Are you a non-technical executive or entrepreneur who wants to be more AI- and tech-fluent? See our program.
This is The Age of AI Series, where we talk to the foremost entrepreneurs and innovators around the planet using ML to transform industries. (Join our special mailing list!)
In today’s episode, we talk about how ML is transforming sports broadcasting!
When you watch a live football game, the feed is usually coming from dozens of cameras all over the place, and being produced by a TV crew of hundreds of people. There’s a lot of real-time decision-making about which camera’s feed the viewer should be looking at at any given time.
For example, when someone scores a goal, takes a corner kick, or clashes with another player, the production team has to quickly identify that happening, cut out a replay section for the viewers, and also choose which of the camera feeds you should view it from. This takes time and manual effort — and must be done in real-time.
Turns out, we can use ML to detect the most important events in the game from ALL camera feeds, and make very good video-cutting suggestions to the production team within just a few seconds. Not only does it save effort, it also improves the viewer’s experience.
My guest is Anri Kivimäki, the CEO of AISpotter, which provides video analysis software to sports broadcasters and some other industries. Their tech isn’t limited to sports broadcasting; it can be used for countless other live video use cases all around us.
Another interesting thing is that they weren’t always in this business; they made a strong pivot during the Covid pandemic.
Here’s what we discuss:
- 1:50 — Different customer segments for AISpotter
- 4:00 — The problem of sports broadcasters
- 8:12 — What’s does AISpotter’s system do; what are the inputs and outputs?
- 15:24 — Inside a sports broadcasting production room
- 18:45 — Finding the most “cinematic” shots, and other things
- 22:20 — How the product works differently for sports coaches
- 25:30 — How the models are trained, and the importance of quality data
- 34:26 — The story and evolution of AISpotter as a company
- 36:38 — AISpotter for the financial sector: security and surveillance use cases
- 40:24 — How the system fights bias in training data
- 42:38 — Why pivoting an AI company is different from typical software startups
- 46:33 — Dealing with non-technical customers
- 48:41 — Special case study: video analytics for horse racing!!!
Aman’s 2-Minute Summary and Key Takeaways
The core system they’ve developed is quite flexible, and its main function is to detect events of interest in a camera feed, which varies by nature. For football, it could be a goal or a corner kick. For a bank’s video surveillance, it could be the drawing of a weapon or suspicious object etc.
To do this effectively, they train a unique machine learning model for each specific event they want to detect (as opposed to training one model for all of them), and in a live stream, each camera feed gets processed by every model in real-time. The detections are then fed to the production team on a simple user interface, so they can quickly make decisions about cutting replays or switching the camera feed that the viewers are currently seeing.
This modular system architecture also makes it easier for them to train models without too much data. This is a simple but computationally heavy system (passing dozens of high-res footage through deep learning models in real time), and AISpotter’s tech is able to get results on the producer’s team in 5 seconds, which they claim is faster than all their competitors.
While talking about AI-powered video surveillance in banks and financial institutions, we touched on the sensitive topic of bias in the training data: many such systems tend to disproportionately mislabel people from certain racial or cultural backgrounds as doing something “suspicious.” Anri’s accepts that it is a genuine concern, but they haven’t had any customers bring it up yet. (We didn’t probe this further, but it reminds us that fixing bias isn’t a one-sided problem.)
What I like about AISpotter is how simple and beautiful the use cases they focus on are. They’re a great example of not getting lost in your own geekiness and keeping the business value proposition clearly in mind.
The company originally spun out of an academic research project (as seems to be the case for a lot of Nordic AI companies), focused on helping coaches improve training. So instead of just detecting events, they were analyzing players’ body movements and form as they played, along with other metrics like heart rate etc. During Covid, they pivoted to broadcasting because people weren’t training in groups anymore.
We also discussed how pivoting an AI company is harder than the typical software company, because ML projects are like mini R&D efforts. It takes a while to choose and train the right models, and even more time to get to a level where you have competitive advantage. I’m impressed that Anri’s team was able to persist through Covid with limited resources.
AISpotter is now expanding to the USA’s huge sports entertainment market. I should also mention that Anri is fundraising for the next year, for any investors interested!
(Ethics Policy: These opinions are 100% my own as an independent observer and educator. I don’t own stock in guests’ companies or their competitors, nor do I get paid by them in any form for any reason at the time of publishing, unless specifically stated. Episodes are also not intended to be an automatic endorsement of any company or its products and services.)
