AI and New Fuseki

[Ian Butler]

I was just thinking about something. Imagine that either the New Fuseki movement got here a little later, or the AI and computer development was several decades earlier, and we humans were still playing old fuseki openings. Imagine that AlphaGo had come along then, and suddenly started playing these 4-4 openings.

It'd have been an even greater revolution than now. People would be extra confused, because they'd be like: why do they even open 4-4 if they're going to jump into that 3-3 immediately? It doesn't make any sense!

Just a fun thought

And maybe if AI development came a decade later, we'd have figured things out ourselves more, and the early 3-3 would be "old news" already.

We'll never know

[sorin]

I wonder if any of the *Zero bots go through any "New Fuseki" stage during their self-play evolution.

[Uberdude]

I wonder if any of the *Zero bots go through any "New Fuseki" stage during their self-play evolution.

Well, the random games at the beginning look a bit like the Sixteen Soldiers

[sorin]

I wonder if any of the *Zero bots go through any "New Fuseki" stage during their self-play evolution.

Well, the random games at the beginning look a bit like the Sixteen Soldiers

Only at first sight

[dangomango]

I was just thinking about something. Imagine that either the New Fuseki movement got here a little later, or the AI and computer development was several decades earlier, and we humans were still playing old fuseki openings. Imagine that AlphaGo had come along then, and suddenly started playing these 4-4 openings.

It'd have been an even greater revolution than now. People would be extra confused, because they'd be like: why do they even open 4-4 if they're going to jump into that 3-3 immediately? It doesn't make any sense!

Just a fun thought

And maybe if AI development came a decade later, we'd have figured things out ourselves more, and the early 3-3 would be "old news" already.

We'll never know

well ancient chinese fuseki is diagonal 4-4

[sorin]

Speaking of New Fuski: one of the top recent bots on CGS is playing the "black hole fuseki" (playing 7x5 and 5x7 with first 4 moves, forming a square), with both black and white, with great success: http://www.yss-aya.com/cgos/19x19/cross/black_hole.html

Also, it is following up with unusual moves in the center (which seem consistent with the fuseki).

[Uberdude]

I presume that's some strong bot that's been modified with hardcoded logic rather than discovering the black hole for itself. This was a nice game with sacrifices and then the attack on Centre group leading to corner profit. Amusing mini black hole in the centre too later, Hawking radiation in action!

[sorin]

I presume that's some strong bot that's been modified with hardcoded logic rather than discovering the black hole for itself.

Oh, yes, I'm 99% sure that it's not something that emerged from pure training, but rather a behavior forced by the programmer (probably in early stages of training, rather than just a vanilla LZ network with hard-coded opening, given that it has good results against LZ on CGS). I wonder what does it do if the opponent takes one of the four initial points.