This article was not produced by an AI. Questionable that it was even produced by an I.
The Jetsons have a lot to answer for.
Of all the space-related culture that came out of the 1960s, the everyday life regular family Jetsons, who just happen to have a flying car and live in weird little towers in the sky, stuck. People say "where's my flying car" when they're dissatisfied by some new tech in large part because of The Jetsons and it's idyllic but boringly hopeful vision of the future.
You've probably seen some screenshots or articles about OpenAI's ChatGPT by now. You may, although this is a little less likely, have seen something about FIFA's Stadium Experience augmented reality feature of their app. It feels like we're on the doorstep of The Future, and there are a lot more companies in and around football a mere stone's throw away from the porch.
Get Goalside will aim to have some fuller pieces on these in due course, but for now let's get something out quick and easy on what some of these things could mean for the near future of football analytics. Call it a beta version, it's what tech would do.
For simplicity and brevity, by 'AI' I'm going to stick to things that OpenAI list as examples you can use their tool for. There are too many to list them all but they include: Q&A, grammar correction, natural language (i.e. human language) to code (Python, R, SQL), summarising text, simplifying text.
If we were being cruel to ChatGPT, the thing that's caused the recent hubbub, we might say that it's a very successful example of Artificial Inintelligence. The Verge went for the term "fluent bullshit". It's read a lot of books and 'speaks well', but it's not too hard to trip it up (which might be reassuring?).
In my post for Get Goalside paid supporters [join for £2 or £6 a month] I managed to, without trying, get it to stumble into making up a citation. When asked to do some maths which required some interpretation, it gave an answer that was the exact worrying mix of surface-level plausible and badly incorrect.
Analytics staff who have to explain concepts to all of their colleagues might have been hoping that an AI chatbot could be used as a kind of FAQ machine. But alas, it'd probably do a bad job; or, worse, a good enough job that its mistakes go undetected.
The bauble of OpenAI that seems more immediately usable would be the set of 'natural language to code' features. Football is a long way from the large analytics departments of baseball, for example, and in many clubs a head of analytics department might be the department. At the very least, analytics staff often have to do their own data engineering, which is a little like asking your head chef to wire up all the cookers. Not that it's beneath them, they're just liable to get themselves killed.
There's been a lot of (justifiable) talk about the threat of AI 'art' to the creative economy, but I think the balance is different when it comes to coding. The old phrase goes 'writing is re-writing': nowhere is that less true than the cheap artistic work at the bottom of the corporate pile that might sustain a lot of artists, but nowhere is it more true than writing production-quality code. I don't know whether I'd trust an unedited webapp built from prompts by OpenAI, but it might save a few hours getting the first draft done.
If the problem with AI chatbots is that they're too smooth for their own good, the problem with VR and AR is the same problem as ever, which is that they're clunky as hell.
As intriguing as FIFA's AR Stadium Experience looks, I'm not hugely sure why fans would want to view stats projected onto the pitch via their phone screen. On the VR side, there are a few systems which aim at tactical analysis and/or reaction training, which have their uses but still sorta seem like they're finding their legs.
I mean that literally too. A favourite of mine (which, to be transparent, is also one that I've tried out personally) is from a company called Rezzil, for the pure fact that they've always hooked up users' feet rather than relying on a headset and those hand controller things. If I may be so bold, involving peoples' legs not only seems pretty useful in football, but absolutely essential.
Both AR and VR feel, to me at a distance, like they're just waiting for augmented reality glasses to get significantly better than they are now. Some current tech lets people see stats and other pop-ups, but it's a little way off Tony Stark's glasses in Marvel movies. I imagine that something like those - which can overlay more complex shapes than rectangles over moving objects - would liven up positioning and set-piece sessions considerably: project tactical pitch lines, opposition mannequins, or assigned run routes onto the grass through players' glasses as they stand on the training pitch.
Off-field, I'm kind of surprised that I didn't see a VR-plus-NFT ticketing scheme during the recent NFT bubble (if you have seen one, please let me know). In the half-year window when every footballer and their cartoon monkey were promoting some kind of collection it just seemed like this was an obvious step to come.
Technically, separating machine learning out might not make sense, but it's different enough for me to justify it. It's a broad term that would technically include expected goals models, and is a step in the AI process, but machine learning itself doesn't necessarily aspire to pass the Turing test. The machines just keep on a-learning, and two things in the past couple of weeks were a great demonstration of where and what.
The first was at Training Ground Guru's Big Data webinar in Jonny Whitmore's talk showing some Stats Perform metrics. The company have demonstrated the underlying models before - expected pass completion, expected threat, expected pass target - but Whitmore broke them down into some really interesting statistics to look (principally) at some central midfielders.
Things like: how often did they make 'safe' passes, how often did they optimise likely completion with threat added, how often were they the target of teammates passes when they were an open option. (As with many things, Messi figured highly on that last one).
While expected goals was both a useful stat in itself and a building block with which to create others, this talk demonstrated how the same will be true of newer, tracking data-reliant models.
The second thing comes from Metrica Sports, who shared news and a dataset from the automatic tagging feature in their software. Their product is a video analysis tool, and this option tags up some basic datapoints when users upload game footage - no outside data source needed. I can't speak for the accuracy or consistency, but the demo example shows how a user could load in a video and let the tool take them to all of the moments when a team enters the final third from the left flank with multiple players in support.
It might not happen for a few years, but it's now not that hard to imagine a world where a lone data employee at a medium-to-small club uses software like the above to get immediate data from video, maybe for scouting purposes outside of their existing data coverage, and an AI bot to write the first draft of a database and webapp to store and present it. Or the staff at a larger club with multiple staff members use it to free themselves up for other tasks.
The 'Will AI kill us all' section
There are understandable worries about what might happen if some of these technologies are simply let loose. Will some precarious jobs just vanish? What happens to the first rung(s) on the ladder if the work usually done there can be done by AI instead?
It's worth saying that, even before talking about their impact, merely their creation is an ethics class in action. Models trained on as much data that can be found on the internet is kind of like the hypothetical we all waved away finally becoming reality: 'You know if you put all your stuff on the internet, it's there for everyone and can be used by anyone'. Lensa is the latest big focus for this, DALL-E has received scrutiny too. Keep an eye out for DeepMind's script-writing Dramatron to make headlines at some point.
If "hoovering up artists' output and creating products which can put them out of an income" isn't enough, there are other serious issues. Lensa has been reported to whiten the skin of people of colour and produce sexualised images of users, even when uploading childhood photos. Dramatron-produced scripts (The Decoder relays) featured some sexism. ChatGPT has some pretty bad biases that can be drawn out too. Although OpenAI have clearly tried to put in some guardrails to their chatbot, users found ways around them incredibly quickly.
This stuff, as far as I can remember from seeing re-runs, never happened in The Jetsons (although we never saw how they tested those flying cars). But whatever skeletons might lie in that society's closet, more than anything the Jetsons family made the future seem pleasantly mundane; they have a lot to answer for for that.
Get Goalside would love to hear from you if you have any thoughts or clarifications. Get in touch at firstname.lastname@example.org