We all know that we tend to see only what we expect to see. We may be primed that way for good reasons biologically, but in truth in go it can be a bad habit, and as such one that is extraordinarily hard to break. I just encountered what was, to me, a devastating example.



This was the start of a Chinese game from the Ming dynasty, i.e. half a millennium ago. It looks weird. Part of that is because group tax applies, which means focus shifts to the sides. Another part of the weirdness is that both players started (after the four starting stones were placed, of course) at the 9-3 points on the lower side. Both these moves were classed as mistakes by Lizzie, giving a loss of about 4 to 5 % points, but that is without group tax considerations. But with the position shown, parity with the starting position had already been regained just before the last move, X. X was another move of the "mistake" level (a loss of 4%, though again with the caveat of ignoring group tax).

It was the square-marked move (the mole move) that I was interested in. This is a very common move in old Chinese openings of various types, even when there is no approach on that side, and I had always assumed it was either just plain bad or bad but connected in some way with group tax. By contrast the triangled move was rare. Nowadays we assume this is a good, or at least valid, move and can even smugly justify it - it makes the opponent overconcentrated.

Now overconcentration is a concept well known to the old Chinese, so they weren't avoiding the triangled move out of ignorance. I therefore did a quick check with AI, and to my astonishment discovered that the exchange on the left is better (marginally, but consistently) for the forcing side than the one on the right.

Despite my initial surprise, I could easily start to see possible explanations, but they were derailed a bit when I further noticed that it was specifically the last move (X) that was warping the assessment. If we just looked at the forcing moves (the triangle and the square), the AI assessment was reversed. But while triangle was better (my life-long assumption), I was further surprised to see that it was only marginal. Putting it all another way, I never expect to see the mole move on my radar. It is so bad (to me) it disappears from view.

But the surprises didn't end there.

I decided to check whether the mole move in this sort of situation had ever appeared in pro games. I expected about 3-4 hits. I got nearly 600, and that included some AI bots.

I found that close to incomprehensible. How could I not be aware of so many moves that I class as weird and so (I reckoned) I ought to have noticed? Yet I had not "seen" (or registered) any of them.

To rub salt in the wound, I found this position from 2018:



Cho Hye-yeon was playing Kim Ta-yeong. Black's next move was A, to which White responded with... B. So this move that was ostensibly bad in the first position was considered good by a pro here.

In fact, in the first position, although Black A was considered the best move by the bot instead of X, White's response to that, as per the bot, was X!

The thought that came to me then was that it may be a mistake to talk as glibly as we do about mistakes when talking about bots. Just because there is one move that is so much better in a given situation does not make the second-best move a mistake. What I am getting at is that we cannot, thinking we are being rational and learning, condemn a move like X as bad because it is, say, on the second line and so extrapolate to avoiding second-line moves in general.

Oddly enough, that same point was made by Muhammad Ali in a YouTube clip I watched last night. The fawning interviewer was asking why Ali bothered to fight the second, third and fourth best fighters when he was so much better than they were. Ali "demolished" (to use Youtube jargon) that argument by pointing out just how good his opponents really were. Just because they were not as good as him did not make them bad boxers.

I've already been staggered countless time by how in tune old Chinese masters were with AI play. This has been my most humbling lesson. Instead of Whack-a-Mole, we need a game called Praise-a-Mole.

KataGo can play with group tax (and no komi) rules so would be a better AI analyser. It makes a significant different to its views, e.g. no more direct 3-3 invasions. Lightvector posted about it at https://lifein19x19.com/viewtopic.php?f=15&t=17674.

Regarding the kick vs footsweep, one obvious difference to me is the kick has a better follow-up (hane / tiger-mouth on top) than footsweep (locally 2nd line crawl, though sometimes can play pincering moves from other side). But once white answers the kick by standing up black has a big weakness at q5 he ought to fix with q6 or r6, so the kick is gote overall. But the footsweep if black blocks does have a follow-up at d3, but this is not as a severe follow-up as the f2 move doesn't hurt d4 much, and is not more severe than if white had simply tenukid and not bothered making the footsweep for black's kindly block exchange, indeed the exchange helps white a little deal this things like d3 and crosscut.

The footsweep is part of my repertoire, but I rarely expect my opponent's to be so kind as to obediently block, expecting resistances like tenuki or double approach. Your pro example the corner footsweeping player already has an extension on the other side of the corner though, so it's quite different to the ancient Chinese one.

Andrew:

I'm aware of katago's extra capability but I can't get it to run on my pc, and I don't like the user-interface in the Master of Go app on my iPad that does offer katago. Sorry.

Your gote point is one I overlooked - thank you. My focus with that right-hand side of the equation was that the 3-3 point seems weaker there than when the mole has been deployed.

Incidentally, my prejudice against the mole goes back decades to a Go Monthly (or similar) article on handicap play (the norm then) in which a pro remarked that if White approached and Black answered with the mole White could treat that as non-answer and tenuki. I think I took it too much to heart.

Your point about the difference between the two diagrams I gave is valid as they stand, but actually the two forcing moves in question are most often played in old Chinese games after a move on the other side, exactly as in Diag. 1, and my desultory AI checks were mostly made for those cases. I chose the diagram here simply because the two corner patterns appeared side by side. I followed the bling

Thank you for sharing this. I am not even half good enough to understand your level of discussion on the goban, but... .
*the in brackets is my guess/addition.

_________________
When you play Weiqi you are joining millions of people across four thousand years of time.
Jonathan Hop - So You Want to Play Go?