Go is my favourite board game. I got really into it in my 20s and was dedicated enough to go pretty regularly to an IRL group in a pub a fairly long bus ride away. Go is often posited as a more ancient Chinese chess-like, but really it’s nothing like it at all to play apart from the obvious (2 players, total knowledge, one move at a time, square board). I’ve played thousands of games in my life now and there is nothing that even feels in the same category to play to me.
What makes Go such a unique experience to play is the absolute paucity of rules. For something with such deep strategy, you can almost fit the rules on a postcard: you can place a stone on any intersection. Groups of stones touching across orthogonal lines form groups. Any group of stones that has no empty intersection next to any of its stones is captured and taken off the board. The winner is the person who surrounds the most empty space. Then there’s a rule to stop stalemates through repeating board positions, and that’s about it. Everything, from then on, is an emergent property of these base rules.
What do I mean by emergent? Well, almost every space is a legal move, and the aim of the game is to surround spaces. Given it takes fewer stones to surround territory in the corners of the board, you generally fight there first before moving to the sides and centre. To capture groups of stones you need to play on every intersection surrounding them, which in turn means you can’t play a stone with no empty intersections (unless it’s a capturing move). So to make groups uncapturable, they must have two internal areas, called ‘eyes’ in game lingo. This emergent requirement to have enough space internally to create two eyes absolutely defines the way the game is played. The real joy of Go is how these simple rules and their direct consequences then have all these unexpected side consequences: it feels a bit like trying to grasp on to a part of a Mandelbrot set or something.
Go was always seen as something out of reach for computers to be any good at — there are something like 10^360 legal positions for Go compared to 10^40 for chess. While Deep Blue famously beat Gary Kasparov in 1996, it took a computer program — AlphaGo — another 20 years to beat the equivalent Go world champion Lee Sedol in 2016. This was the first real world application I think I remember seeing of the modern version of AI.
The company behind this, Deepmind, was bankrolled by Google for the purpose and they created a documentary about it. I found this documentary profoundly uncomfortable to watch, the closest to big tech propaganda I’ve seen in a while. An office of entirely white and asian men of about the same age and bodytype all operate out of some huge office with seemingly almost unlimited resources at Google HQ. This context of who funds this, why, who this staff are, why they are doing it, is never discussed. Go pros are reluctant to play it as it’s a bit lose-lose for them: if they win it’s expected and if they lose it’s humiliating. So when it finally gets its first win it’s actually really embarrassing for the ex-pro who volunteered to give it a go. And when it finally beats Lee Sedol you can see almost a sadness in the room, the mixed emotions of this incredibly human game that resisted analysis for so long finally being beaten.
The wildest part though after this is that it’s revealed that no-one on the Deepmind team plays or cares about Go. They just did it because they could. AlphaGo’s playstyle was initially incredibly boring: if you tried half of its moves every game in a club setting I think people would have got pretty pissed off with you, doing the same 3-3 invasion that leads to dry, technical and uninteresting games. I think over time the meta has somewhat recovered but to my mind at least when I was following the pro scene it just made games less interesting to watch.
To me the whole AlphaGo phenomenon has been an incredibly pure exercise in representative technology. Google wanted to prove itself in the field of AI, and completely transformed a millenia-old boardgame that no-one working on it cared about to do so. The benefits for people actually playing the game are swings and roundabouts. The main beneficiaries are, as always, Google shareholders.