VAR: An exercise in Artificial Intelligence?

VAR, Video Assistant Referee, Robocop, montage

When The Guardian (26 Feb) announced Jonathan Liew’s silver award – congratulations, by the way – in the British Sports Journalist Awards, it published a column by him that opened: “I don’t really have a position on VAR.”

That’s an odd attitude to take when your stock-in-trade is to hold an opinion. Very few people, surely, “don’t really have a position on VAR”: those with negligible interest in football, mainly. Until the turn of the year it may have been legitimate to suspend judgement, arguing limply that it was ‘too early to tell’. But the season is now two-thirds over (nine-tenths, from a Liverpool point of view). It’s high time opinion-formers in football decided where they stood on VAR.

The difficulty they face is that VAR is not consistently atrocious. Yes, its decisions are occasionally baffling. But occasionally it comes up with what appears to be if not the right answer then at least a reasonable stab at it. And therein lies a plausible explanation.

Many times this season it has been almost impossible to imagine that the Video Assistant Referee is watching the same incident as the rest of us. Is it possible that in imagining this we’ve stumbled on the truth: that nobody in the famous Stockley Park Incident Room is watching it? But an experimental Artificial Intelligence (AI) system is?

VAR has all the hallmarks of AI:
Mystical faith in technology on the part of the authorities (see also NHS records, smart motorways and, looking nervously right, left and right again, autonomous cars)
Nit-picking Assuming its lines are accurate, as VAR has to, the width of an armpit or an instep is as good as a mile
No understanding of the soul of the game Some goals (eg Teemu Pukki for Norwich (2:10) against Spurs in December 2019) are works of art and should not be ruled out for anything less than the personal involvement of Vladimir Putin in the build-up
Incredibly slow.

In addition, three hitherto puzzling factors can now tentatively be explained:
The reluctance of referees to consult pitch-side monitors. The interface must still be in beta testing. The referee would have no control over the process and would be obliged to wait for VAR to shuffle its replays, lines and angles, pausing occasionally to flash up an eternally gyrating icon
The inability to distinguish between clear and marginal errors by the referee. Computers are much better at black and white than grey, even when the grey might be mistaken for Farrow & Ball strong white
The abandonment of any attempt to police penalty-area grappling at corners. VAR is clearly programmed to regard arms as a legitimate part of the game except when the ball strikes a Bournemouth defender’s shoulder.

If VAR is an AI application, should we expect it to improve? Typically, AI systems use vast quantities of incoming data to build on the original algorithms their creators supply and to fine-tune their own performance. Unfortunately, that can mean their mistakes become more ingrained and alarming, depending on the mentality of those creators. In this case we should assume teams of programmers and referees. It doesn’t augur well, does it?