Is AI ready for its primetime sports production closeup?

Much is being made of the potential of artificial intelligence and machine learning to dramatically scale up live sports production. However, caution is being urged by the very vendors developing AI workflows.

“Sports rights holders and broadcasters hear about AI everywhere but they don’t have a clue how much it will cost them or how much they will use it or what benefit they will get from it,” says Jérôme Wauthoz, Tedial VP Products.

“AI can do a very good job of speech to text making thousands of hours of footage searchable – something not possible a year ago. It’s not caption quality but it is low cost, about 1-2 dollars per hour of footage,” says Sam Bogoch, CEO at video search specialist Axle ai.

“The question is really not what works but where it makes most sense for AI to be applied first and most cost effectively. Many media organisations are simply not geared up for it. They need to get their metadata in order first.”

“AI is a marketable buzzword,” critiques VSN Product Manager, Toni Vilalta. “It’s critical with any new technology not to overhype it and be brutally realistic about what we’re talking about.”

Tedial, which has married AI engines to its MAM software to augment and speed-up live production, takes issue with the claim that A.I is cheap.

“AI only makes sense when it costs less than a human – otherwise customers are better advised to hire a freelancer to do the job,” says Wauthoz.

While the All England Lawn Tennis Association and Fox Sports have used an AI system from IBM for production purposes, Wauthoz still feels these and every other AI for production are proof of concept.

“AI is still at the marketing stage. Being used at major sports events to create a buzz. As it stands, AI is not a sustainable product.”

Digging into this further, Wauthoz suggests that a freelance logger might cost £125 as a daily rate.

“That means your AI system must cost less than that otherwise there’s simply no way it makes sense from a budget point of view,” he says.

While top tier sports like the Premier League are awash with cash and may be prime candidates to pioneer AI adoption, he feels the technology is out of the budget range for less popular sports.

“Take volleyball for example. It will be challenging to monetize even AI logging since the margins on this will be low. You need to train the AI on sufficient and relevant data which could take anywhere from three weeks to three months and you would need to do it each time for each sport after which it should learn [from the data] itself. Even then you have to budget in the processing and hosting costs which means, in the end, it can be expensive.”

This view runs contrary to the prevailing view of the benefit of AI to sports. Automated production systems like that of Pixellot have found a niche in lower tier sports and club training scenarios.

“At the moment, the technology is not ready to replace for primetime sports programming,” says Bogoch. “But, if you took a second-tier sports event and you use AI to kick out highlights to the Web then there may be benefit on the basis that anything being better than nothing.”

A.I is far from perfect and faces a number of general challenges. “The first of these is that organisations need huge amounts of data to train an AI,” outlines Kevin Savina, Director of Product Strategy at Dalet. “The data has to be well documented, which can be expensive and hard to do.

“Secondly, there are still holes in the technology. Sometimes we don’t know why it works or why it does not. Some models are biased, others are not robust. Organisations need to understand that sometimes you should not trust the math – it can be wrong.”

“There are still holes in the technology. Sometimes we don’t know why it works or why it does not” – Kevin Savina

Tedial’s Smartlive uses speech to text to automatically transcribe the commentary on ingest. It also applies logs object and facial recognition to log players, jersey numbers, whether the shot is wide angle or close-up, where slow motion is used and detection of key actions (penalty, foul and so on) then combines the results for assembly of highlights packages at any length desired.

It’s still proof of concept though Tedial says it has several sports rights holders and producers interested in its commercial launch in November.

“The big advantage we have is the Tedial MAM and bpm engine means users can create a lot of different workflows including delivery social to digital platforms which is a main focus of sports rights holders,” says Wauthoz. “The automatic highlights engine can package a storyline in a few clicks, or a draft EDL can be handed over to the production team to add voice over, graphics or any other beauty shots.

“Plus, we can operate this in on prem in the cloud or in hybrid fashion.”

He says Smartlive can be applied to non-sports live programming like The Voice. “You can definitely do it. All I need is the data feed on which to train the A.I.”

Crucially, he says, the AI is an option: “You are not obliged to use it.”

“A machine will never be as creative as a user,” he says. “To really talk about highlights you need to tell a story. AI is a tool to help the production to produce more content and faster but the final touch – the creative and artistic part of the editorial — will always done by a human.”

The price of hosting on AWS and Azure will inevitably reduce. “When the cost of using AWS is really low -– then the industry will really find a way to monetise AI. But so far as 2018 goes I am not convinced there is a great benefit simply because the cost of AI is too expensive.”

The industry is still at the beginning of its journey into AI and likely won’t see more complete glass-to-glass automated solutions in primetime for several years yet.