The Shintaro Higashi Show - How Can AI Help Judo?!? 🤔

Episode Date: March 5, 2025

AI is rapidly changing the way we train, learn, and analyze Judo. Peter breaks down the fundamentals of AI, how machine learning can enhance coaching, and its potential applications in skill development. Shintaro explores practical uses, from video analysis to training optimization, while discussing AI’s limitations and ethical concerns. They also touch on the future of AI in Judo, including data-driven strategies and predictive modeling. Whether you're curious about AI’s role in martial arts or its broader impact on sports, this conversation offers valuable insights and thought-provoking takeaways.00:00:00 Introduction00:00:30 Peter's research and collab with IJF00:06:33 Research timeline00:08:10 Applications of this Research00:15:00 What's next?00:17:29 Hakuin.ai (Sponsor AD)00:17:57 Commentators vs AIIf you're in business, then you have customer churn. Whether you're building a startup, growing a mom & pop shop, or operating in a fortune 500 powerhouse, Hakuin.ai measures, predicts, and improves your customer retention. https://hakuin.ai

And then it's integrating with the scoreboard because it shows exactly when the thing was done, when the penalties were given. And that gets locked automatically probably. Probably, yeah, that would be my guess. It would be the most efficient way. And if I was building that product, that's exactly what I would do.

I should be on Silicon Valley. Yeah, exactly. I could probably fake it. That's actually, you know, but you can't even imagine how valuable that is to AI research. All this like, you know. Hello everyone. Welcome back to the Shinto Higashiyo with Peter Yoo.

How can AI help Judo? What a question. Yeah. And I think in the last episode where we talk about Chachi 50, some people are joking about, we talked about chat 50 people, some people were joking about, we talked about too I guess, can there be an AI, like a robot that can do judo physically with you? I mean, that's like the cool idea, right?

But I think that's like, I think decades away, but I briefly mentioned that I might do some collaboration with the IEJF with my research so I wanted to kind of tell you guys about that how that's going so yeah that's uh and then no I connected you to the guys at IEJF and we're having meetings about it and stuff like that I don't know how far down the line that's went you know but uh yeah let's kind of talk about your research first and how it could help judo and Then you know, maybe we'll come up with some ideas about putting LIDAR in this the foot of that

Yeah, robot. So then when you attack it with the dashi or something it Distance and then can counter some of it crazy. Oh, so yeah, so my research is a That would be crazy, huh? So yeah, so my research is basically the basic idea. What if ChachaPT could watch a movie with you and talk about it? But, and that's like the basic idea, but I think I want to go further and then maybe what if we could ask the ChachaPT could sit next to you and then maybe do help you cook something too. like see what you see and then you know, oh hey, you're doing that wrong, you gotta cut the onion this way or whatever when you make a mistake. So that's kind of like the setup. Yeah. So integrating the large language model, which is

words and binary code and all this stuff, and then integrating visual thing to do you convert it into code and all this stuff and then integrating visual thing to do you convert it into yeah texture something and then it converts into that and interacts that way. Is that the idea? We don't. So my thing is more like let's kind of align video input and the text input together so that they can kind of be processed together instead of some approaches take that where you know you translate the video into like describe the video and then use a description to talk about the movie but that's just a kind of like a you know not a sexy solution right

like you're like you want the robot to actually watch the video and then directly talk about it or like what watch what's happening right now yeah so that's the thing. So I think there are a lot of technical challenges. I mean, you know, we don't have to go into the details about that. But that's why I contacted you to help me, like asked you to help me work with the IGF because judo, you know, has all this dynamic sports in general, has all these dynamic elements. And then you have to be,

it's even harder for the AI model to understand the movies because there's so many dynamic physical movements that happen and then, how can we even develop a model that can actually understand Judo by watching it? That's the whole premise of my thing and then even further okay I'm kind of losing you right now. No, no, I'm listening. Okay, okay and then so let me know yeah because I could like I

tend to like just go into too much technical detail and stuff. But you are partnering with Judo TV and Judo TV has so much of these sports broadcasting videos. You know what they do a good job? They do a good job with denoting when something happens. I know, that's so valuable. I actually don't know how they do that. Do they have people who actually watch all the videos and then mark it? Yeah, I don't know. I really don't know. Maybe when they have the software during the matches that are coming up, the upcoming matches, it kind of integrates with it. And then Ozari scores and then the person who's playing Ozari. Oh, maybe, yeah. It's all integrated. And then it's integrating with the scoreboard because it shows exactly when the thing was done,

when the penalties were given. And that gets locked automatically probably. Probably, yeah, that would be my guess. It would be the most efficient way. And if I was building that product, that's exactly what I would do. I should be on Silicon Valley. Yeah, exactly.

I could probably fake it. That's actually, but you can't even imagine how valuable that is to AI research. All these videos that are annotated properly by people, and then it just happens naturally, right? Because, you know, IGF is really big on this and they're generating data all the time. So my idea was, what if we could have an AI system that could kind of act as a commentator? And we're not there yet. That's actually a lot of technical challenges to be solved to get there. But it's a good testing ground for an AI model that could actually work with you in the real world.

I have an idea. You ready? What's up? This is the new product. Every time, you know, I get pretty hyped at these competitions. Whoa! Blah! You know, like doing this kind of thing. So they're going to take recordings of that and they're going to rate it, you know, one to five. You get what I'm saying? Right? And then whenever a score happens and then the person you just insert one of those. Oh, yeah, that guy excitement like an exclamation generator. Yeah, yeah. I mean that really adds to the, you know, I feel like, you know so So that's a setup so I asked basically through you the IGF like they could provide me with the footage So I can kind of research and then maybe we publish

The results and the data set how soon can I do a research something like that? What do you mean? How soon can you bang that out it depends? I mean I don't know I? Think this type of research you need a lot of time beforehand, like try to formulate the problem, they say. Like, because you have to, you're basically creating a new sport, basically. You have to come up with the right rule set that everyone can buy into. So I had to do a lot of reading and thinking and talking to a lot of people to refine that rule set. And I think I met a good, that took me, I think a couple months.

Of course I wasn't able to focus on it solely, but yeah, I think. Wait a minute, how are you gonna do that if you have a new job? What do you mean? Oh, once I graduate? Oh no, so this is one more, this is what you have to do to graduate.

Yeah, this is my, this will be the last chapter of my dissertation, yeah. Oh, okay, so how long is, when project you have to do to graduate. This will be the last chapter of my dissertation. Okay, so how long is... when are you going to graduate? This May is the plan. So you could bang this project out by May? I think so. I better. Wow, so you gotta work every day. This is a grind. PhD is always a grind, man.

This is why it's... you know that phrase publisher perish. Yeah. Yeah. So it's like it's that it's a grind. It's not I don't think you can do it's not a sustainable model honestly in my opinion and that's why once you become a professor you start taking on other duties because and then let the grad students focus on the actual work. So it's a grind. You rely on this brilliant people who are more brilliant than me to produce this scientific discoveries. But yeah, that's interesting.

But yeah, so this project, I think I'm at the point where I think I've talked to enough people and gotten enough feedback that the rule set of this new sport I'm trying to create is good enough, I think. So tell me about that. So I think I'm focusing more on, so this has to be a little technical, but basically if you think about chatbots, right? It's term-based, right? While the chatbot is generating its answer, like Jadgiputhiradipstik, you can't talk.

And while you type your response, it can talk, right? True. But that's not how it works in the real world. Things come in, the video, the things change, the visually all the time while you're talking, and people might say something while you're talking, like you get interrupted or something like that, right? So currently the models we have are not really able to handle this real time interaction natively. So my idea is to suggest testing ground using judo videos or soccer videos, broadcasting videos, to test this real time interaction because it's kind of perfect for it.

Like things are changing rapidly in front of your eyes and then you have to adjust your response based on it. It's not turn-based. You know, Judo athletes are not gonna wait while you say, oh, you know, Blue Gi is trying to do an osu! Togari. It's gonna like go for osu! Togari, but you have to just say, oh, it might be a fake. So if it was a fake, you have to just say oh it might be a fake so if it was a fake you have to adjust your what you say in real time so i want to set set up this environment where people can i'm gonna suggest i'd suggest a simple approach to maybe address this issue yeah but the goal is that by suggesting uh proposing this type of setup and with the data set

proposing this type of setup and with the data set, people can work on this problem. Interesting. Yeah, like they can propose. Okay, so how does this benefit IJS? So I think so that was like a Mattias that I talked to. I think he's the head of IT infrastructure, I think. Yeah. Yeah, yeah. He's sharp guy. I work with him when I'm... Oh, good. Yeah. We're in the same, We're in the same booth. Ah, okay. Awesome guy.

I had a short meeting with him and I explained what I just described to you. And he's very excited and he was thinking about... So I thought maybe like, you know how not all matches are commentating right now? Because you can... So I think there's no way we could replace all the commentators. So your job, because no, you want the human touch. So I think for the finals and like the important matches, you know, we'll still have like professional commentators like yourself to commentate.

But like if you think about the qualifying rounds where you just can physically there are multiple matches that are going on and you just can commentate everything. Right now we have three maps going on simultaneously and then we decide and usually the rule is whatever championships rounds that's the furthest gets the first dibs on right and then sometimes That's the furthest gets the first dibs. Yeah, right and then sometimes you know, I see an American I'm like, let's go. Yeah. Oh, you're like that's like You can't go there, but he'll let me do it. Oh, yes. I'd say huh? Yeah. Yeah, we're just You know jumping around jumping around, you know, and then you know, you could pick your favorites to raise like oh this guy's Yeah, I'm at one of this this quarters but that's a round of 16 but you know he's Japanese and he's a rising

star I'm sure everyone at home wants to see that guy let me go over there like we can have a little bit of flexibility and stuff like that but a lot of like you know two-thirds of the matches don't yeah exactly commentating love because we're jumping around so I I thought maybe this could... Except at the World Championships because every mat has a... Okay, okay, I see. So in like smaller tournaments or qualifying rounds it'd be kind of great to you know have if we could develop an AI system that can commentate. That's cutting edge, huh? That will take a lot of work I think. I mean that's...

But you're setting the groundwork for something like that to be able to be built on this. Exactly. What I'm saying is this type of problem, I'm trying to convince other scientists that this type of problem is worth investigating. And here's a way to test it. I've collected enough data. Yeah, you gotta be like, there's 50 million judokas around the world that would really appreciate that. Hopefully, yeah. An AI version of Neil Adams.

I mean, that would be so cool. Yeah, just like Neil can... Maybe you can switch between Neil and Shintaro's voice. They say Neil's voice is the voice of judo. I mean, it really is. He's been doing that for so long. Just give everybody a British accent. Yeah, and so there's that and I also I talked to Matthias about this because he was curious about the applications and his suggestion was also very interesting. It's a live

translation of commentary. So right now Junotv supports what? English obviously, Japanese, Georgian, well, French, Russian, major languages, right? Yeah, they do, yeah. Let's say I don't think Korean is supported yet, but imagine a system that runs alongside you. So while you're commenting in English, your voice gets translated into Korean or whatever language of your choice

and then comes out of your voice while you say it like a live. So something like that. I thought that was very cool. I mean that would be very it would be a feat to pull that off. But yeah, so things like that. So I mean that's like the vision. And also like you know this will also has other you know although I'm using sports broadcasting videos, but this has some implications into you know autonomous vehicles or robots and wearable headsets because all these settings have similar concerns they are real-time interactions that can change things around you while you're saying something

so that's the idea interesting, that's awesome and then so you have another meeting lined up with those guys. So I sent to Not yet, but I did send him. Mattias asked for like a little Proposal a page proposal so that he can talk to you know the higher-ups So I sent him that and we'll see what happens I'm hoping that I just I just need access to the videos and the annotations that you talked about and yeah well, that'll really help me.

And yeah, I think judo is a very unique sport because a lot of times people will be focusing on like team sports like soccer in this area. But I think judo offers more, because they're only two athletes and then it's more granular, I think, like the way. Also each move is all annotated I know like people get annoyed when judokas are like oh is there a Hanegoshi is there Uchimata but that the fact that you can precisely

pinpoint what the move is it's very I think valuable yeah yeah for data person yeah yeah they're very data driven yeah yeah hired a lot of data scientists and they use a lot of statistics and math and yeah you know I think visor was a casino yeah you told me it's just he was like a heavily math yeah based casino dude in Romania and you know so he's using that background knowledge to kind of yeah make everything data-driven you know when they took out the last minute you know the women's yeah yeah I remember fight they were like nothing happens in that last minute you know and then everyone's like oh you're gonna cut the women's to four minutes and not the men and they looked at the men's thing

they're like yeah you know not as bad as the women in terms of like nothing happens the last minute but men too they're much less likely to bomb someone in the last minute if you have four minutes to throw someone on in the last minute most of the time it doesn't go that way let's just cut it to four you know so very data-driven you know it's great that they do that the pen it was the pinning it bone was decreased from 30 seconds to 20 because of that, right? Like no one escapes in the last 10 seconds. Yeah, not no one, but it was very little.

You know, statistically significant with the standard deviation. So I'm very excited about this. So that's why research is. And yeah, I just got to do the work and yeah. Speaking of AI, thank you to Hawqueen.ai. If you have business you have churn. Thank you Drew for sponsoring the podcast. If you have a business that produces churn, Hawqueen.ai will use AI to determine, analyzing, what to do with that churn. Customer churn. And then thank you to Jason and Laval thank you very much

HigashiBrand.com thank you judo TV of course with the IJF use discount coach and taro watch all the top level IGF competitions and you know if the IGF is listening I think this would be very very cool to get Peter on board to replace me. Hey I'll say I think that's a you know yeah chest engines are no one can beat chest but we still play chest. It's the same way so I don't think we'll ever want to just get rid of human commentators. You got all those exclamations you had are what people right. Well people want their genuine reaction, you know Yeah, I love all Sheldon's tidbits on the thing because he goes into stories and history

Yeah certain sayings and he's the most interesting dude. You know, I know. Yeah, I love working with that guy You know, it's just kind of like start telling you stories about certain things and you know, sometimes the matches aren't exciting You know, sometimes they kind of that's an art man that's an art yeah and he just fills that time you know and he like tells a story and you know he'll say it as it is too he's like oh not the most exciting match here you know I haven't had any scores all penalties you know it's two minutes into golden score let me tell you something and you're like oh what does he got to say you know so that's yeah that's the human touch I think I don't think you know it can replace yeah I can't really replace that you can replace me whoa whoa that's a reaction

you know if it comes from a robot it doesn't feel genuine you know it's like that's very true I get excited that's why you know so yeah once yeah yeah but yeah and again like judo team I can't thank them enough and you know so yeah once yeah yeah but yeah and again like judo team I can't thank them enough and you know at least like at least you know meeting with me and they're talking to me about this and they are sitting on this they're doing a great job this is a unique unique product it's a big time connector I know yeah it's like you're you're advancing AI yourself. Silicon Valley man. Let's go for it. Next season I'm going on. Is that even a show still going on right now?

I think that it's over. I gotta watch it. That's a funny show. Alright, cool. Alright man. Thank you Peter. Thank you and we'll see you guys in the next episode.