Quote:
1. It has to be a valid domain, i.e. luck or randomness don't play a role, or a minor one; skill is the determining factor
2. There is one or more validation criteria
3. You get many repetitions
4. With instant feedback
and the fifth point is more an instruction than a criterium: you perform "deliberate practice", i.e. you regularly set the bar (a little) higher than your current level of skill
With some minor quibbles (e.g. "with instant feedback" can be replaced by "with feedback" and it might be argued that "determination" is the determining factor, not "skill") this is both nothing more and nothing less than trial and error. The one learning strategy that all creatures have to be rather good at. Dressing it up in pseudoscientific language or a "logical" list doesn't really add anything to it Indeed, doing that can cause harm in a complex process. Simple can be too simple.
Classic example of trial and error: babies learning to walk? They can't be instructed verbally. They don't understand language. They wouldn't understand a criterion if it hit them in the face, and if it did they
would burst into tears - negative feedback. They can imitate, but imperfectly, and it in any case seems a stretch that to assume they make the connection, at least initially, between lying on their stomachs and older people walking upright. But they are curious and hungry, or whatever, and something in their DNA impels them to make movements towards satisfying whatever their cravings are. And that's all. They have no specific goal ("learn to "walk"), just a craving - probably food, usually. Most babies don't even learn to walk to satisfy their craving. They learn to crawl first. They have no detailed list of criteria to use as feedback. They learn to move, in whatever fashion, towards an object of desire that might be food. At least it goes straight into the mouth. But it might be a plastic toy. Inedible. Failure? Baby doesn't think so. He'll go on and try to eat another toy, or the puppy, or the carpet. He might never find anything to eat until Mother comes along with a juicy nipple (which looks nothing like most toys, puppies or carpets but he knows what it is!)
But babies are good on repetition, and they are therefore also good on consistency and persistence. More to the point, they don't actually need a teacher. A parent might think they are helping by giving feedback such as "What a clever boy" (meaningless to the boy and probably confusing the puppy in the process). They might give physical help: holding the bay's hands to keep him upright as he toddles along. I have no doubt this benefits the child, but suspect it may not be in actually teaching him to walk. It seems it's more useful as a way of bonding parent and child in a loving cocoon. Encouraging him simply to continue his curiosity peregrinations, perhaps?
And not all babies learn in the same way, so that parents who think
they they are succeeding in teaching the tiny tot may be deluding themselves. I had two daughters. One learned to walk via the crawl. The other learned to pull herself up a sofa leg and then learned to walk along the sofa. Al0l by herself. She skipped crawling.
Is this method of learning relevant to go? Strictly, I'd have to say no, because playing go involves some rules, such as capture, and that's above baby level. But if you change the game to the "fuseki game", where the criterion, or "food" is juicy-looking clumps of territory, you can get a simulacrum. I believe that this sort of experiment relates to what kvasir had in mind, and also overlaps with the Kansai Ki-in method I mentioned. In short, we make or goal learning to walk in go (ikken tobis and kosumis) then to run (keimas) and
that is the sole goal. You are not trying to become stronger, get a diploma or crush the other guy. You are simply learning how to move forward reliably in good balance, gradually getting faster. Creating good suji. My experience of western go is that too few players can actually walk properly on the go board. They keep falling over or dunchi8ng into walls. Trial-and-error never seems to work for them. Why?
I think if you go back to being a parent you can see part of the reason. One thing parents like to teach is how to catch an object. Biologically useful. But many kids find it very hard. There's a multiplicity of reasons, I think. One may be negative feedback: you get bopped on the nose trying to catch a ball thrown to you. It hurts. Whatever goes through a toddler's mind equating to "sod that for a game of soldiers" makes him decide to do something else. Like eat the carpet. Often the activity starts too early. Baby has just learned the joy of walking and wants to indulge. When you go to toss the ball, he walks towards you instead. Usually with a heart-melting smile. See, feedback works - for you! Another problem is expecting Baby to have control of his fingers. He can't really cup them and doesn't yet understand why he should.
We see many examples of premature instruction (to coin a phrase) in go. Many go instructors have tried to refine the process. I'm not convinced they succeed. For example, a Taiwanese teacher devised an imaginary environment in which territories became pools where fish could swim freely or be trapped. I've got the book somewhere but all I remember really is that he designated his method as "scientific." This was on the basis he had identifying labels and lists, and he had applied "logic", all the while overlooking the constant problem with logic, which is that you can get totally different results depending on where you start - and deciding where you want to end up can distort the process even further.
I'm having to generalise a lot in this limited space, but I do think that the problem with this second-tier approach (i.e. the one after pure trial-and-error) is that too many people are trying to "teach" go. They are like parents, and fall into the same trap of thinking that if they "teach" something and the child "learns" something there must be a cause-and-effect. Pure logic, right?
Going back to learning to catch a ball, not all kids become proficient at it (perhaps because of things like the well-attested being-hit-in-the-face-or-the-goolies syndrome) but those who do, normally learn by imitation of their peers and by constant repetition (i.e. a lot more than 5-munutes parent-bonding time or 10-minutes attention from a teacher in a class). In fact, interference from a parent or teacher can cause precisely that: interference in the learning process. This becomes a major problem when we try to scale it all up to helping adults learn go. They usually create the interference for themselves. They think, "I've got a Master's in pantomime. I'm brainy. I just need to understand what you are telling me then I'll know it." Codswallop. You might "understand" some of it but you won't "know it". Human Big Bang theory (a source of hubris vaster then the fabled Siberian salt mines) versus the discoveries of the AI-equivalent James Webb telescope?
What is the proper way to put this second-tier
learning (nb not teaching) process into operation for kids. I have seen this in operation in one way in the UK by watching Peter Wendes and his wife Sheila take go to schools all over the country with impressive success (though apparently not built on by the BGA). Peter and Sheila were both highly qualified teachers but once they got over the necessary part of telling the kinds what a go board and stones were, they let the kids learn by playing and listening to, or copying, other kids. I would listen in to the kids and was totally fascinated by how successfully the kids interacted.
A more normal way of kids learning go, however, is one that we cannot replicate easily in the west, though it is one we are familiar with from chess. That is, we are not taught the game specifically but pick it up from our siblings or child peers or by imitation of grown-ups. The story of a famous go player (e.g. Kitani Minoru) learning go by watching grown-ups play (in Kitani's case by watching games in his father's barber's shop) is told of many people, and that's how I learnt chess - watching my father and his friend play. It's common in chess. A variation on this, not at all common in chess, apparently, is that Baby heard the sound of go stones being played while he was in his mother's womb, and so developed an attraction to the sound and thus the game. This story is told of Shusaku, but also of others, and I have seen a claim that this has been supported by scientific research. They key point, though, in every case is that the child is learning, not being taught.
Another thing often noted in the go world (and I believe it applies in chess, too) is that it is the youngest child in a family that becomes the best go player. The exemplar is Go Seigen. The usual explanation put forward for him is that the eldest son had to prepare himself for the responsibility of taking over the family and had to secure a livelihood. The middle son was the heir and spare (in European terms he would join the army - Wu's middle brother became a revolutionary). The third son in such cases no-one cares much about. He's just spare. So he can do what he likes. Even play go. Without rejecting that entirely as part of the explanation, I think the fact that the phenomenon is noted among women speaks against it. Clearly, external factors (such as wars!) have influence, but if we accept it as a working hypothesis, possible explanations would be the wish of the youngest sibling to imitate the elders plus willingness of elder siblings to encourage the younger ones (Confucian sense of duty?). If there is something in it, we can expect Ueno Risa to rise above even Ueno Asami! Poor Sumire has to do it all on her own - or go to Korea where more young kids are playing?
While hoping to be excused for leaping over some hedges and avoiding normal pathways, for space reasons, my conclusion is that the key element for children is self-learning, not being taught directly. Teaching should just be regarded as a word to loosely describe creating a suitable environment. I further conclude that a major weakness in western go is that almost all players have had to skip this self-learning process, and this has been combined with too many people wanting to be their "teachers" (the cynic in me thinks that, as with the nanny state, ego may come into the equation here). As I already mentioned, I also think a further handicap is that too many adults impose the "I want/need to understand" mindset (or, worse, the "I want to be logical" mindset) on the process and effectively put up a permanent bar to progress.
That is why I see kvasir's suggestion as the possibility of a revelatory process that may help western players go back to self-learning the important bits of go (i.e. suji) that have been lacking in their make-up.
You might then argue that, while AI can't teach you, you can do this self-learning by using it as an assistant. It would be nonsensical to deny that completely. After all, Go Seigen learned about go from Chinese classics (which rarely had commentaries) or from go books from Japan, such as Shusaku's collected games, that he couldn't read. The game records were his katago. But I think the big difference is that Go had already learned to crawl, walk and run in go. Most western players have only got as far as using crutches, and insist on using them on staircases at that.
My own suggestion for the "adults learning to walk" process is to forget the theory of putting each foot about one foot in front of the other - you might "understand" that, but it's a hubristic waste of time. Instead, practise, repeat, practise, repeat, and don't stop once you think you've got it right. You must never stop until you never get it wrong. Practise what? Well, pros recommend repeatedly doing what Go Seigen did - play over games over and over again. Essentially, you are looking to be able to say "in this sort of position, top players do that." Just that. No codswallop about understanding. After all, that's how AI learns, doesn't it? - it doesn't understand either. Build up your intuition. Once you never get the basics wrong, then of course all the various other ways of studying are not just still available to you, according to your preferences, but are more likely to work because you've nailed the basics.
Quote:
the fifth point is more an instruction than a criterium: you perform "deliberate practice", i.e. you regularly set the bar (a little) higher than your current level of skill
I know there are currently big academic debates about what constitutes "deliberate practice" and I'm not sure whether setting a higher bar level is now seen as an important factor. But FWIW, people will have noticed that I often write fairly lengthy pieces here. It's partly because (as a journalist) I can do this very quickly, and it's partly to make up for the lack of contributions by lurkers, but often another reason is my version of deliberate practice. By collecting all my thoughts on a topic down somewhere together, I find that I am helping my intuition or subconscious. I am feeding it. I am not pretending to understand what I've written - I often don't and happily expect to change my mind at some point. I don't consciously mull over it again. I let the clever part of my brain - my intuition - do all the hard work of parking all this data in my head. I expect to find it there, neatly sorted, when I next need it, or when I next need to modify it. I find this procedure works far better than just reading swathes of material in books or on the internet.