Time to Go

Time to Go

This will be my last BaconLOG entry. It was started while working at the Atlanta Chess & Game Center as a way of getting the word out in the way IM John Donaldson does with his excellent Mechanic’s Institute Newsletter. After leaving I used it, for the most part, to comment on the world of chess. The chess world has changed dramatically since playing in my first USCF rated tournament in 1970. One of the biggest changes has been the rise of the machine. Often called an ‘engine’, I have come to think of the computer chess program as the ‘Oracle’. The Oracle has altered the way the game is played. It used to be that the final word on a move, or position, was given by a Grandmaster. The joy in analyzing chess was to study the analysis of a GM and possibly finding a mistake. Today GM’s write in their commentary things like, “Deep Purple says this”, and no one questions the Oracle. It has been said that ‘beauty is in the flaws’ and if the Oracle makes no mistakes, where is the beauty?
Former World Champion Kasparov has been working with the young GM Magnus Carlsen recently. The pictures shown on websites and magazines all show a third entity working with them, the Oracle. World Champion Viswanathan Anand was quoted as saying recently: I use computer a lot, I must admit. I check analyses, variations, and I have to do this, because everybody else does so, and one has to check and re-check everything. But I use computers a little strangely, because while I am looking at a position with one eye, I can be watching a film or doing something else as well.
I read recently that the Oracle had produced a theoretical novelty on move 34. THIRTY-FOUR! GM Vlastimil Hort was asked in New in Chess, 2009/3: If you could change one thing in the chess world, what would it be? He answered, “I would strickly expel and forbid all computers. Using them is a surrender of the human brain.”
It is not just chess that has been altered by the Oracle. I played backgammon with Dan Heisman at a World Open earlier this decade. We only played for low stakes, only one dollar a point, as it was the first time I had played for money in almost two decades. After our session he asked about my past. I said that I had previously played professionally in the late 70’s and 80’s. He said, “I could tell. You’ve got that 80’s style.” He went on to tell me that a computer program, ‘Snowy’ had altered thinking on the game. As an example he said it was now commonly accepted that most players would make the two point with an opening roll of 6-4. I cringed at the thought of making that play, as it far too early to make that inner point as it limits one’s options considerably. I mentioned something I had read on the Chicago Point website: “Back in the 80’s players had style; now they are all techies.” Something similar could be said about chess.
Hans Berliner wrote in the NYTimes, Feb 6, 2003: “You don’t have to be really good anymore to get results. What’s happening with chess is that it’s gradually losing it’s place as the par excellence of intellectual activity. Smart people in search of a challenging game might tray a game called Go.”
I have recently begun to study seriously the ancient oriental game of Go. Although I learned how to play decades ago, I have only played a few dozen games in my life. I now have two games ongoing on the Dragon Go Server. Because of the Oracle, that is simply not possible with correspondence chess. Fortunately, Go programs are not very strong, offering little, if any, help. I must think for myself and it’s the same for my opponent. Peter Shotwell writes in the forward to Go! More than a game: “It’s almost infinite complexity has defied computer programmers attempts to ‘memory crunch’ the game as they have done so successfully in chess. Low-ranked amateurs can beat any program, and the situation is unlikely to change much in the foreseeable future as it would take more than a lifetime to play a program that plays Go the way Big Blue, Fritz, and Deep Junior play chess. A chess champion who aided in the development of Deep Blue recently commented that computers have changed the way championship chess is won, because all the top players must now employ them to study complex combinations. On the other hand, only human minds can play Go well, making the Go board one of the last places on earth that has been unaffected by the incursion of modern machinery.”
I participated in many studies at the Georgia Institute of Technology psychology department, with many having to do with memory. Not only was I paid, but I also received results of the studies. One of the most important things I discovered was that, as one grows older, it is very important to try and learn new things. In the forward to Go Fundamentals by Shigemi Kishikawa, John Fairbairn writes: “Go has apparently been shown to provide beneficial intellectual stimulation that aids in staving off senile diseases. This may (like many of its benefits) be because it is a game that relies heavily on pattern recognition rather than pure analysis-right brain over left brain.”
The beauty of an idea is that it was discovered by a human mind. Computer programs do not have an idea; they only produce what it has been programmed to compute as the best move in a given position. It has been written that the difference between chess and Go is that while chess is akin ten to the twentieth power, Go is ten to the power of two hundred. Again from Peter Shotwell, “Part of the mystique of modern computer Go is the game’s sheer insolvability. Ever since Wang Ni noted in 1050 that no Go game had ever been repeated, many statistics and ‘folk lore’ have accumulated. One popular adage is that ‘There are more possible games than atoms in the universe’. Because the board is so large, even after pruning, the first 14 moves of Go produce a search tree with ten-thousand trillion leaves. It would take Deep Blue, which analyzed two-hundred million chess positions every second, over a year and a half to play one move of Go. Still, it would not know if that was a good move, because, unlike chess, Go is so vague in terms of profit-now versus influence-later calculations, and is so complex on a local scale.”
What seems now a lifetime ago when I was playing both chess and backgammon can best be summed-up by words from a Bob Dylan song, Shelter from the Storm:
'Twas in another lifetime, one of toil and blood
When blackness was a virtue and the road was full of mud
I came in from the wilderness, a creature void of form.
“Come in,” she said,
“I’ll give you shelter from thestorm.”

The love of my life read to me, what has now become a classic quote, from a book she was reading, by Trevanian. “What Go is to philosophers and warriors, chess is to accountants and merchants. “ The quote usually ends there, but it continues: “Ah! The bigotry of youth. It would be more kind, Nikko,to say that Go appeals to the philosopher in any man, and chess to the merchant in him." But Nicholai did not recant. “Yes, sir, that would be more kind. But less true."
She seemed to derive satisfaction from the fact that what she read bothered me…Now that I have delved more deeply into the game of Go, I have a much better understanding of what the author meant by the exchange. Players of Go consider chess in much the same way players of chess consider checkers. Zhang Yunqi lists the qualities required to excel at Go as, “The tactic of the soldier, the exactness of the mathetician, the imagination of the artist, the inspiration of the poet, the calm of the philosopher, and the greatest intelligence.”
Microcomputer executive and expert Go player Nolan Bushnell said, “Those interested in impressing others with their intelligence play chess. Those who would settle for being chic play backgammon. Those who wish to become individuals of quality take up Go.”
During my studies of theology and philosophy, I have been most attracted to Taoism. Several years ago while on retreat at a monastery I read a book that changed who I am. That book is, Change Your Thoughts - Change Your Life: Living the Wisdom of the Tao by Dr. Wayne W. Dyer. I have heard it said that you are the books you read read the people with whom you associate. My path first led me to chess, then backgammon, and now to the beautiful, nebulous, mystical game of Go. The Great Man, World Chess Champion Emanuel Lasker said, “If games are played by sentient beings on other planets, then they play Go.”
Michael Bacon

I only played Go a few times, but I wonder why it’s not possible to derive positional principles that a computer could use. It seems from my brief experience that one can guess several things that might happen in a position, but which of them is important depends on detailed calculation. In chess, the themes are not simpler but one has a better idea which ones are dominant in a given position.

So Go would require exact calculation interspersed with positional judgment. The successful chess programs calculate til the end, not pruning the tree very much, then evaluate. More pruning would be required with Go, or maybe we just have not identified the right underlying principles.

It’s surely true that computers have changed chess. As a competitor, one has to use them. As a now-spectator, I marvel at the level of GM competition these days. (I was never a GM but I got beyond the pushover stage.) The computer has changed the way the game is played, and allowed us to identify new and better “religious” principles to modify our older ones. The game is more dynamic than it ever was before, with lots of temporary sacrifices and even permanent ones. Things that would have been considered very brave are now seen as rather ordinary.

How do top Go players study? Isn’t there a way for them to use computers to enhance and develop their intuition, too? I predict that in a few years, computers will start having an impact on Go too. And I didn’t say anything about memorizing openings; that is not the main point.

I would agree that Go is not unsolvable. My very limited understanding is that it is a matter of: a) computing power being able to handle the complexity, and b) I don’t think the resources have been thrown at developing computer Go that have been thrown at computer Chess. If there’s a company with the size and resources of ChessBase, Convekta, or IBM for that matter that has tried taking on Go as a project I don’t know of it.

Then again, the last time I looked at the state of computer Go, Cosmos was “it,” and I enjoyed playing it on a 286. Is “Many Faces of Go” still the current leader?

ETA… As to move pruning… I wonder if that 14 move figure takes into account that any given board position in Go can be repeated in four directions and still carry exactly equal tactical/strategic weight and influence?

(Example: Black plays a first move on any corner star point. This board is identical if Black had played to any other of the corner star point. What would be four nodes in Chess are one in Go. If Black had played to any of the four central star points, the result is identical. Eight nodes are thus two.

On White’s first move, if White plays the the star point opposite, all those boards are ‘identical.’ So, given a play other than the central star point, what would have been 16 nodes - two ply to 8 possible points… are in actuality only 2 nodes. While I don’t believe initial play to a star point matches any Joseki I’m aware of, the principle still applies. I would wild-***-guess that Go will be mastered in computer terms by Joseki mastery coupled with fuzzy logic decisions on how those Joseki will interact with one another.)

It’s true that computers don’t “think,” they act in accord with their programming. However, brute force board analysis is not the factor in computer Chess that it once was, either. Power in modern computer Chess comes from variation pruning of one sort or another. So it will be with Go, someday - the move search algorithms will be improved by elegance as opposed to brute force programming methods.

I’m not trying to denigrate Go - it’s a great game. Some of our local players enjoy playing it, including this one occasionally.

Here’s the thing, from my perspective… One does not need to own or use a computer to enjoy chess. To improve, probably. But to play and enjoy a casual game? Or tournament play in one’s rating bracket?

That’s another beauty of the Oracle: One can switch it off and ignore it. Maybe you won’t be able to be a Grandmaster that way… but who wants to be a GM? :wink: :wink: :smiley:

The other nice thing about Chess, Go, Backgammon, Cribbage, or Scrabble: It will always be there, waiting for a return if you want to come back to it. :wink:

. .

At some high level of complexity, advice about an isolated move or two during a game of GO would cease to be of any use to even the “dan” (?) level Go players: it would no longer be plausible to understand why the computer is recommending that move.

It is similar to chess wherein a grandmaster or Fritz could tell a class level player what the best move is, yet it would be best only because there is a complex strategy or tactic that then has to be followed forward with the next several correct moves. The class player cannot find these moves.
So again, advice from Fritz on an isolated move is sometimes useless to the club player.


It is false that grandmasters takes the advice from Fritz in an “unquestioned” manner.
The next crucial milestone is when a computer vs. computer+HumanWhoCanOverrule is won by the computer alone. Today the human+computer team still wins.
. .

Michael - thank you for the time and thought you’ve put into your blogs and other posts.

Best wishes in whatever you do, and Happy 2010!

Either that or Handtalk.

One of the challenges inherent in programming a computer to play go, as opposed to chess, is that possible moves vastly outnumber sensible moves, and you have to evaluate them all – either that or reduce the computer’s analysis to a form of pattern-seeking that has difficulty accounting for the whole-board position. Go often rewards intuitive play, and it’s hard to endow a computer with intuition.

The game of Chess has always been (to me) a strange and beautiful thing. Sort of like watching the few wild mustangs that are left, or the wild elk, and buffalo. I realize at any minute I might turn to see a road full of cars, but for just a moment, a fleeting period, I can see the beauty as it was, before it was lost.

I refuse to fool with the computers, but just move though some of the Masters games(some new, some old) and there for just a moment, a fleeting period, I see the “GAME”. As the old “Gentlemen of the GAME” played it. I truly pity anyone who has not had that “Moment”. Just a dumb ol-cowboy’s take on it.

True, but computer Chess acts in this manner as well. There is certainly an order of magnitude difference between having 20 candidate moves and 360 candidate moves. (Literally!) But even 20 candidate moves going 3 to 4 ply deep is a pretty awesome number to reckon with, and going 12 ply deep is an even more awesome number. (If one considers large numbers awesome. :smiley: ) Yet Chess computers manage to prune the decision trees reasonably well at this point. To the point of rejecting a candidate move on 6 ply only to revive and play it on 12 ply because a sacrifice becomes a mating net - if the computer has time to get to 12 ply.

I can think of two advantages that computer Go programmers may have: First, excluding captures (which does muck up this advantage quite a bit,) there are only ever 361 possible moves, ever. (That is, 361 possible places to place a stone.) In Chess, the sum total of possible potential single moves for one side is:

64 King positions
64 Queen positions
64 Rook positions
32 Light-square Bishop positions
32 Dark-square Bishop positions
64 Knight positions
56 Pawn positions (a player’s pawns will never be on their first rank… This could be reduced to 48 if one considers the pawn starts on second rank and will never be able to ‘play’ to second rank.)
Equals 376 possible piece locations.

But in Chess, where a particular piece class is, and the move sequencing of how it gets there, is paramount. Go has sequencing issues, I know, but does not have (in my inexperienced opinion) the same degree of sequencing complexity. (It matters very much, for example, that the Queen is not brought out early, that the Knight gets to an outpost when the board position is such that the Knight will be secure there, and that flank pawn moves come after central power is settled. It doesn’t matter at all that a stone is played to B3 or H12 on the first or last move, as long as it has had a vital role in either establishing the player’s territory or denying the other player territory.)

Piece placement in Go is thus almost, but not quite, a state function. Piece placement in Chess is never a state function, except that the computer carries out its analysis given current board state.

And this brings me to Second: In Go, “winning” relies upon calculation of total area one controls in a living status, less prisoners sacrificed to reach that area amount. Computers already seem to be very good in making those calculations in final board positions. Even AIGO on my Palm rarely makes mistakes determining whose territory an area is in final scoring. (And, of course, the computer knows exactly what the score is, given that perfect estimation.)

Success therefore is a function of maximizing that area for a computer, with the occasional ‘rogue’ move thrown in to deny the enemy territory. As opposed to move sequencing to achieve a mating position. That does seem susceptible to brute-force calculation, which is a function of computer power resources.

But the algorithms are completely different, to calculate optimum maximization of territory versus move sequence to achieve checkmate. And it wouldn’t surprise me at all that territory maximization algorithms are already present in real-world applications. And the equivalent to “intuition” in a computer system is currently various AI routines, including fuzzy logic.

All this long discourse (that I could well be wrong about) brings me to a question… I agree that for a kyu level player that intuition plays a large role. But I wonder how many dan (and advanced dan) players would agree that they play by intuition, or make purely intuitive moves?

It still will not surprise me in the slightest that someday there will be an expert system geared at playing Go.

Funny you mention that. I just touched on that topic in my blog in regards to game postmortems in a database… My conclusion, for the club player, is that one could ‘dumb down’ the computer a bit - at least to the point where at least one of the lines recommended make obvious sense to the player. Either by ply depth limiting, or reducing the engine’s Elo strength if that is an option. Ideally, one would analyze with a program cut-down to one’s level and also a full-strength program. One can only improve if one understands the improvement and can apply it.

I’m trying to remember what book it was I read on Kasparov-Deep Blue that concluded that Kasparov wasn’t just playing a computer, but was essentially playing the whole team behind Deep Blue.

The conclusion behind your statement, given where each side is using an identical program and hardware, is that this will be the point where human involvement no longer matters.

Given different programs, that would (if repeated with different GMs,) establish that the other program is stronger.

Still, one can enjoy chess without a computer at all, as Harry suggests. All one has to do is stop caring about getting one’s rating as high as possible by any [legal] method possible.

Yes, this is a good point.

A notable exception is endgame planning, where it’s often right to spend time imagining where you want your pieces, then later worrying about the details of getting them there. And until recently, when computers just seem to dominate humans in all aspects, computers have been weakest in (non-tablebase) endings.

Has pattern recognition been tried in Go programming? There are less dimensions to the pattern, since each “square” has only three states: empty, black or white.

I was at a tech conference recently in LA. One of the presenters stated that Intel would likely have an 80 (yes 80) Core chip within a decade for the PC. Put a handful of those in a server class machine and even Go’s size starts to look a wee bit smaller.

----- Ed

You know what I forgot? Castling! :open_mouth: :blush: :smiley: I suppose castling on either side and either color ought to count for four more ‘locations’, even though the locations of the Kings and Rooks would be implicit in the above.

And I agree that pattern recognition is a key in Go. It would also be a critical element in my rought concept above of using fuzzy logic to negotiate the differences between joseki (book) positions.

And, on the meat of the topic, there are times for a lower-rated player like myself to disregard the Oracle, as well. For recent examples I’d refer to:

This post on my blog - when the computer analysis goes beyond the player’s capability to actually understand the recommended candidate moves, and this post on my blog for an example of disregarding the computer’s recommendation in postmortem analysis.

3^361 = about 10^172. That’s unimaginably big, even if one assumes that only one in (fishes around in hat) 10^50 possible positions makes any sense. And that’s just the positions; the play will be a much bigger number. After only 120 half-moves (ignoring rotations, reflections, captured stones), there are something on the order of 10^297 games.

Pattern recognition is a stepping stone to true intelligence. What chess software uses pattern recognition extensively? Think of the blockade positions in chess that human players recognize at a glance, but still mystify Rybka.

That’s a lot of positions, and to develop positional principles, one could never look at all of them. But one doesn’t look at all chess positions either, even all reasonable positions.

This would probably require human intelligence, like examining master-level Go games and seeing which positions end well and which don’t. Rybka is trained by having GMs evaluate positions and somehow inserting that wisdom into the positional evaluation function. The “somehow” in the case of Go might be prone to a systematic approach, due to the structural simplicity of the game.