AI at any cost: Mimir on why artificial intelligence needs a buzz cut
By André Torsvik, Mimir EVP operations.
There’s no bones about it; calling 2024 a major sports year is an understatement. The Euros, the Copa America, the Summer Games, the cricket T20, not to forget all the usual suspects like the Grand Slams, the Tour de France, the Super Bowl, and Wimbledon, and all of the above except the Superbowl happened in June, July, and August.
You’d think no one was crazy enough to try to implement new technologies in such a busy year. You’d be wrong. Sports broadcasters were no different in a year where everyone looked at AI. Moving on from a mostly hype-cycle-manic 2023, we saw some excellent real-world implementations in 2024, albeit not always well-thought-through ones. The (mostly good) examples are numerous: ESPN using AI for real time analysis, pioneers F1 using predictions for a few years already, NFL creating real time graphics overlays with AI, and NBA implementing it in the draft process.
This is in addition to the prodigious use of transcription, translation, automated highlight editing, and more, leveraging technologies that have all been around for a long time and have little to nothing to do with the beautiful bauble of generative AI.
A big sports year always brings significant investment and allows broadcasters’ production crews to renew, to experiment, and, frankly, to try out some stuff they would not be able to afford otherwise. And while I can’t see inside the accounts of most broadcasters, there is a clear indication that there was some overboard-going. Compute costs money, and so does experimenting your way to functional implementations. But after a glorious spending spree comes a cold morning, and I think 2025 will be just that. The buck, as it were, will stop here.
Cost efficient AI
2025, on the other hand, is not a huge sports year. And the global economy is challenging. And broadcasters, by and large, are in a position where they need to look hard at both their top and their bottom lines. And AI and cloud have gotten reputations – somewhat unfairly – for being a dangerous signal in spending. But they don’t have to be; when used right, these technologies are what can save you money rather than burn it, and that is where the tech bros of our little world need to make their arguments and make them well if they are to have any hope of making it to the other side of the budgeting process with any toys left in their prams.
AI usage needs to be taken in hand and given a good talking-to. It needs structure, guardrails, and guidelines. It needs to get a buzz cut and a real job.
Because now we’ve had some real experimentation with AI in all its forms, one thing is more transparent than everything else; there are some benefits too good to give up. I’m not necessarily talking about auto-generated graphics, generated audiences for empty stands (remember that one?), stance analysis, or even the ability to fill in that missing shot you’re short for the highlights reel. I am talking about way more basic stuff; stuff that will let you be both efficient and innovative at the same time (yes, with AI).
The ability to crunch masses of data, and fast. For many, this is where the rubber hits the road (certainly in motorsports and others using predictive analytics). And to leverage masses of data efficiently and on-demand, you need masses of scalable compute. You certainly don’t want the massive CAPEX investment of getting this inhouse – so you can only get it from a cloud platform.
The possibility to make your assets more accessible and useable. I am talking about metadata and how “simple” AI applications such as transcription, object labelling, and face detection and recognition can help generate massive amounts of this – automatically and (if done right) at a reasonable price. The extrapolation here is of course, searchability, discoverability, reusability. Your assets will no longer be single-fire cannons of very expensive content but repeated, double, triple or multidip troughs of value that can be accessed and used repeatedly.
The growing potential in automated workflows involving AI. Once you have a rich metadata set – much richer than manual logging could ever get you – the road does not just have rubber on it; it’s been paved and prepped as well. By applying analysis and a little bit of automation, it is not a stretch at all to make LLMs describe the scenes in your videos, to use metadata to tag all the goals in a match, and then a little bit more allows you to auto-create packages based on this. And here’s the kicker; with some clever setup, you can create multiple ones simultaneously for multiple platforms, leveraging your content better and more efficiently.
So, what does it take to make something like this work? Understanding the technology, applying the elbow grease of setting up automated workflows, and heeding two pieces of advice:
Choose cloud-native tools. There is no other way to properly leverage cloud compute, access AI platforms and vendors, and connect with emerging technologies. Doing all this with legacy tech is just too hard or even impossible and rigidizes your tech stack. And you will not have the time, bandwidth or even the patience to deal with once-in-a-blue-moon updates when the technology you are trying to leverage is changing at 100 miles an hour. The only way to counteract this is to use tech from vendors who have embraced continuous deployment and are willing to help you keep up by deploying new builds not yearly, quarterly or monthly but weekly or even daily.
Avoid AI vendor lock-in. When you choose the AI-enabling cloud-native technologies to build your tech stack, you must ensure they don’t tie you into single AI vendors. There are multiple reasons for this, but here are two: with lock-in, you also have cost-lock-in. And while you may know what their services cost right now, you cannot predict their next offering. AI is becoming a commodity, and you should treat it as such.
Also, you cannot know today who will do what best in the next year or even six months. So why would you tie yourself to a mast that might go down? AI is still the future. But in 2025, we need to answer whether it will be used, but how it will be used prudently and efficiently.