[–] SwiftLion 0 points 5 points (+5|-0) ago  (edited ago)

So, there are a couple of things that are different between Go and Chess that make this make sense, so let me cover those first. The first thing is that, in Go, you can make move after move after move that, ultimately, does not affect the final score. This is much harder to do in Chess. The second thing is that, in Go, the computational bound of "seeing through" the board is sufficiently huge, that both players have to understand in their minds exactly where the "fights" are on the board, and then consider them in relative isolation. (Any Go players that read this understand how that's an oversimplification, but it's useful.)

Firstly, let's talk about moves that don't help. Go is 'the game of encircling', where only the open spots are counted. If you reach out and put a stone into your opponent's empty scoring spaces, and your opponent needs to remove that stone, then he'll have to place four white stones to capture it--so, you lost one stone against your own score, but it made your opponent lose four stones from his own. So, that's a net win, right? Well, no. Because if you have no open path to creating what a Go player would call an 'immortal structure', or at least a 'living' one, then the opponent actually doesn't even have to answer your move--they can say that it's useless, and if you both agree, then the pieces are considered "dead" and removed at the end of the game as if you'd defeated them. If you don't both agree, then you play until it's obvious who was correct.

This leads naturally to the second part, about how vastly huge a Goban really is. In Chess, there are 64 squares, with an average game consisting of about 40 moves. A Goban has 361 spaces, with an average game (between experts) consisting of about 200 moves. When it's time to make your next move, Go averages about 250 choices per move, but Chess averages around 30. The important thing about these numbers is to realize that the game is truly beyond human comprehension. Your brain simply cannot evaluate all of those choices, because remember, the number of possibilities multiplies after every move--so 40 times 40 times 40 for chess, means that there are about 64,000 combinations available for your next three moves. With Go, however, the 'next three moves' are chosen from a possibility space 250 times 250 times 250 in size, which is 15,625,000 possibilities.

As a consequence of this, playing Go by trying to dominate the board, and completely understand the entirety of your position, is functionally impossible. If you read up about Go, you'll see that nobody discusses "making up" moves or strategies in Go. They are "discovered". Learning the process of playing Go is about cutting down that decision tree through human agreement--because, in Chess, it really is possible to make the "best" moves. In Go, by contrast, the game is kept playable by humans through agreement of its two players. If either of the two players just begins making legitimately random moves, the game will devolve, and even the professional will be unable to determine what the final score will precisely be.

Instead, both players very carefully decide exactly which areas to develop and when. You'll start off by each player sorta "picking" corners, and then they begin building structures there, much like you do at the beginning of an RTS. Then, they'll identify each other's structures, and decide how to attack them. Unlike Chess, an attack in Go only succeeds once you have ensured that there is no possible way for a group to 'un-encircle' itself--a process which can take dozens of moves. In order to determine whether this will be the case or not, you get "tunnel vision" onto the area you're going to be working on, and you think about a few dozen moves' worth of progress in that singular spot, and your attention to the rest of the board fades more with each spot farther from the center of the group.

After all this exposition, we're finally on the explanation of move 37 and its significance!

There is a concept, in Go, known as a 'God's Hand' move. This is when you examine the board, and you aren't playing joseki, you aren't thinking back to what the masters did or what your teacher did or what's correct. You look down at YOUR board, in ITS own entirety, and then you reach out and lay down a stone in a place that follows no pre-existing pattern. This is considered a discovery. If you dedicate your life to playing Go, you may make a single God's Hand move, and if so you would be very proud and consider yourself a notably accomplished person. In the anime 'Hikaru no Go', the whole purpose of the Go-playing ghost that's come back is that he died without making a God's Hand move, and can't be satisfied with the afterlife as a consequence.

When 'Deep Thought' placed move 37, it did not follow any existing moves or games. That particular choice of move had not ever happened before. In that moment, it is not incorrect to say that 'Deep Thought' created a truly original idea. Deep Thought played a move that no human had ever conceived, placed or evaluated in any living game.

The move is called 'beautiful' because, as humans looking at a Goban, we are aware that the game is beyond us. That's why it's worth staring down into, and dedicating one's life to: no matter how closely you look, what you see and perceive with your mind is a rudimentary mockery of the delicate balance of hundreds of stones, each influencing one another and the spaces around them. Playing the game is like walking into a dark cave with the combined maps of thousands of years' worth of explorers, taking one step and then another, inferring as best you can what the whole cave must be like from the surface of the wall to your left and the ledge of the cliff under your right foot.

So, to see a machine look deeper into that abyss than you, the world champion of Go are capable of . . . that is a truly humbling experience for us as a species.


[–] SwiftLion 0 points 2 points (+2|-0) ago 

As a curios sidenote, Deep Thought was also aware that it was making an unusual move. One thing they offer for each move is a rating of how likely an arbitrary human player might be to make the same move. For most of the game, its moves were relatively likely for a human to make, but when asked to rate how likely a human might be to play move 37, it rated the likelihood at 1 in 10,000.


[–] n1L [S] 0 points 0 points (+0|-0) ago 

Thanks. That really made sense and makes me want to try learning go again. Really great answer.


[–] Chiefpacman 0 points 2 points (+2|-0) ago 

Perhaps you coild specify the move in question? A google search actually yields this submission first, followed by this interesting article on the topic- https://gogameguru.com/alphago-defeats-lee-sedol-game-1/

After reading it I'm still wondering what is seen as the unexspected move.


[–] Fibbideh 0 points 3 points (+3|-0) ago 

I think he's referring to move 37 in game 2. The announcers and everyone in the room thought it was a mistake.


[–] n1L [S] 0 points 1 points (+1|-0) ago 

You're right. That's the one I'm talking about. Thanks.