Life In 19x19

RobertJasiek wrote:IMO, subconscious thinking also relies on reasoning. Of course, I can't prove it down to the neural net's level:)

In this context, you must say something about what reasoning is. If you mean that a neural net (effectively) computes a function, then it is obvious. If you mean something more than that, your claim may or may not be true.

Edit: changed "define reasoning" to "say something about what reasoning is". The former is far too stronger a claim.

Ok, so it seems like we have a bunch of different lines of thought/ arguments about intuition here.

1. Intuition is a mere ruse of reason whereby it deceives itself by setting up an other to be ultimately reabsorbed into the universalizing activities of the understanding. Sure, sure, but maybe not so relevant to playing go?

2. Intuition is a way of playing that in fact is different from playing by consciously directing oneself to consider general principles; but in fact, intuitive play is merely the preconscious or subconscious consideration of the same principles. This strikes me as entirely dubious, at least the subconscious part. One notorious example: native speakers are brilliantly faithful to the rules of their dialect, and unerringly identify good and bad usage, but are incapable of figuring out what the rule in question is.

3. Intuition is a way of learning go that involves exposure to a huge amount of material, as opposed to memorizing rules that give good results; but in fact, the process of intuition is simply a long, laborious way of inducing these rules, which could so easily be learned directly. This seems wrong to me, and it seems to be pretty parasitic on the account of play style in #2. It certainly suggests that mc bots are a detour from the true path of go ai; and that immersion is the wrong way to learn languages; and that both dogs and quarterbacks should avail themselves of a kinematics course.

jts wrote:...It certainly suggests that mc bots are a detour from the true path of go ai...

This is the main reason I've always been sad about the use of Monte Carlo in bot AI. I look forward to the day when Monte Carlo is succeeded by more advanced heuristics. Even though the move tree is both deeper and wider and Go, I see no reason why we shouldn't eventually see the same story in Go (when Rybka et al broke down the number crunching wall with outstanding evaluation logic).

I think that "intuition" is both real and necessary. "Logic" falls down in many cases where there are simply too many possibilities to consider. That's why there aren't any really strong go programs yet on the 19x19 board. Every move in Go is solving not one but dozens of problems simultaneously: balance of influence, balance of territory, security of groups, command of tempo, and so on. At some points, the state of flux is so great that anyone who claims to be playing by anything other than intuition is deluded, because they eliminated a dozen equally good moves with their single move that fits their style.

I can understand how this could be uncomfortable, and I suspect Robert is going to tell me that I'm a moron. However, if you look at game commentaries even by the strongest players (Go Seigen, for example), there are not-infrequent comments such as: "I played 103 because 1 in diagram 14 felt negative for Black," where "negative" is just an over-pompous translation of "bad." An Younggil's comments are some of my favorites for this, because he will lay out a variety of alternate moves at some positions and say: "This is also thinkable, but it's a different game." It's also interesting to see other pros reviewing someone's game and say: "I would have played here, but it's not (Gu Li|Lee Sedol|Lee Changho)'s style."

Blake wrote:"Logic" falls down in many cases where there are simply too many possibilities to consider.

Logic does not fall down but a) must be better worked out and b) apply hierarchical decision making.

That's why there aren't any really strong go programs yet on the 19x19 board.

There are no really strong expert system go programs yet because the existing programs have implemented only a tiny fraction of human go knowledge. There are no really strong 19x19 MC programs yet because a) more MC design tricks are needed or b) expert system knowledge is also needed.

Every move in Go is solving not one but dozens of problems simultaneously:

Not simulatenously in general, but hirarchically or - for same level problems - in any order or - ,if parallel algorithms are allowed, only those at the same level simulatenously.

that fits their style.

When go knowledge is not advanced enough, then equally valued moves can occur and style preference is (still) an option. Choosing among equally valued moves is not intuition but application of reasoning to the currently available extent and precision of evaluation.

if you look at game commentaries even by the strongest players (Go Seigen, for example), there are not-infrequent comments such as: "I played 103 because 1 in diagram 14 felt negative for Black," where "negative" is just an over-pompous translation of "bad."

Weak teaching.

An Younggil's comments are some of my favorites for this, because he will lay out a variety of alternate moves at some positions and say: "This is also thinkable, but it's a different game."

Reasonable teaching. When there equally valued alternatives after one's judgement, then one should not pretend to have greater precision.

It's also interesting to see other pros reviewing someone's game and say: "I would have played here, but it's not (Gu Li|Lee Sedol|Lee Changho)'s style."

See above.

Intuition is just "unconscious thinking".

If you use only intuition or only logical thinking, you use only half of your brain. Use both !


Intuition is really strong, and can solve complex problem really fast but is unreliable
Logical thinking is slower, but far more reliable

That's why, in my opinion, the most efficient use of your brain is to pick your candidate moves with intuition, then verify if they really work with conscious and logical thinking.

Logical thinking can be used to simulate what some consider "intuition". The process is about equally fast. E.g., use Local Move Selection: set a sufficiently large environment where to look for a move for a local purpose, then filter quickly the obvious failures, then consider the remaining interesting moves more carefully. For global move selection, similar processes can be used, e.g., by applying the additional mask "identify the biggest space, in which to search for the next move". (If there is time, then every move should also be verified by reading.)

@Jasiek if it doesn't reveal too much from your books I don't have, how are you picking interesting moves to examine? I know someone who I always feel is holding himself back when he says; "I would never think of playing there; I don't know any rule or guideline that suggests it (at least over these other guidelines)".

I just look at the obvious moves (to me, at the time) and then any other ideas that jump out at me. These tend to be normal looking ideas (ie it's easy to come up with (or remember) real-sounding proverbs to justify them). I can't imagine a heuristic that would describe how to look in normal-looking spots.

I seem to be firmly in the "clearly everyone is doing both" camp (headed by president Nixon).


Edit: Oh, and I forgot the real reason I wanted to post. I super-disagree that negative is just fancy-talk for bad. Negative suggests playing reactively to your opponents' strategy rather than making a good plan of your own. ...To me, at least.

Loons wrote:Edit: Oh, and I forgot the real reason I wanted to post. I super-disagree that negative is just fancy-talk for bad. Negative suggests playing reactively to your opponents' strategy rather than making a good plan of your own. ...To me, at least.

I agree completely. The problem with not being a native speaker means that nuances of language are missed. Japanese has loads of nuances that I need explaining to me by people who understand them, and English is no different.

If I am to believe pro reviews of my games, I am both.

1. I use my intuition to a select a move which is usually pretty good.
2. I use my "logic" to convince myself that some other, inferior, move is better.
3. Then I play the "logical move."

Loons wrote:@Jasiek ...how are you picking interesting moves to examine? .... I can't imagine a heuristic that would describe how to look in normal-looking spots.

Perhaps you haven't noticed, but Robert went on to answer your question elsewhere. It does seem to me though that asking one's (s) elf what's urgent and then examining clumps of stones for ways they might lose a base etc., and then finding nothing, to scan the board for large places without stones would be a fairly logical heuristic to find some normal moves.

Robert does go into specifics on how to arrive at such decisions with the help of logic (principles). On p. 94 of Joseki 2, he introduces the concepts "'stability,' 'urgency' and 'investment'" which "prepare the concept 'playing elsewhere'," and includes a method for numerically evaluating the instability of a group, thus - at very least in theory - facilitating the decision whether to look elsewhere.

Edit: Oh, and I forgot the real reason I wanted to post. I super-disagree that negative is just fancy-talk for bad. Negative suggests playing reactively to your opponents' strategy rather than making a good plan of your own. ...To me, at least.

Just so you know, this did not go unappreciated

I use both. I am probably more logical than average, but I do not believe anybody possesses the capacity to analyze it as a game of pure logic in a game so complex. Tic-tac-toe, sure. I cannot hold the actual, full reality of the positions on the board in my head, so I analyze a puzzle based on my "feelings" on the positions and my interpretations of the whole board, as well as my "goals". Sometimes the "bottom line" representing a certain group of stones or objective in my head is wrong, or becomes wrong without me noticing, so I make a "logical" choice that is dependent on faulty variables. I also try to eliminate bad results or favor good ones through reading where I can, but it is much too difficult to be comprehensive for someone of my skill.

What a funny word "counter-intuitive". I definitely consider myself a counter-intuitive player, however I don't believe that go is "logical", because even though go is discrete in nature, the game has not been solved so there are no "Truths" only hypothesis and conjectures(or strategies).

There are truths for a few partial solutions.

You are confusing known truths with truths. There are obviously many unknown truths, for instance the value of x = 2349820948239048239056829046837279325897893472854789234798532 * 23904802981290582903859023890582349085903482590345890238542, which had not ever been computed* prior to my posting this comment. But this morning it was still true that there was a truth there.

* At least, it's extremely unlikely that it had been.