BACK INTO THE MAILBAG

by Steve Lopez

Several weeks ago a received a really interesting letter from Don Prohaska on the subject of artificial intelligence (A.I.) and computer chess:

T-Notes might want to think about this. I've read most of the stuff about this matter but I am not convinced. I used the top 4 programs on the Swedish list going over the 6th game of Kasparov/Deeper Blue. On Kasparov's h7, each of these programs played the same move as Deeper Blue, and instantly several of the following moves. But, if I turned the opening book off, none of the programs sacked the knight. Forcing the programs to make the knight sac, each program gave black a very large advantage. Now, I have no way of knowing how Deep Blue plays, but it seems to me if a program analyses a position and calls it bad and doesn't make the move, but forces itself to make what it thinks is a bad move because its look up table tells it to make that move, that is not playing chess. At least if we're trying to see how well the computer plays. Further, Inside Chess in going over one of the games said that it was not known that Deep Blue had one of the perfect end game modules and so Kasparov played on when he had no chance of winning. Now, if I play an end game against a computer and it has a perfect end game module with the configuration on the board, does the computer beat me because it plays chess of it just selects from its look up table. I like all this stuff in my computer program, but if I were trying to see how well my program played chess, it would have nothing but the analysis engine in it. I can find no good argument that tells me that a computer beat Kasparov at chess or how well a chess program really plays against humans.

This is a pretty interesting letter. Don raises some points that have been the subject of much contentious debate in the year since Garry K. and Deeper Blue squared off. Is a computer "thinking" when it plays moves from an opening book? Is it "thinking" when it plays moves from an endgame tablebase?

To answer that, I have to ask myself whether I'm "thinking" when I bang out the first 10 moves of the Worrall Attack from memory. I probably had to ponder each move when I was first learning the opening, but now I just fling the pieces out there without too much consideration until my opponent plays something "out of book" (e.g. weird).

Back in 1993, I bought a copy of a shareware chess program from a local computer shop. I was startled to learn that the program always responded to 1.e4 with 1...e5. If I continued with 2.Nf3, it responded with 2...Nf6. The program had no opening book; hence, I played many a Petroff against it. I've since learned from my father (who has the same program) that I bought a defective copy: his has an opening book and will considerably vary its play.

I'm bringing this up to illustrate a point: that without an opening book, computers will tend to play the same thing over and over. It was determined a long time ago that computers need an opening book to be able to vary their play. It's also generally been accepted that a computer playing from such a book is no different from a human player banging out the moves from memory by rote.

But there's a deeper (no pun intended) question here: how large should we make a computer's opening book?

Back in the days of Fritz2 and Fritz3, a lot of users were pretty hot to get the optional Power Books in order to expand the program's opening library. But users who got the whole set noticed an interesting phenominon: Fritz was sometimes playing into disadvantageous lines and opening traps. The reason for this was becuase the Power Books were designed to make Fritz a more effective training tool for the human user. The program would occasionally play less-than-optimal lines to simulate the play of a typical human opponent: our tournament foes very seldom play the best possible moves. Learning to bust a particular line at home should give you an edge should that line be played against you in a tournament.

So in this case, bigger is not necessarily better. Users who wanted to have Fritz play only the strongest possible lines were directed to use one of the smaller opening books, which were fine-tuned to provide Fritz with lines that would maximize its winning potential (e.g. sharp tactical lines at which computers tend to excel).

Another factor is the strength of the opposition. I remember another shareware program that I used to fool around with that allowed the user to customize the opening book. It was pretty easy -- you just used a DOS text editor to type in the lines that you wanted the program to "know". It had a maximum capacity of something like 100 different lines of play. For a fish like me, that was plenty. My copy eventually wound up with around 75 lines or so in the book. But a master would have taken this limited repertoire and blown holes in it big enough to drive a truck through.

Better opposition means that the opening book must be larger. So where do you stop? Excellent question. The basic opening book and database for Deeper Blue included every master/grandmaster game that IBM could get their hands on. Duplicates were eliminated (a mammoth job that took several weeks). This database was then turned over to Grandmaster Joel Benjamin, who fine-tuned it to include many off-beat lines (including responses to moves such as 1.e3). The end result was an opening book that included all known chess theory. Overkill? Perhaps, but then again, they were about to face the world champion.

Was this fair to the human player? Grandmasters know tremendous amounts of opening theory. I once read that a typical GM is familiar with a quarter-million opening setups (tabia). It's probable that no GM knows as much theory as Deeper Blue "knew" going into that match. In the case of Game Six, it's probably a moot point. I was watching the game on ChessNet and as soon as Garry pushed the h-pawn, all of the GMs and IMs bugged out and started screaming that he'd blundered. It's a pretty well-known trap and I know a few class-level players who are familiar with it. Garry just plain screwed up.

By the way, while Deeper Blue has access to the five-piece endgame tablebases, they weren't used in the match. ChessBase sells these exact databases on CD for use in ChessBase 6. None of the endgames covered on these CDs (the now-famous Ken Thompson endgame CDs) occurred in the match. Garry has had these CDs for years. Inside Chess is mistaken in their assertion that one of these games could have ended with Deeper Blue consulting the tablebase. (No offense to Yasser, but he's not exactly the most knowledgable guy when it comes to computer chess).

So the basic question we have to ask is this: does Deeper Blue (or any computer) actually play chess? I've maintained for years that no computer "plays chess". Computers crunch numbers and manipulate symbols and other data. When you fire up Fritz for a game, the computer isn't playing chess; it's simulating playing chess.

In order to create a better simuation, programmers give computers access to opening and endgame libraries, two areas of the game in which it's possible to play by rote or by concrete principles (for example, look at the King, Knight, and Bishop vs. lone King endgame. Either you know how to play it or you don't. The odds of a player being able to make that mate within fifty moves without prior knowledge of how to play it is only slightly higher than the odds of the Earth being destroyed by a comet within the next couple of years). The idea is to level the playing field for the computer, which cannot learn, remember, or memorize these rote principles and variations the way a human player can (and despite it's massive computational power, no computer can figure out the Bishop, Knight, and King vs. lone King endgame "on the fly" within the fifty moves dictated by the rules of chess).

Is it fair? I don't know that "fairness" was ever part of the equation. Humans and computers can do many of the same things, but they do them differently. I'm personally of the opinion that IBM did the research and built Deeper Blue for exactly the reasons they said they did: not primarily as a publicity stunt, but as a means of legitimate research into the field of expert systems. I'm sure the publicity aspect was considered, but it wasn't the main goal of the experiment. Why chess? Because it's more interesting and accessible to laymen and the general public as an example of an expert system application than, say, a competition consisting of a computer matching diagnoses against a medical expert. There's also the fact that chess has a clearly-defined goal as its object (checkmate) with a specified set of parameters (the laws of the game) used along the way.

And why won't they play Garry again? I don't know about you, but if somebody I beat mouthed off the way he did, I wouldn't want to play the butthead again either.

Either way, both sides have legitimate points to make. No Deeper Blue wasn't playing chess, any more than it's "sons" will be treating patients or creating molecular patterns. But it did simulate playing chess well enough to beat someone who is arguably the greatest practitioner of the game in history, and that's pretty much the bottom line.

Of course, all of this is just my personal opinion. There are many excellent arguments on both sides of the debate than can be included in the limited space we have here.

Don, thanks for the great e-mail. I'm sorry I can't really do justice to it. The questions you raise are more deserving of a book than of a few short paragraphs, with much space being devoted to both arguments.

A question I've been hit with a few times: how do you use a database with both Fritz and ChessBase? Both programs use the same data (with the caveat that you have to use .CBH format data if you're using ChessBase Light). You can have both programs use the same database. For example, I can have Fritz analyze a game in the database ALEKHINE.CBH in a folder called \OPENINGS on my C: drive (e.g. C:\OPENINGS\ALEKHINE.CBH) and when I look at that game using ChessBase, the analysis will be visible. You can't, however, have both programs access the same database simultaneously (that is, you can't open a database using Fritz, minimize Fritz, start ChessBase, and open the same database).

Wolfgang Krietsch writes:

"If I have a database with keys for Tactics, is it possible to review the search setting that define this key?"

Meaning that once a user creates an entry in a theme key, is it possible to go back and look at the Search Mask entry used to create that key? Yes, you can. Highlight the key in question, whack [ALT-W], and the Search Mask appears. You can then view the parameters for the key and even change them if you want.

David Schreiber writes:

"Do you have any ideas on how to use Chessbase/Fritz to improve visualization skills? This is one area where I need a lot of work."

I hear you, oh my brother. Visualization is an old bugaboo for many of us. You want to hear the really good news? Your visualization skills decrease with age. About the time that your back and knees start to really hurt upon getting out of bed in the morning, about the time that you notice your sex drive being next-to-nonexistent, you also notice that you can't see seven moves ahead in the King's Indian Defense like you used to. It's a wonder that more middle-aged chessplayers aren't leaping out of windows.

I have two pieces of advice, only one of which is directly ChessBase-related. Try the "Training" display mode in the move window of ChessBase6. Open a game window and click on the second button from the left at the bottom of the window ("Show/Hide Notation"). Select "Training". As you play through the game, only the current move (and any text notes) appear in the move window. Play through the game exactly as you would play through one of your own games -- trying to anticipate the next move(s), plans, and variations for both sides.

The other advice I can give applies both to chess books and games within ChessBase/Fritz. When you come across a variation line, try to make the moves in your head before playing them out on the screen or on a second board. This may seem like rudimentary advice, but it works. Look at the process the way you'd look at lifting weights. Visualization is mental weightlifting. It hurts like the dickens at first, but the more you do it, the less painful it becomes and the better you get at it.

Try using chess books that give you a diagram every four or five moves and read them without using a board. Also, try Lev Alburt's technique from Comprehensive Chess Course: learn to memorize the board without pieces on it. Know the color of every square. Lev suggests playing a game that reminds me of Battleship: have a friend call out the name of a square (such as "e5"). You must correctly call out the color of the square and name all the squares of both diagonals that intersect at the square your friend called out. This technique works better than you'd initially think. Give it a whirl!

Here's one from Phil (just plain "Phil"):

"Regarding Fritz, what good does a correspondence analysis do me if I can't drag it into the game or otherwise save it? If I can save it, how?"

Just highlight the position on which you want Fritz to perform correspondence analysis, activate the feature, set your time controls, and let 'er rip! When the program is done, use either "Save Game" or "Replace Game" to save the analysis to disk as a permanent set of variations within the game.

These questions came from Alejandro Nogueira:

1- Can I "Color my notes" in Fritz 5 as you have told in CB 6.0 ?

2- How do you perform the "Clean up" function to delete "deleted" games on a database in Fritz 5 ?

3- Is there a way to print out diagrams with "graphical" features (arrows, coloured squares, etc) when you print games with diagrams ?

4- Can you (or CB Germany, maybe ?) give a ETN on MEDALS, their interpretation (Codes) and how they can be used for learning, searching, classifying, etc..?

5- I have problems running videos from CB Magazines. My Win95 onfiguration seems to have problems to perform video files with .WAV extensions : HELP !!!

I donīt know how to edit in HTML or send you some funny(!!??) games Iīve lost and won.

Want to tell you another nice guy on the Net is MIG (Michael Greengard) and his articles on CB Germany and at TWIC.

Taking the questions in order:

1. No, and that's a shame, too. Coloring the notes (as described for ChessBase in a previous ETN) makes them much more readable. I hope that Matthias will consider this feature for inclusion in Fritz6.

2. After opening the game list for the database in question, you mark games that you want to delete: highlight the game and click the "Delete" button. You'll still see the game in the list, but it will be "crossed out" (it will appear with a line through it). The game is still in the database, however; you can still open it up and play through it.

To get rid of these games permanently (and I do mean "permanently" -- once you've eliminated them, there's no way to get them back!), just click the "Clean Up" button. Fritz will wipe out all the games you've marked for deletion and then reorganize the keys so they point to the correct games (the key information is indexed by game number, so once games are deleted, the keys will point to the wrong games unless Fritz sorts the keys again).

3. Unfortunately, no. Bruce Pandolfini once asked me this and was very disappointed to find out that he couldn't do it. A workaround is to convert the gamescore to text format (such as PGN). Then you can use the [ALT-PRINT SCREEN] key combination to send a shot of the screen (showing the board complete with colored arrows, etc.) to the Windows clipboard. You can then use a graphics program (such as Paint Shop Pro) to import the screen shot and edit it down to where just the board is shown. Then use a word processor to import your text, insert the graphic file of the board at the place where you want it, and print out your game.

The printing features of ChessBase and Fritz are pretty good, but word processors are still vastly more flexible in what they can achieve. By using a word processor, you sacrifice some time but gain the ability to manipulate text and diagrams in more creative ways than our programs allow. The printing functions of ChessBase and Fritz are geared more toward printing personal hardcopies of games and analysis.

4. I did an ETN on what the medals mean. I'll try to remember to do a full articles later on searching for medals and other uses for them.

5. Egad! Looks like a job for John Maddox. You can e-mail John at jmcbase@intrepid.net with specific tech support concerns like this one.

Anyone who wants to send me games (funny or otherwise) for inclusion in ETN is encouraged to do so. You don't have to send them in HTML format; I'll take care of the formatting. Just fire them off in either text, PGN, or one of the ChessBase formats (.CBF or .CBH). I'll convert the format and include them in an upcoming issue.

And I have to give a quick plug to my pal Mig (a.k.a. Michael Greengard). Ol' Miggy is one of the funniest chess writers around and his stuff is definitely worth checking out. He posts to the German ChessBase web site (http://www.chessbase.com) and to Mark Crowther's excellent publication The Week in Chess (which hits the web every Monday night, around 8 or 9 PM Eastern Time). Mig has a very irreverent outlook on the chess world. He's a hoot and I encourage you to check out his stuff (players with no sense of humor may safely ignore this entire paragraph).

Paul Massie comments on a topic from last week's ETN:

I wanted to add my feeling to your comments about getting publishers to include disks with their books. I have a relatively large chess library, as well as a large amount of software. Although for a long time I would buy a wide assortment of books, these days I rarely buy one at all, since I discovered I simply never open them. Some of the ones I have bought that were particularly useful I then spent many hours transferring the contents into CB format. This is really somewhat ridiculous, since the content of many of these books was originally created using CB. It would have been little additional expense or effort to include a disk, and would have saved me the effort of re-entering it.

I would buy a great many books now if they included a disk with the contents in PGN or CB, so at least in my case the publishers are missing many sales by not doing so. I see many interesting books now I don't buy, but would if they had a disk. I have no idea how many other people are in my situation, but I suspect the numbers are larger than the publishers realize. They're actually missing sales by not including the disk.

No argument with that! I also believe that publishers could actually generate a few extra sales by including the games on disk with their chess books. I think that part of the problem lies in the multitude of available data formats, but using .PGN on the disk (putting the games in a self-extracting .ZIP file, if necessary) would take care of that concern, since nowadays dang near every chess program reads/converts .PGN (and there are some darn fine freeware/shareware .PGN readers available too).

The only argument I can think of against this practice is the relative fragility of disks (get one too near a power source and you get a blank disk, so tell the stockboy not to stack the cases too close to the power box on the stockroom wall). But this didn't stop the publishers of computer books from including floppies all those years...

Finally, this from Ted Summers:

I am a owner of Fritz 4 and I am currently using ChessBase light. I was wondering if there is a utility for changing formats from *.CBF to *.CBH file format and vice versa. In this way one could exchange data between the two programs and keep their notes also.

Ted, so far the only programs I know of that will do the conversion are ChessBase6 and Fritz5. I'm sure someone will come up with a freeware utility at some point. In the absence of that, the upgrade to Fritz5 is definitely worth the price, for "Friend" mode and "Sparring" mode alone (not to mention the ability to convert data between formats).

As the great bluesman Sonny Terry said at the end of his autobiography: "Well, that's about get it". Until next week, have fun!



You can reach me by e-mail with your ideas and suggestions. Hammer me with them! I'm listening!