AI Analysis: Ear-Reddening Move

AI Analysis: Ear-Reddening Move 1

Honinbo Shusaku 本因坊秀策 born on 6 June 1829 in Innoshima, Japan is a really famous Go player who played one of the most well known moves in the Go history – the ear-reddening move. I got to hear about him very early when I learned about the game. Nowadays, we have many tools to study Go, so it would be good to see what the AI would consider instead of Shusaku’s most famous move.

Below you can find analysis of Leela Zero 40 blocks, ELF V1, AA Master, Master 8A, Crazy Stone Deep Learning, Leela 11 Deep Learning, 5 block special network.

This is deep analysis, which contains 400k visits till 2.2 million visits.

The reader should be aware that these analyzes are done with komi, for no komi analyze you can check this wonderful article on lifein19x19 forum.

AI Analysis: Ear-Reddening Move 2

This move is called the ear-reddening move. Gennan’s disciples were watching the game and not one of them doubted that Gennan would win. But a doctor, who also had been watching the game, thought that Gennan would lose. When pressed for an answer he replied: I don’t know much about the game, but when Shusaku played the move, Gennan’s ears flushed red. This is a sign that he had been upset. (source)

AI Analysis: Ear-Reddening Move 3

Let’s start from the weakest till the strongest AI. This is 5 blocks special trained and you can find it on Ah Q Go. On over 1,2 million visits, you can see that it prefers to play on top left and take the big move, it does not think that Shusaku move is possible. Black’s winrate with komi is 33,7.

AI Analysis: Ear-Reddening Move 4

This is Crazy Stone Deep Learning and for this position it searched over 2 million playouts and it also does not find Shusaku beautiful move. With winrate of 42,4% for black with komi, it thinks black has to cut at J4, which is below move 126.

AI Analysis: Ear-Reddening Move 5

This is Leela 11 Deep Learning and it searched for about 2,2 million simulations and it thinks that black has to extend, so still no Shusaku move. According to Leela 11, black’s winrate is 50,33 with komi. Very interesting that they think that pretty much without komi, black is quite ahead.

AI Analysis: Ear-Reddening Move 6

More about Pangafu Networks you can find here. This is network entirely trained on human games, 100% human games. Once again with komi, black’s winrate is 32,2%, which is pretty much meaning that black is ahead if there was no komi. This time as you can see, on about 800k visits it considers the cut on the lower side, but it slightly prefers to play on the left. Both shoulder hits are possible. The push on the top left is not that much preferred by this network. It searches for some moves on the center, but not really the Shusaku special move.

AI Analysis: Ear-Reddening Move 7

This is Master 8A and it has 80% human style game. On over 700k visits it thinks like other AI, that the push on the top left is good and the second option is to play shoulder hit. Black’s winrate is 25,5 %, slightly lower than the other analysis. However, without komi, I believe black would still be ahead or the game would be at least even.

AI Analysis: Ear-Reddening Move 8

Here we are with the strong ELF V1 and it’s positional judgement which on about 400k visits, it thinks that black has only 7.1% chance to win. However, this is with komi and as we know ELF drastically changes it’s winrate for every one point. So once we remove komi, we will never know what percentage it would show. It agrees with Leela 11 that black has to extend on the right, interesting. However, still no Shusaku move.

AI Analysis: Ear-Reddening Move 9

That’s Leela Zero 40 Blocks Network 199 and that’s our last hope ! On over 2,2 million visits, it still believes that the top left push is the biggest move, however, as you can see on the diagram on lower left, it plays the Shusaku move after some exchanges on the left ! Indeed it is completely different than playing it first, but to me it was quite exciting to see it anyway being played.

AI Analysis: Ear-Reddening Move 10

This is another analyze with 15 blocks trained on 40 block games and after the move is being played, on over 700k visits it thinks that it is about -4,3%. So it is definitely not that bad move.

This article took a whole night, so I really hope it was helpful to all of you who like this move.

Follow me in social media:

Sharing is caring!

Leave a Reply

Leave a Reply

Your email address will not be published. Required fields are marked *