In the latest episode of his fascinating series on AI go theory, Shibano Toramaru examines the following position.

Click Here To Show Diagram Code

[go]$$W
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . c 2 . . |
$$ | . . . . . . . . . . . . . . . d e . . |
$$ | . . . . . . . . . . . . . . . f g . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . O . . . . . , . . . b a , . . . |
$$ | . . . . . . . . . . . . . . . O . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

He observes that when AI first took over the go world, the usual Black answer to White 1 was (unlike human pros) A or B. The human pros then followed suit, of course. and the various pincers such as Black 2 disappeared into the shadows. But Shibano also notes that recently Black 2 and the other pincers C to G are being played again.

Two questions arise. The first is: why did the pincers die out? The second is, of course, why are they being played again?

Those who are studying go with AI might like to have a go here at how players like Shibano answer these questions.

Click Here To Show Diagram Code

[go]$$W
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . c 2 . . |
$$ | . . . . . . . . . . . . . . . d e . . |
$$ | . . . . . . . . . . . . . . . f g . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . O . . . . . , . . . b a , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

AI, with sufficient playouts, does not play the early corner enclosure but rather invades at either 3-3.

In AI, pincers did not die out but became infrequent.

Early pincers, such as in the position, AI rejected because they lose about 0.1 (near), 0.2 (far) or 0.3 (very far) estimated points according to AI.

Played again by whom? AI or pros?

AI: I do not constantly install recent nets. Rather I prefer to run a familiar net deeply. AI rejects such pincers. Maybe some recent nets have changed their mind?

Pros: fashion. Playing one move in the game that loses 0.1 or 0.2 is immaterial, especially if the unprepared opponent is caught by surprise. Some of the modern pincer josekis are taisha to the third power - countless chances to make mistakes. Some amateurs like to trick their opponents so play such pincers. Maybe some pros try alike?

The very far pincers 2 and C are mistakes, AI judges clearly.

If we shall trust Shibano, what are his minimal numbers of playouts for second move candidates, on each turn of each variation? I do not trust anything below 100,000 but, if necessary, go beyond 5,000,000 for at least each of the first and second move candidates, on each turn of each variation, preferably in several whole board positions. If he does not state any numbers of playouts, he acts as a magician.

You want us to try by ourselves. So... Just a wild guess, to give some food to the discussion, although I am pretty sure I am missing some big picture AND many subtle nuances here :

1. The pincers may have become rarer because the keima-push-tenuki sequence, which was previously seen as circumstancial because of the resulting solid black territory, has been reevaluated as an effective forcing sequence, given the way the AI used it.

Click Here To Show Diagram Code

[go]$$W
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 2 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . O . . . . . , . . . 5 3 , . . . |
$$ | . . . . . . . . . . . . 6 . 4 X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

2. They may be popular again because of the new tendency for Black to refuse the aforementioned sequence (going from "I'll take the cash, thank you" to "No, you won't have me being forced on the third line!") and launch some complicated (but heavily studied) lines after he pushes and cut, some of these lines conducting to interesting, if chaotic, results ?...

Click Here To Show Diagram Code

[go]$$W
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 2 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . 6 5 1 . . |
$$ | . . . O . . . . . , . . . . 3 4 . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

Chess players have had access to large databases since the 1980s. Even without AI, it's clear from the stats that some openings are more effective than others. Basically, if you play a king's pawn opening, or if you respond to your opponent's king's pawn with anything other than the Sicilian defence, you're clearly making a mistake. The evidence has been out there for decades.

Yet people have kept on playing diverse openings, and chess openings now are just as varied as they've ever been.

Why? Because style -- getting into a type of position that suits your strengths, or that avoids your opponent's strengths -- is sometimes more important than accuracy in a game between two fallible humans.

Is it possible that go players went through a phase of reading too much into minute differences in AI evaluation, and are now starting to recover?

My general feeling is the pincers fell into disuse because the AIs didn't play them, and playing AI moves was a fad. Professional play seems to be a bit faddish as well. However, when you look with an AI, the pincers don't come off too badly. They're certainly in any reasonable list of candidate moves.

Without any further insight I wonder whether humans are going back to the pincer because the "best" move is only best if you play like an AI. Much like the moves suggested by a scrabble playing computer might only be best if you have perfect word knowledge, I am wondering whether human pros are finding the game a little difficult. Or perhaps they're now more comfortable with the fighting that comes after the pincers. I can only speculate, but I will be interested to read Shibano's thoughts.

On another note, October of this year will mark the 10 year anniversary of AlphaGo's win vs Fan Hui; the decade anniversary of Lee Sedol's match will be next March. Anecdotally it takes about ten years for a player to become a pro, which means that we are about to mint the first generation of pros who have only ever known AI. I'm looking forward to seeing the insights of these "AI-native" go players.

This argument has always struck me as weird, if not fallacious. Every player, recreational or professional, will partly play moves they "understand", partly go for some adventure and in between play moves they might not understand but have seen from pros or AI. You don't know in advance whether it will work out like you intend, because Go isn't solved. If both know where a situation will end, it's tenuki time. Otherwise you are on somewhat unfamiliar ground, where your inspiration will come from all kinds of places, including prior AI analysis.

On another note, AI doesn't (only) play narrow paths towards victory that are impossible to emulate, but (mostly) moves that increase the likelihood towards victory. It's more probabilistic than deterministic. So, if you know such moves, why would you deliberately not go for more probability to victory. Yes, you will at some point fail to play like AI would. But that still doesn't mean you should have deliberately not played like AI when you still knew what that meant.

Certainly a reasonable argument, but I'm not sure it jives with what happens in real life. Players tend to play to their strengths. Players good at fighting often play an opening that tends to lead to fights. Territory players play territory openings. I think it's plausible that perhaps pros have looked at the types of games that follow the AI move and then looked at the games following a pincer and decided that--for a fraction of a percent drop--they are comfortable with the games following a pincer. It's not like they'll deliberately play 'suboptimal' moves after that, so it's a one-time drop.

I can give a single hypothetical scenario (based on absolutely no research): AIs have very good positional judgement and we've seen them use large sacrifices to great effect. Suppose a kosumi tends to lead to games where such sacrifices are common, and a player figures that they'll screw up these kinds of positions 1 time in 100. They'd be better to take a 0.5% hit in winner percentage by playing a pincer if they're treading more familiar ground. It would let them defer the game's critical moments until later.

The position John shows is common, but the AI blue move hasn't suddenly changed. Why would pros start playing what the AI thinks is a suboptimal move? I can think of a few reasons
[list=]
[*] Irrelevant--the difference between the blue move and a pincer is so tiny as to be irrelevant
[*] Humans think the AI evaluation is wrong
[*] Boredom--the kosumi and knight's move are getting stale and played out
[*] Research--players are looking to catch an opponent off guard with a new move in a pincer
[*] Comfort--the types of games following a pincer are easier to play
[/list]

Maybe someone else can add to the list?

What goes up must come down!

Soba Go has been ascending. When this kind of game no longer gives the top players the desired results they will play a different kind of game. Everyone will follow their lead when the top players are successful with a more original approach.

A similar question is why approach move is not played often when there is an empty corner. Many top players did this all the time in the past. Openings like the following are obviously fine and have been played in the Ing cup final. Maybe one should instead of 'Why not?' ask 'Why?' when the best players play something.

Click Here To Show Diagram Code

[go]$$B
$$ +--------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . , . . . . . , . . . . . 1 . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . , . . . . . , . . . . . , . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . 2 . . . . . , . . . . . , . . .|
$$ | . . . . . 3 . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ | . . . . . . . . . . . . . . . . . . .|
$$ +--------------------------------------+[/go]

Some background points first.

No discussions by pros that I've seen waffle on about decimal points. They are trying to see the big picture. The big picture varies greatly according to whether you have near-impeccable technique, as most pros do, or you are an amateur who can't tell his hane from his elbow. Shibano appears to be trying to help the latter group.

Underlying this, I think, is a realisation that even pros are not normally playing against AI. They are playing against other humans with foibles who will have a tendency to make mistakes just as much as they do. As amateurs, we can make that same assumption big time. The loss of a game is nearly always dependent on who made the last mistake, and that mistake is more likely to be because of psychological or environmental factors, such as even what time you went to bed last night.

Also underlying discussions of AI for amateurs by pros is language, not numbers. Here is where Shibano seems to score highly. The present case is one in point.

Recall that josekis, at least in those joseki dictionaries that amateurs used to swear by, were traditionally evaluated in static terms: profit versus thickness, overconcentration, good shape, aji and so on. It seems that this approach has to be thrown out now. It is not the first time we have been in this position. New Fuseki had virtually the same sort of bling effect as AI, and even most pros were at a loss in trying to come to terms with it. One approach was in a book, Integration of Joseki and Fuseki, by Kitani. It was largely ignored, perhaps because he lacked the skill to find the right words for amateurs.

We are now back in that position, except that Shibano seems to have, in my judgement, more aptitude for expressing new ideas.

Obviously, you have to read his whole work to get the true value of those ideas, but the position shown here gives a flavour of his approach, I think.

Pincer moves such as Black 1 below were traditonally seen as worth playing because they combined a pincer with an extension, yes, but also because they constituted an attack, and so kept the initiative.

Click Here To Show Diagram Code

[go]$$B
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

The expectation was that White had to respond while Black went gaily on attacking, with 4 as below.

Click Here To Show Diagram Code

[go]$$W
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . 5 . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . O . . . . . , . . a 3 1 , . . . |
$$ | . . . . . . . . . . . . 4 . 2 X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

AI has shot down that notion by playing moves such as White 5. The point appears to be that, if Black continues attacking the White group below, White can cope (sabaki) with moves such as 'a'. The end result appears to be evaluated by AI in a way that Shibano expresses as 'an attack that has no effect' (攻めが利かない). This is a dynamic evaluation rather than the usual static evaluations of josekis. This kind of evaluation is also expressed (in other contexts) by criteria such as 'how might play develop after this?' In other words, evaluating the subsequent play is more important than evaluating what has gone before.

Obviously, it is not literally true to say that the attack above has no effect. It really just means that the effect is nugatory compared to the effect of the tenuki move (5 here). We have a fuseki proverb that tenuki is recommended when the local move is worth X points or less. Various numbers have been suggested for X, but 15 seems a popular one. In any event, AI seems to be acting on some similar principle. Maybe we need to downgrade X. (Rob van Zeijst's QARTS can also be taken into consideration.)

However, a new approach to Black 4 above was conceived some years ago and that was to try Black 1 in the diagram below.

Click Here To Show Diagram Code

[go]$$B
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . O . . . . . , . . . O O , . . . |
$$ | . . . . . . . . . . . a . . X X . . . |
$$ | . . . . . . . . . . . . 1 . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

White no longer has free forcing moves against Black ('a' leads to a cut and an unfavourable fight, apparently), and so he is now judged (by some, at least) to be prone to an attack that does have an effect. One example given is where White is driven upwards, allowing Black to enclose on the right side while continuing an attack. Even without that, it becomes hard for White to invade on the right side.

The latest thinking appears to be that, if pincered, White will avoid playing as in the previous diagram, but will tenuki even earlier, as in the next diagram.

Click Here To Show Diagram Code

[go]$$W
$$ -----------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . X , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . a O . . |
$$ | . . . O . . . . . , . . . b . , 2 . . |
$$ | . . . . . 1 . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ----------------------------------------|[/go]

We can expect Black 2, with Black hoping for White 'a', Black 'b'.

However, AI has more tricks up its sleeve, and other skill sets, one of which is sabaki. Shibano gives the following example for the above position.

Click Here To Show Diagram Code

[go]$$W
$$ ----------------------
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ , . . . . X , X . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . 3 X . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . O . . |
$$ , . . . . 1 2 X . . |
$$ . . . . . . X . . . |
$$ . . . . . . . . . . |
$$. . . . . . . . . . . |
$$----------------------|[/go]

Please don't rubbish Shibano because of what you read here. Read the original.

You can rubbish me, but I'll add another thought to the mix to whet your bloodcurdling appetites. Much of the way AI plays can be understood in terms of the well-known human one-weak-group strategy, which implies great skill at shinogi. But AI adds some layers. One is that it starts much earlier than the human version (in the fuseki rather than the middle game), and it can perhaps be described as a several-weak-groups strategy, with a choice of which one to leave really weak till later. This in turns requires a skill set rather larger than just shinogi. One example is the sabaki just mentioned, that being used to settle the weak groups that are going to be settled before the isolated hare is left to run from the hounds. To how many pros, let alone amateurs, would that ever occur? Even when you see it, it is hard to accept, no?

Almost all decisions among top candidate moves AI makes due to decimal points. Decisions due to integer differences are the rare exception.

My study with AI includes both (a humanly manageable simplification of) decimal points and the 'big picture' (which I prefer to call 'go theory'). AI-driven go theory is an implication and interpretation of AI's decimal points. Without underlying data, interpretation is weak. With underlying data, interpretation is strong.

Although as a historian you ought to know, let me explain to you the importance of data evidence in the field of AI study. For little data (typically fewer than 100,000 playouts for each of the two candidate moves in a particular position), AI frequently changes its mind on the relative order of the top move candidates. This is often so regardless of whether they are all local or globally spread all over the board, that is, there still is no clear strategic decision (aka big picture) by AI. Accordingly, interpreting players (pro or amateur) can pretend whichever big picture they want to make their listeners believe as long as their knowledge relies on too little data.

Whichever teacher lacks support by reliable data cannot be trusted. Selling something as 'big picture' does not replace the necessity for evidence by reliable data but is weak, regardless of the playing strength of the teacher.

During decades before AI, I made the mistake to learn opening theory from professional players or their media. The more such opening theory I learned the less I understood because everything contradicted everything else. In the end, I felt like knowing nothing useful about opening theory. For the sake of improving my opening play, I will not repeat my mistake in the AI age: I will not learn opening theory from professionals having used AI but with insufficient data because necessarily the result would be the same: everything will contradict everything else.

Evidence is mandatory!



This is useful and important, of course.



Numbers can be translated to simplifying language, such as: good, fair, bad; correct, intermediate, mistake; Black favourable, even game, White favourable. It is unnecessary to bombard learners with numbers but teachers can translate their experience of AI bombardment with numbers to words. The opening (informal approximative words will do) is not the endgame (values can be necessary).



How are thickness, aji etc. static? They are dynamic concepts! Thickness describes a player's potential development and aji describes potential exploitation by the opponent. It is all fine and well that you have raised our awareness for dynamics for decades but please avoid downplaying concepts that are already dynamic!



No. AI dislikes moves such as 'a' because they give Black unnecessary extra safe territory where otherwise White can prevent it.



Not cute. Both are relevant for winning a game.



You do not want to hear it... AI does not use principles but uses decimal points. Principles can be human interpretation of what AI does.



No. According to AI play, 'a' can lead to such or Black taking the corner or Black pushing up then attaching at 'b' for peaceful settling. Not to mention Black crawling on the 2nd line being another AI possibility.



Not for AI.



This is what happens if people (pros?) do not study deep enough with AI. Instead of move 1, AI almost always chooses White c, also for 2-space pincers, with tenuki or other local moves being mistakes.


EDITs: typos.