53 AI Moves Worth Trying In Your Game

53 AI Moves Worth Trying In Your Game 1

Pretty exciting to have this first article about an overview on AI moves finished! I would say it was really refreshing for me to do it, although it took quite some time. I really hope you guys will benefit and try some of these moves in your games. Some of these moves mentioned in the article, you have most likely seen already or even tried if you are fan of AI games. However, there are some people who are not that much into AI Joseki researches, so this is also for them. Even to these who study AI quite deeply, I am sure there will be several suggestions over here which you haven’t seen and tried yet.

I want to thank to Viktor Lin, who let me use GTX 1080 and made the process of this article faster, because on my GTX 1050TI, it would have taken at 4x more time.

15 Blocks network trained on 40 Block games was used and average visits of the positions were 150k.

I hope to be able to build a second article soon. Enjoy reading !

#1

53 AI Moves Worth Trying In Your Game 2

So, next time before extending, think twice, connection is also an option!

#2

53 AI Moves Worth Trying In Your Game 3

That’s very interesting, it happened only once in a professional game database and it is after AG Teaching tool was released, so we can say that this beautiful move is an AI move (the blue move).

#3

53 AI Moves Worth Trying In Your Game 4

That’s a beautiful blue move, the kosumi is AG joseki, however, in some situations like this one, the turn is a big gain and a good idea to consider! Actually, during my time in China I tried it twice, and both times I got 90% after the joseki finished, simply because my opponents didn’t know this special turn. Stanislaw Frejlak 7-dan who showed me this move, learned it the same way that these kids learned it from me.

#4

53 AI Moves Worth Trying In Your Game 5

So it seems like playing on the bottom is too slow, attach !

#5

53 AI Moves Worth Trying In Your Game 6

Once you defend on the lower side, for white it seems also a bit slow to defend, however, 90,8% of human professional games defend on the right side, ahm.

#6

53 AI Moves Worth Trying In Your Game 7

This position I have seen in books and lessons and they always say that capturing the one stone leads to a disaster. You know what? Leela seems to disagree and says ” Not capturing the stone leads to a disaster”

#7

53 AI Moves Worth Trying In Your Game 8

That’s a beautiful move to defend, indeed.

#8

53 AI Moves Worth Trying In Your Game 9

Yes, this cut exist, apparently. Scare your opponent, try it out.

#9

53 AI Moves Worth Trying In Your Game 10

Why do you extend if you can fly like this? It’s a cool possibility indeed.

#10

53 AI Moves Worth Trying In Your Game 11

Here, we have two moves actually. It seems like hane is only possible in special cases, although I still do not quite understand when, but when I learn, I will surely make an article. However, AI prefers keima instead of hane.

#11

53 AI Moves Worth Trying In Your Game 12

Seems like even push and keima is fine.

#12

53 AI Moves Worth Trying In Your Game 13

But once you hane and connect, your opponent shouldn’t connect as well. He should play the beautiful blue move, which most of you already know.

#13

53 AI Moves Worth Trying In Your Game 14

It is wonderful move that I learned in China, however, professionals didn’t try it a lot in actual tournaments yet. I have seen only several games. The idea is to not help your opponent too much. Sneaky way.

#14

53 AI Moves Worth Trying In Your Game 15

AlphaGo move innovation, quite impressive, hard to understand.

#15

53 AI Moves Worth Trying In Your Game 16

Even pushing here is totally fine according to AI, it even thinks it is first choice in many positions.

#16

53 AI Moves Worth Trying In Your Game 17

If black extends in the corner, white pushes, black peeps and we get to AlphaGo joseki, however, Leela seems to have a bit of trouble understanding it, so it thinks it is a minus. Even at 1.2 million visits, it still has troubles, however, once I go over the variations, it thinks it is a plus sequence.

#17

53 AI Moves Worth Trying In Your Game 18

Next time before playing hane from below, rethink that you actually have a better move possible, even though it is quite weird. I used to know it is trickplay, seems like not for Leela.

#18

53 AI Moves Worth Trying In Your Game 19

Simple keima as we have been told to be a normal move is too passive according to AI sensei, seems like attachment is easier to make Sabaki, that’s something we actually know as proverb. “When you have weak stones, attach to make them stronger”

#19

53 AI Moves Worth Trying In Your Game 20

This move was actually one of the biggest changes in my Go view at least. I had really hard time grasping that this kind of peep can actually be good.

#20

53 AI Moves Worth Trying In Your Game 21

However, this keima (blue move) was a move that I loved to play before AI. So, now that AI likes it and it thinks it is first choice, I am really happy about that.

#21

53 AI Moves Worth Trying In Your Game 22

AG showed us that even tenuki at this point is possible, you don’t need to play on the top right immediately. Interesting how natural this kind of move feels now, but when it was played for the first time, felt really awkward.

#22

53 AI Moves Worth Trying In Your Game 23

Seems like Leela is not a big fan of connection on the top, so it pushes.

#23

53 AI Moves Worth Trying In Your Game 24

Yes, extension! Cut leads to a disastrous percentages, well known as one of the “common kobayashi joseki’s” for sure not for AlphaGo and Leela Zero.

#24

53 AI Moves Worth Trying In Your Game 25

AlphaGo enlightened us that we do not need to hane and connect, but we can either push once again or even tenuki..

#25

53 AI Moves Worth Trying In Your Game 26

Haha, of course peep or push, how you dare play hane and a normal joseki that we used to play for years..?

#26

53 AI Moves Worth Trying In Your Game 27

According to AG and Leela Zero, push is simpler and better than hane.

#27

53 AI Moves Worth Trying In Your Game 28

Hah, of course I expected hane for white’s move, probably most of you as well, however, it’s not really a Leela’s taste to simply hane.

#28

53 AI Moves Worth Trying In Your Game 29

Don’t always connect blindly, although this move is a minus in this situation, I have seen it being a plus in other situations, so add it in your list of possibilities. Wonderful analysis of Uberdude here/

#29

53 AI Moves Worth Trying In Your Game 30

Forget about the wedge, play hane or kosumi says AI. The sequences afterwards are a total mystery for me still, however, definitely an exciting suggestions, so I am gonna study them.

#30

53 AI Moves Worth Trying In Your Game 31

That’s better than two space extension according to AG Teaching Tool and it seems like Leela Zero also agrees.

#31

53 AI Moves Worth Trying In Your Game 32

But that’s something I totally disagreed with. However, time passed and I somehow get the idea, what about you all? Would you play this extension?

#32

53 AI Moves Worth Trying In Your Game 33

Don’t connect on second line, fight back instead !

#33

53 AI Moves Worth Trying In Your Game 34

Never thought this attachment can be the first choice of AI, how weird..

#34

53 AI Moves Worth Trying In Your Game 35

That’s AG suggestion, it looks cool and is either developing into a fight or giving black a move on the right side.

#35

53 AI Moves Worth Trying In Your Game 36

Even this move is possible in some positions. I tried it out on a real tournaments, only against 5D+ and I wish I could make a photo of their reactions, they were priceless, try it out !

#36

53 AI Moves Worth Trying In Your Game 37

So, so.. AG says the bump is better.. LZ agrees. The thing is that it might develop into a fight or just take points in the corner and even after that, I am not really sure if I am going to actually take that joseki over the one with the ordinary extension. What about you?

#37

53 AI Moves Worth Trying In Your Game 38

Again an AlphaGo change over here, no hane, but push inside instead, it changed the way we thought about this joseki, although humans have played it before.

#38

53 AI Moves Worth Trying In Your Game 39

It appears on the AG Teaching Tool ( I don’t remember what position exactly), however, LZ doesn’t like it for sure and I still haven’t really found a position where it likes it (over the past months). If anyone finds, please write a comment below and show the position, I am curious and I believe more people as well.

#39

53 AI Moves Worth Trying In Your Game 40

That kosumi over there is shining, I really want to try it out ! In human games, atari or extension up is most popular choice.

#40

53 AI Moves Worth Trying In Your Game 41

That’s something I wrote about in my book Go Studies: A History of Adventure, you can read a detailed explanation there about it. Stanislaw Frejlak 7-dan found it while we were in China for studying Go in Ge Yuhong Go Academy. The idea is that if you want to take solid outside strength, then this move is a good choice (all bots have different percentage and opinion of this move).

#41

53 AI Moves Worth Trying In Your Game 42

The push became popular due to AI, otherwise we used to defend by playing nobi up (or down, depends how you look at it) with the move which has only several visits.

#42

53 AI Moves Worth Trying In Your Game 43

This cut is a totally new joseki, however, when you review with AI, always check the ladder yourself, never relay on the bot.

#43

53 AI Moves Worth Trying In Your Game 44

Such a nice move, isn’t it? I saw it in AG Teaching Tool, however, LZ doesn’t like it a lot, however 0,5% is not much. The idea is to not leave too many aji’s to your opponent, otherwise, of course you can push once again and jump.

#44

53 AI Moves Worth Trying In Your Game 45

AI showed that this is effective way to deal with Low Chinese Fuseki !

#45

53 AI Moves Worth Trying In Your Game 46

That clamp and being pressed down on the second line, probably has something magical about it, but I am probably too much “human” so I don’t really understand it.

#46

53 AI Moves Worth Trying In Your Game 47

AG flexible choice on the other side, no need to add a move to the top hm

#47

53 AI Moves Worth Trying In Your Game 48

If you hane instead of cut, be aware that you might get lucky and your opponent still hanes, however, if your opponent read this article or studied with AI, he will connect or tiger and protect his weakness and build the bigger side, so cut first (create weakness first).

#48

53 AI Moves Worth Trying In Your Game 49

Truly love it when AI likes moves like this one.

#49

53 AI Moves Worth Trying In Your Game 50

Another keima which is quite a good move, not leaving many possibilities to your opponent, no more pushes !

#50

53 AI Moves Worth Trying In Your Game 51

A complex joseki arises and instead of atari and connect on the top, AI prefers to push here and build more influence and strength on the right side. Although this joseki is made by AI, it still finds new moves in his own sequences, so it is pretty interesting.

#51

53 AI Moves Worth Trying In Your Game 52

Haha, time for some action. Leela suggestion is cool, indeed, I like the blue move, however…

#52

53 AI Moves Worth Trying In Your Game 53

AG suggestion is much more intense, I guess that’s why Leela doesn’t like it. However, you can try it in your games and surprise your opponent!

#53

53 AI Moves Worth Trying In Your Game 54

As a big fan of Sanrensei, to see that keima being a blue move is really tough for me to grasp, but wouldn’t mind playing it.

Follow me in social media:

Sharing is caring!

Leave a Reply

3 responses to “53 AI Moves Worth Trying In Your Game”

    • There is embed sgf on the website on many articles, but as you can see these are diagrams with percentage from Leela Zero, so there was no point to embed 53 SGF files this time. The point of the article was anyway to just bring ideas, I hope you liked it !

  1. #38
    q16 d4 q3 d17 d15 e15 e14 e16 b15
    r16 d17 q3 d4 d15 e15 e14 e16 b15
    q16 q3 d16 c4 e4 e5 f5 d5 o3 r6 e2
    q16 d4 q3 d17 d15 q5 p5 e15 e14 p6 p4 e16 b15
    r16 q3 d17 c4 d15 q5 e15 p5 p6 e14 e16 p4 s5
    q16 q4 d16 c4 k16 c14 e4 e5 f5 d5 o3 r6 e2
    most likely every variation in the alphago teaching tool where the joseki occurs

Leave a Reply

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