brave new world it appears to seek

boxingscene.com

66 pointspseudolus66 comments

I didn’t see it mentioned in the article but rumours are the tech is DeepStrike by the Danish company Jabbr.

Fury’s expletive laden review (video): https://www.sportbible.com/boxing/boxing-news/tyson-fury-ole...

> “Absolutely s*t. Here's what, f*k all the computers. Keep the humans going. More jobs for humans, less jobs for computers. And f*k electric cars too while we're at it.”

The AI appeared to be absolutely right, however. There wasn’t a single round where Fury landed more punches.

Are we sure the CompuBox punch stats weren't part of the input for the model after each round?

Most sports would profit from a deeper understanding and application of statistical inference. Or even of descriptive statistics. The fun lays in the operationalization, a step most applications of AI will likely omit, since AI will work out the dimensions of the data by itself.

Boxing for example could calculate the amount of kinetic energy brought to the table and how much of it landed or was sidestepped. How much was absorbed? Not easy to do, but also not impossible.

Frankly, boxing fans seem the least likely to care?

No, this is a stupid idea for people who don't know anything about boxing.

Effective aggression and ring generalship are subjective human judgements. Boxing results are varied because in a championship fight, many times rounds will be completely subjective.

If you made boxing judged by AI, the rules would quickly be gamed. I suspect it would turn into a contest of jab/output and completely ruin boxing.

What ultimately makes boxing great is a fight today doesn't look that much different than a fight 100 years ago. That is the whole fun of it.

It is a really a symptom of a culture obsessed with scientism. As if adding a bullshit layer of scientism makes things somehow better and "smarter".

I don't understand. Why would an AI judge inherently be more perceptible to being gamed than a human judge?

I believe the argument would be that humans are inconsistent, fallible, and gameable in idiosyncratic and individual ways not that they are less susceptible than AI.

Computer systems fail systematically, humans fail more randomly (for most classes of failure)

> humans fail more randomly

And often intentionally

Yep. The in-a-nutshell argument for AI is that there's an unusual amount of bias and corruption in boxing judges, who usually determine the victor. The in-a-nutshell argument against is that less subjective scoring systems like punch counting have already been tried as an alternative in Olympic boxing, but fans didn't like the adaptations in style that resulted. (Of course, dropping the punch counting for the 2016 Olympics immediately resulted in subjective judgements that were interesting at best...)

Sure but then you at least have basis to complain instead of “that’s just how the game is now.”

I don't see why we would need to say that, when we can debug and fix the system.

Not sure this holds, especially given the context of boxing, which is notorious for corruption.

A computer can only judge based on things that can be measured. Martial arts (and really sports in general) rely heavily on things that can't be measured such as aggression, spirit, toughness, etc.

They get measured when it matters - in who gets beaten down at the end, yes?

I count myself as a pretty decent fan of boxing. To argue that the rules are not already being gamed is pretty shallow. Winning by points is pretty much standard. Providing one additional judge that auto-tallies wouldn't kill the sport.

And I do believe the sport would benefit from, say, 3d reprojection to see a different view, or a heat map of hit vs throw locations, or a reproduction of movement throughout the ring.

Other sports have had these deep stats for a long time, and boxing has what, jab counts, points, and knockdowns? Come on. I love the sport and I love data.

A toubling aspect of the "bullshit scientism" is that confidence/belief in "AI" seems proportional to human mistrust of other humans. Therefore it's success rests on fomenting division and mistrust amongst people and iconoclasm of traditional skills/knowledge. Artificial ("intelligence") is, in it's very definition, deceptive.

Which is ironic because trusting AI is fundamentally trusting the humans that made the AI over those that the AI replaces...

Good comment. When we parse, data science, and algo a barely structured brawl, sports are donezo and we get too far away finally from being human.

> …completely subjective.

That may be. But all the gamblers demand objectivity. Whether or not this technology provides objectivity is not knowable now, but it has the empathetic storytelling that it does, which is all that matters for the marginal gambler.

> If you made boxing judged by AI, the rules would quickly be gamed.

I though boxers were already gaming rules and this is effectively what gave rise to stuff like MMA.

Fans got tired of seeing a victor based upon technicality and skill and rules. Fans wanted a victor based upon being beaten down.

It's kind of interesting in that I'm not that interested in this for boxing but would rather see these kinds of systems in stuff like gymnastics or ice skating. Gymnastics and ice skating get absolutely skewered for the fact that you have to be "politically connected" in order to score well over time. I suspect that having an AI scoring system that winds up scoring performances correctly regardless of political connection would be godsend to those kinds of sports.

>Boxing for example could calculate the amount of kinetic energy brought to the table and how much of it landed or was sidestepped. How much was absorbed? Not easy to do, but also not impossible

It would be pretty difficult to do in a combat sport without sensors. As I understand it fencing has participants' blades wired up for this exact reason. How would you measure how hard a punch landed from a distance? You could put sensors in the gloves, but some of the time the punches bounce off the opponent's forearms, so you could get a false 'powerful' rating from a blocked punch.

Worse, some of the best punches aren't the most powerful in a kinetic sense, but just happen to very accurately land in the right spot on an opponent's chin. Or they're well timed and the opponent doesn't see it coming, so their surprise makes the strike more damaging. Even if you could precisely measure punch impact to the head from a distance, you'd be missing out on some excellent punches that are less powerful but set up by accuracy & timing. So yes it's basically impossible to measure in any way but subjectively

F=MA Several very high speed and high resolution video cameras could collect all this information based on how glove, forearm, head, etc accelerate/rotate in real time.

You don’t need to process it all in real time. A blow by blow after round highlights real based on the most damaging blows could grow the sport by making it more interesting to watch lesser matches.

The discussion was about scoring. In order to score the fight, you do need to process it in somewhat real time. I don't think waiting a few days to find out who won is going to be acceptable to fans

Sure, which is why AI scoring is a more difficult problem than creating an automatic instant replay with approximate forces.

I’m still thinking minutes here which kind of blurs the long with real time scoring.

You could probably do it acoustically to some degree. A punch is going to violently displace some air proportionately to its impact.

Also, It might be undetectably negligible, but it would be an interesting experiment to see if a sufficiently sensitive thermal camera would be able to gather information on how much the air is getting compressed in advance of a punch.

Respectfully, I think the noise of the crowd + the fighters might make that a non starter :(

You bring up a good point, it definitely a difficult environment. However there’s a lot of techniques for isolating sounds in a noisy environment.

Getting clean audio from performers at concerts is only moderately challenging these days, with plenty of great examples to point to. So I wouldn’t be so quick to dismiss the idea.

There's a lot more to it than just measuring force. For an extreme example, imagine delivering the same punch (force, target, etc.) to Mike Tyson in his prime versus a 90 year old lady.

People who know how to absorbed and redirect a blow minimize how much their brain gets tossed about, but measuring peak acceleration of the skill is going to show how much someone’s brain got rattled. It’s a solid structure and your brain is inside

Also a you don’t see 90 year old grandmother’s in the ring. So it’s true someone with a larger head has an advantage here, but mechanical properties of tissues should be fairly similar between fighters and acceleration accounts for the mass of the head.

Or apparently Tyson in his prime vs Tyson in his late 50s.

It's not only about power but about unbalancing the opponent in order to obtain a knock-down. A weak hit making him fall is better than a massive blow (especially on hitting a strong body area).

You mentioned fencing: the electronic scoring system in fencing is pretty primitive, and has not at all replaced the need for a referee. It determines whether a hit has landed, and that’s about it. Unfortunately the sport still has a big problem with how subjective the refereeing is (particularly in sabre, which is my weapon), and that’s driving corruption at the highest levels [1].

https://www.nytimes.com/2024/05/09/world/europe/fencing-olym...

Modern fencing is also just about the worst case scenario for a martial art being watered down in the name of sports. At this point it's basically glorified tag.

I strongly prefer watching kendo over fencing because of that; the human element in fencing is essentially a technicality as far as I'm concerned as just a guy watching.

What's competitive kendo like?

You might enjoy this video! The actual topic is the history of Aikido, but it shows some neat historical footage of, for lack of a better term, pre-Kendo Kendo. That is to say, Kenjutsu/Iaijutsu schools sparring using Kendo gear.

https://www.youtube.com/watch?v=D7yyqLyPGDk

Accurately assessing the quantity and quality of punches landed is entirely tractable. If you can accurately track the movement of each fighter's joints (plausible with camera-based CV, almost trivial with LIDAR) then you're just solving a fairly straightforward dynamics problem. A well-timed hook or uppercut to the chin is more damaging for predictable biomechanical reasons - rotational trauma causes more damage than translational trauma because it results in greater shear forces within the brain, particularly the brain stem. It isn't a massively more difficult problem than the doppler radar systems used to track ball movement in sports like golf and baseball.

I think the harder problem is assessing the subjective factors mentioned by shadowerm, but that's also a hard problem for human judges.

Some sports ready have - baseball and golf famously record effectively every action any player does. American Football makes a lot of claims/hype but we don’t get access to data like in baseball/golf.

Geopolitics and many other extremely important domains would too.

I think it is going to become increasingly difficult to keep the various manufactured illusions our culture is composed of together. Hopefully the transition isn't too tumultuous.

Analytics has arguably made basketball worse, where everyone stands around the 3 point line, and the offense is mostly trying to shoot a 3 or go to the rim. Game used to be more diverse.

Analytics changed a lot of things in baseball too.

For example, analytics showed that stealing bases was a waste, and we saw a 6+ year window where hardly anybody tried to steal bases. But then, in 2023 steals were back in fashion. Why was that? More advanced analytics showing that the original analysis was wrong? Teams realizing that after years of not being forced to defend against the steal makes them more likely to be successful?

Basketball will adjust. If nobody has to defend against a midrange jumper then eventually nobody will be in the position to defend against a midrange jumper and it becomes an easier shot to take.

I’ve seen the complaints: the Bulls and Celtics are ruining basketball by shooting 50+ 3’s a game. But if that ruins basketball then basketball was weak to begin with. I mean seriously, neither the Bulls nor Celtics are getting anywhere close to the championship. Let them shoot 100 3’s a game if they want. Who cares?

I don't entirely disagree, but your point is undermined somewhat by the fact that the Celtics won the championship literally last year, and are currently favorites to win again this year.

That's a good point, and a serious brainfart on my part :)

I'm not sure that's realistically possible. The effect of a punch is going to depend on where exactly it landed, the angle of the punch, whether it landed solidly or glanced off, whether the person receiving the punch rolled with it, the physiology of the person receiving the punch, what punches the person has already received, how the person receiving the punch breaths, etc.

Debatably there's other factors that should also factor into a judge's decision such as how aggressive a fighter is.

Baseball already does this for things like sprint speed, bat speed, contact point, attack angle, exit velocity, and launch angle with Weather Applied Metrics added to show how the wind robbed your team of a home run. For pitches there is trajectory, release point, spin axis, seam orientation. Then on defense, a players starting location and their path distance speed to the ball, catch probability and arm strength. That's not even all of it. MLB parks have many multiple high frame rate cameras installed along with their own datacenters. They are also working on replacing the umpire & strike zone with an Automated Ball/Strike System (ABS); which I am personally not to happy about--that would be like taking those awkward fist-fights out of hockey.

https://technology.mlblogs.com

All that, while very cool technology, is going to be much simpler than judging how hard a punch hit. You could gauge how fast a punch was with the same sort of tech as that, but how well it connected, and how much force was applied, are much more difficult.

The ai judge scored it 118-112. Usyk won the fight, but maybe by 2-3 rounds and not 6. Boxing is way too subjective.

I had it 116-112 and had the first as a swing round that I gave to fury. To me it was usyk controlling the fight. The commentators would have you believe otherwise. Remember that, when ring side, fights look different depending on where you sit, which could explain some of this.

Deep Strike was only run on only the public feed and not a quad feed as is preferred.

Shitty hit piece, nothing to do with brave new world (doubt the author has even read it), you can go claim your check from whatever think-tank hired you.

As someone whose favorite book was Brave New World growing up, I've found that it's unfortunately become a catchall for any technological breakthrough with (ostensibly) chilling ramifications for the future.

They should put impact sensors inside the the gloves and use those readings to flash points on the screen video game style.

Too easy to game since the sensor couldn't detect where the glove hit.

Watching two boxers hit themselves and the floor would be quite funny.

Fencing manages to detect a touch, so this combined with a force sensor could be a pretty neat idea.

For a sport as impure as boxing, I feel like there's nothing left to lose. We might as well use it as a test bed for experimental ideas.

Can it also detect traumatic brain injuries?

Ah. The dissolution of the 'the sweet science' by a science that is indifferent to taste. AI continues to draw attention to both the flaws that age-old customs accumulate over time and remind us of the more intimate aspects that have been lost as custom turns into commodity.

Congrats this website features a video where it provides the same as an audio only interview. But if you scroll down or up it stops. If you leave the tab it stops. Thanks for nothing. I hate the internet. Stop forcing me to change my behavior when you are the one who is wrong.

"Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting."

https://news.ycombinator.com/newsguidelines.html

Reply on news.ycombinator.com