Rational choice by amateurs

[RobertJasiek]

The tentative wall consists of three strings.

You do not understand what is alive? Come on. It does NOT mean to have surrounded eyes. Do not pretend to be an absolute beginner.

That a string is a string does NOT mean that the three strings are connected to each other. Yes, we know that a string consists of stones. Their string-constituting connection (string-connection) does not automatically establish a connection between the strings.

Yes, you misunderstand.

[jlt]

I may also have misunderstood your definition. Is it correct to say that, according to your definition, the group consisting of the string of three stones is thick, the group consisting of the two strings of 5 and 2 stones is thick, but the whole group of 10 stones is not?

[ez4u]

In terms of the OP, we seem to be confusing rational choice with reason-based choice. The many undesirable effects of the latter are quite well reported over the last thirty years.

I found the original paper ("Thinking Too Much: Introspection Can Reduce the Quality of Preferences and Decisions", Wilson and Schooler 1991) but only behind a paywall. One interesting point was

"...The Consumer Reports rankings were based on the ratings of seven consultants who were trained sensory panelists. These experts rated 16 sensory characteristics (e.g., sweetness, bitterness, aroma) of 45 jams; these ratings were averaged to compute the ranking of each jam..."

In other words the experts agreed on a list of reasons in order to rate the jams. Did this make the experts' list less reliable?

One of the sources cited in the original paper was "The disruptive effects of explaining attitudes: The moderating effect of knowledge about the attitude object". The abstract of that paper explains... [Emphasis added]

Previous studies have found that asking people to explain the reasons for their attitudes can change these attitudes and lower attitude-behavior consistency. We found that people's knowledge about the attitude object moderates these effects. In a reanalysis of an earlier experiment, we found that analyzing reasons reduced the correlation between dating couples' attitudes toward each other and their break-up rates only for couples who had been dating for a relatively short period of time. In Study 1 analyzing reasons reduced the correlation between undergraduates' attitudes toward a political candidate and the number of fliers for that candidate they were willing to distribute only for those who were unknowledgeable about the candidate. In Study 2 analyzing reasons changed undergraduates' attitudes toward a set of political candidates only for those who were relatively unfamiliar with the candidates. Several possible reasons for the moderating effect of knowledge are discussed.

I found all this interesting. Of course YMMV.

[RobertJasiek]

Is it correct to say that, according to your definition, the group consisting of the string of three stones is thick, the group consisting of the two strings of 5 and 2 stones is thick, but the whole group of 10 stones is not?

Yes.

Next, the question arises why would you even consider thickness of part of the tentative wall? Concentrate your study on important aspects of a position.

If we only assessed thickness of the top right two strings, we would only be judging on how well these two contribute to protecting the right edge. If we assess the three strings together, being or not being thickness affects most of the board. This is important.

***

Similarly, imagine a large tentative territory region of yours. If only fragments of your surrounding strings are thickness, they protect little territory. If all essential surrounding strings are thickness, they have the potential to protect the large territory region. It is important to have them all as thickness and not allow large parts of your surrounding strings to fall into pieces. (As a related study, we must also assess whether - despite the surrounding thickness - an invasion life is possible.)

[John Fairbairn]

This portion of the discussion has turned into a discussion of thickness, and a discussion of a highly idiosyncratic definition of thickness at that. FWIW, the pro does not mention thickness at all and instead mentions that if White is allowed to play at A that would make Black thin at the top. Furthermore, he mentions a "weakness" at the top which Black A helps to cover. Embedded in all that is the implication that there are tactics available here. None of this would allow us to apply the standard definition of thickness here.

IOW, for him, it's not a thickness problem, it's a direction of play problem.

That in turn means the true decision has been made at a level one notch up. It was a problem of perception. The pro perceived it is a DOP issue. The amateur perceived it is a faux-thickness problem. Take your pick.

[jlt]

FWIW, the pro does not mention thickness at all and instead mentions that if White is allowed to play at A that would make Black thin at the top.

I probably misunderstood something again. Isn't thinness related to thickness?

[Bill Spight]

FWIW, the pro does not mention thickness at all and instead mentions that if White is allowed to play at A that would make Black thin at the top.

I probably misunderstood something again. Isn't thinness related to thickness?

Thickness, as an English go term, usually refers to a thick shape or local position. The Black wall on the top side is too weak to qualify as thickness, whether White plays at or around A or not. It is not a matter of degree. OTOH, the Black position at the top can be to a greater or lesser degree thick or thin.

Edit: I see from a previous note that Jasiek defines thickness as something that a position has. The standard usage is that thickness is something that a position is.

[John Fairbairn]

I probably misunderstood something again. Isn't thinness related to thickness?

It's related but there's a dichotomy in Japanese. The adjective atsui can be converted to two kinds of noun, atsumi and atusa, which are both used in go but are different but both usually translated as thickness in English. That's the real source of confusion (even before other people come along and try to create new definitions). But usui = thin converts to just usumi in go.

That's understandably too esoteric for most people, so I recommend just going with what Bill said.

[Elom]

From my understanding of rational, the better a choice's reasoning, the more rational a choice is. Amateur jam tasters can talk themselves into reasoning far worse than the subconscious. So the conscious cannot do the subconscious job well without training.

I'm not too competent with go engines, but it seems self-learning AI are based upon the human brain. For any position, your brain comes up with the best options in an instant. Most moves are rated as really low (such as first line plays and senseless self-captures), leaving a couple of candidates in the green. Second, you verify them through search, the ones you feel to be most likely best getting searched more than others, each position repeating the process for the first one. Then you change the values the more you search. Better players usually have fewer candidate moves, but sometimes weaker players discard moves that a stronger player wouldn't. This can be due to assuming a move is bad because it doesn't seem to follow some strict rule— and could explain why many amateurs are often surer of their opinions than pros!

I guess when speedily solving many easier tsumego or when playing a fast-paced game, you have to rely more on perception, and your reading must be accurate and quick. Solving a hard tsumego or playing a long game, you need the will to last and conduct a thorough search many times.

What about a situation where A wins more against B, B wins more against C, C wins more against D and D wins more against A? This hearkens back to the style discussion, but it may also show those players have developed different perceptions. I wonder what kind of perception today's young future pros would have from studying AI games.

Maybe it just means that we should be wary of our reasoning, rather than abandon it. More accurately, be wary of oversimplifying. So far this discussion tells me that when reviewing games, you should try to describe your feelings during the game, then explain your reasoning as to why you felt that way, then explain your verification process. In addition to you avoiding false reasoning, reviewers may be able to identify separate problems with perception, reasoning and reading.

[yakcyll]

Maybe it just means that we should be wary of our reasoning, rather than abandon it. More accurately, be wary of oversimplifying. So far this discussion tells me that when reviewing games, you should try to describe your feelings during the game, then explain your reasoning as to why you felt that way, then explain your verification process. In addition to you avoiding false reasoning, reviewers may be able to identify separate problems with perception, reasoning and reading.

Most definitely. It is especially important to follow such a process when using a bot to review games; it's easy to fall into a trap of looking for justifications and explanations of the percentages and their deltas instead of analysing the process of selection of a move, i.e. becoming biased. First go through a game and pick out the points you focused on during the match, then perform a more thorough, perhaps written analysis and only in the end should you look at whatever feedback a bot can provide you with. At least I find this to be a much more efficient method of review than only utilizing the last stage.

[Gomoto]

How about just playing the practical move.

The reason for my move: It works.

[Elom]

Mistakes in rational choice by amateurs often occur through taking isolated aspects of something and moving it away from it's practical effect and towards a general idea of the whole being good or bad depending on one part.

I feel the use of checking our reasoning can be found in the meditative schools of Kyūdō— shin-zen-bi, truth-goodness-beauty.

It says an empty true mind naturally hits the target.

So making sure a good method is used to find out what should work and get it to should improve one's accuracy and speed .

edit: I left out half of the first sentence. Whoops.

[EdLee]

the meditative schools of Kyūdō— shin-zen-bi, truth-goodness-beauty.

It says an empty true mind naturally hits the target.

True, it's not the conscious mind, but rather the subconscious -- the deep neural net after 10,000 hours of efficient training -- that hits the target.

Applicable to many disciplines: photography, boxing/martial arts(hitting the targets), car racing(hitting the optimal curves, speeds, braking, accelerations), billiards, skiing, writing, cooking, carpentry, painting, etc. etc.

弓道; 真善美

IMG_0650.JPG (30.81 KiB) Viewed 13235 times

IMG_0651.JPG (145.84 KiB) Viewed 13232 times

[Elom]

the meditative schools of Kyūdō— shin-zen-bi, truth-goodness-beauty.

It says an empty true mind naturally hits the target.

True, it's not the conscious mind, but rather the subconscious -- the deep neural net after 10,000 hours of efficient training -- that hits the target.

Applicable to many disciplines...

Yes , I often feel like there are many ways to go wrong, and the entire purpose of practice is to make all the mistakes you may, so that your subconscious can register them and avoid them automatically. A bit like building immunity to different mistakes.

So in archery, the concept of hitting the target seems straightforward. But in practice, there are so many ways to not hit the target, and the closer a shot that misses seems to one that will actually hit it, the more likely you are to mistaken it for a shot that's successful.

In physical sports, there are many mistaken movements, and in mindsports, there are many mistaken plays but AI can self-play for millions of games and make all the mistakes that it needs to .

[John Fairbairn]

I mentioned earlier the word "perception" as a word that came to me after being reminded of Gestalt psychology and that seemed apposite to go. I have pursued the word a little further and it now seems even more fecund in its applications to go.

It seems that one of the key elements of perception is the human ability to infer: to fill in information that is not there. Apparently the experts are not yet sure why we do this, but their best guess is that it's an evolutionary adaptation brought about by the fact that so much of what we see or hear comes to us as incomplete information. Certainly, what most of us see on the go board is a potage of incomplete information! We'd love to adapt!

A classic example of perception seems to be Kanizsa's illusion:

img081.jpg (7.28 KiB) Viewed 13144 times

We "see" a white triangle imposed on a black one, but in reality there are no triangles in the figure.

My guess is that this resembles what is going on on the go board with this importance difference: pros see the triangles, we amateurs see only the Pacmen. My starting point for this guess is something I realised about go quite a long time ago, and for me it was the most important insight I've ever had into go. It didn't make me stronger (and as to why I'll mention that later) but it allowed me to appreciate the game at a much higher level. The insight was this: go is supposed to be the surrounding game; pros play the surrounding game; amateurs play the counting game. Or, to repeat the comments on the figure: pros see the triangles spanning the whole board, we see the Pacmen territories in the corners and on the side.

Or, to belabour the point a little, amateurs think surrounding applies just to countable territories whereas pros realise that you can surround abstract things like influence.

Always assuming my insight was on the right lines, why didn't it make me stronger? I think it's because you only learn to "see" (infer) the
ley lines if you put in the notorious 10,000 hours.

That then raises the question as to what benefit the ley lines give the pros. I think the answer lies in the very important remark Bill made earlier on: reading (calculation) is best used for confirmation of ideas we come up with first.

No-one can read out everything. We approximate by using tricks to prune the tree. Every player beyond a Day 1 beginner has experienced this. You look at a tsumego problem and if you try to work out every move on a "if he plays there I play here" basis, you soon run into the sand. But if you recognise a shape (e.g. a space where you can play a nakade) you can choose the right move instantly, just performing one or two lines of analysis to make sure there's nothing hidden there, such as a possible seki.

In other words, if you have a specific goal in mind you can read ahead much further and more easily. My guess is that pros can do this for areas way beyond the scope of tsumego and also beyond the scope of josekis and middle game josekis. They can do it for the whole board because they can see the invisible ley lines there. AI bots are effectively doing the same, but even more reliably.

I believe that counting - the be all and end all for many amateurs - is likewise just another confirmation tool for a pro, and again they can use it in the centre of the board in ways amateurs can't.

Obviously I've no personal experience of how to become really strong at go, but everything I read (or infer!) points to building up perception or something like it by playing over countless pro games. That's certainly what the pros have done. Some amateurs I know have tried this and often claim it doesn't work. Some desperately try to memorise pro games and even more say that doesn't work. I believe that's because they think it will work by osmosis.

The way to make it work is by effortful practice. T Mark Hall's now famous experience of improving two grades just by playing over the games of Go Seigen was an example of that. The point is that he was transcribing the games into sgf files from densely packed diagrams. He had to be "effortful" to find the next move on the diagram without scanning every line. He learned to look in the right area of the diagram to find the next move. That alone made him stronger. He was, if you like, laying down ley lines.

Not everyone wants to transcribe games as a way of becoming stronger, but they can't avoid putting some effort into playing over games. Just thinking about what is really being surrounded, how that little defect on one side of the board can have a tsunami effect on the other side, and things like that, may be a way of learning where the invisible ley lines should be. Details aside, that seems the real rational choice for amateurs here.