Monday, March 23, 2009

Highlights from the career of Ed Trice

Ed Trice is an American inventor, artificial intelligence researcher, software developer, and entrepreneur. He was born on December 5, 1966, in the city of Philadelphia, Pennsylvania. He graduated from Monsignor Bonner High School in 1984 at the age of 17. In his senior year of high school he took two college science courses, Chemistry and Physics, while earning varsity letters in the sports of rowing (4 years, 3 years varsity) and football (1 year). In the varsity heavyweight 4-oared shell with coxswain, his boat won the Bronze Medal in the 1984 National Rowing Championship. Ed Trice was inducted into the National Honor Society in 1984. An an electrical engineering major at Drexel University, Trice quickly gravitated towards programming on his 128K Macintosh. He wrote a chess program, named The Sniper, while sick with pneumonia in 1985. Gradually the program was improved, and in 1987 it became a member of the United States Chess Federation (USCF). See this link for The Sniper's USCF member details.

The Sniper won its first USCF tournament in April 1987, scoring 3-0 in the open section of an event held at St. Joseph's Prep High School. It earned a provisional rating of 2207, making it the first ever software program to eclipse a 2200 (Master) performance rating. The Sniper exceeded the 2200 mark for several more tournaments, including the 1989 Pennsylvania State Chess Championship.

Shown above at the bottom right is Hans Berliner, the well known creator of the program "Hi Tech". Hi Tech is shown playing John Mongilutz (who goes by "Jack") while The Sniper (above and to the left) is playing its game against Valerie Face, who lived in Pennsylvania at the time. Both programs won their games that round.

The Sniper enjoyed an undefeated streak lasting its first dozen games before losing first to Stefan Rakowsky and then to Mark Eidemiller both of whom were rated over 2300 at the time. After these losses, Trice made some changes to the program that were not fully tested, and a bug resulted in The Sniper drawing a string of games against players rated in the 1900-2000 range, eventually bringing its rating below 2200.

It should be noted that The Sniper ran primarily on the Motorola 68000 32-bit microprocessor at speeds no faster than 7 MHz. Ironically, in the tournament that The Sniper lost 2 games, it was running on a faster 16 MHz Macintosh SE/30 machine. Such speeds are thousands of times slower by today's standards. A 30-minute search on such hardware would examine the same number of positions as a machine today could do in just one second. In this context, The Sniper was a remarkable achievement.

In 1989 The Sniper also played in the Drexel University Chess Club Championship.

By virtue of scoring 6-0, The Sniper earned the right to play for the title of University Champion in 1989, which caused a stir among the players. It was the first time there was a 3-way tie for first (Trice was 6-0, as was Scott Richter). So, the Championship format was a Round Robin, rather than a set match, for the first time ever. Trice was in the uncomfortable position of having to play against his own program. When he won games against The Sniper, players remarked that he knew the weaknesses of the program better than anyone else, and therefore he had an edge over Richter. Trice's individual score against Richter is what clinched the title for him, so their arguments regarding The Sniper were essentially moot points.

It was also in 1989 that Ed Trice played the chess hardware machine named Deep Thought, later named Deep Blue by the IBM corporation. Deep Blue later went on to defeat World Champion Garry Kasparov in a legendary match.

In between rounds at an event in 1989, Trice told the Deep Thought operator that it had a bad move in its opening book, based on what he saw it play in a prior game. The operator said something dismissive to Trice, who claimed he could win the game from move 7 onward. Deep Thought was set up to play a game in 30 minutes with the position after

1. e4 d5 2. d4 c6 3. exd5 Qxd5 4. Nc3 Qd8 5. Nf3 Bg4 6. Bc4 e6
7. O-O Bxf3
and Trice went on to win the game by move 20:

8. Qxf3 Qxd4 9. Nb5 Qd8 10. Bg5 Qb6 11. Rad1 Na6
12. Be3 Nc5 13. Rd6 Rd8 14. Bxc5 Qxc5 15. Nc7+ Ke7 16. Rxd8 Kxd8 17. Qxf7 Qe7 18. Nxe6+ Kc8 19. Rd1 Nf6 20. Rd8+ 1-0

You can see this game online and by clicking here you can replay it.

In between rounds at the 1990 World Open at the Adam's Mark Hotel in Philadelphia, National Master Alex Dunne was invited to play some informal 10-minute games against The Sniper. After playing 2 games while seated at the computer (without the aid of a regular chessboard), it was suggested that The Snipershould play Alex in a correspondence game. Trice agreed, but suggested the program take the Black pieces in a Pirce Defense. When asked why, Trice said because the program often misplays such positions over-the-board, and he wondered if it would at the longer time controls of (roughly) 72 hours per move.

Alex wrote a 2-page article on the experience in the December 1991 issue of Chess Life Magazine. In it, Alex asserts that the program was playing "Master Level" chess, despite his victory in the correspondence game.

In the 1990's, Trice moved to New York and designed software as a consultant to Fortune 100 companies in midtown Manhattan. In his spare time, Trice released a Blackjack card counting package as "shareware", which relied on users to pay on an "honor system" if they enjoyed the program. His Blackjack Deluxe program and shareware distribution metaphor contained an idea that Trice parlayed into a windfall of profit: any user who sent in a design suggestion that was good enough to be integrated into the program, or who thoroughly documented a bug, would have their names permanently enshrined in all subsequent releases of the program. The result was thousands of great ideas and high quality feedback. The program became feature rich and was extremely stable. People identified bugs on literally every type of Macintosh in existence, and Trice paid attention to all of their gripes. After three years, there were relatively few requests to be made (most were incorporated into features already) and there were literally no bugs left to fix.

It was around this time Trice installed some clever code into the program. It would start displaying "please register me" messages with increasing hilarity the longer a user did not register the program. The program would also start to deal "5's" more and more with each launch that remained unregistered, until the user had an entire 6-deck shoe filled with nothing but 5's. Downloading a fresh copy would not cure this, and mystified users wondered how Trice thwarted all of their attempts to keep free versions of the program. (Later Trice described how an "invisible file" was written to the hard drive. Upon being "seen" even by an unlaunched version of the shareware, the program would immediately "self modify" so that even copying it to another machine lacking the file would result in the same behavior. The befuddled users were left with only one course of action, to pay the $20 fee eventually. Once registered, the shareware messages were removed, the features reset to normal playing, but only if on the same machine as it was registered. Trice required a $20 fee per machine, which prevented illegal copying of the software.)

In 1996 it was voted one of the "Top 50 All Time" shareware programs in the September issue of MacUser magazine. Of the 50 programs, only 12 were games, and Trice's Blackjack Deluxe was given a perfect 5.0 out of 5.0 rating.

In the article, the MacUser staff also mentioned that Blackjack Deluxe was the only piece of software they encountered that ran on every type of Macintosh, from the original 128K Macintosh up to the newest machine released in 1996. For those who might mistakenly believe this was a sparse list, Blackjack Deluxe reportedly ran fine on the Macintosh 128K, Macintosh 512K, Macintosh Plus, Macintosh 512Ke, Macintosh II, Macintosh SE, Macintosh IIx, Macintosh IIcx, Macintosh IIci, Macintosh Portable, Macintosh IIfx, Macintosh LC, Macintosh Classic, Macintosh IIsi, Macintosh Classic II, Quadra 700, Quadra 900, PowerBook 100, PowerBook 140, PowerBook 170, Macintosh LC II, Quadra 950, PowerBook 145, Macintosh IIvi, Macintosh IIvx, PowerBook 160, PowerBook 180, PowerBook Duo 210, PowerBook Duo 230, Macintosh LC III, Macintosh LC III+, Macintosh Color Classic, Centris 610, Centris 650, Quadra 800, PowerBook 165c, Workgroup Server 80, Workgroup Server 95, PowerBook 145b, PowerBook 180c, Macintosh LC 520, Workgroup Server 60, Centris 660AV, Quadra 660AV, Quadra 840AV, PowerBook 165, Macintosh Color Classic II, Quadra 605, Quadra 610, Quadra 650, PowerBook Duo 250, PowerBook Duo 270c, Macintosh LC 550, Macintosh LC 575, Power Macintosh 6100, Power Macintosh 7100, Power Macintosh 8100, Workgroup Server 6150, Workgroup Server 8150
Workgroup Server 9150, PowerBook 520, PowerBook 520c, PowerBook 540, PowerBook 540c, PowerBook 550, PowerBook Duo 280, PowerBook Duo 280c, Quadra 630, PowerBook 150, Power Macintosh 6200, Power Macintosh 6300, Macintosh LC 580, Performa 5200, Power Macintosh 9500, Power Macintosh 7200, Power Macintosh 7500, Power Macintosh 8500, PowerBook 190, PowerBook 5300, PowerBook Duo 2300c, Workgroup Server 7250, Workgroup Server 8550, Performa 5260, Performa 5300, Performa 5400, and Power Macintosh 7600.

When asked why he made sure his "small program" worked on so many different machines, Trice said: "With the exception of the Workgroup Server models, I've received hundreds of shareware registrations on every type of machine, and thousands from a few of the very popular Macs. At $20 a copy, each model it worked on represented potentially tens of thousands of dollars."

This revenue stream allowed Trice to become one of the first software providers to enter the mainstream distribution channel purely from the shareware market.

From 1993-1996 Trice published three software titles distributed through Ingram Micro (MacWarehouse): The Sniper, Blackjack Deluxe, and Poker Deluxe.

At about this time, Trice began playing checkers recreationally on the internet against the very strong program named Chinook (pronounced like Sha-nook). The Chinook program had already won every title that could be held by a checkers program, including the World Champion title of "Man vs. Machine." The clashes between Chinook and Marion Tinsley, the greatest checkers player ever to have lived, has been documented in two books by Dr. Jonthan Schaeffer of the University of Alberta.

One Jump Ahead: Computer Perfection at Checkers (the newest book) and the older One Jump Ahead: Challenging Human Supremacy at Checkers. The new book has an ISBN-10 of 0387765751 (and therefore the ISBN-13 is 9780387765754) and I highly recommend it.

After playing the program every day for 2 months straight, then playing for several months in the Microsoft Gaming Zone, Trice was "immersed" in the game and finally gained the understanding needed to puncture Chinook. Trice became the first player to defeat Chinook twice in one day (see this link scrolling to October 20, 1996 in the Amateur Level and Novice Level groupings) and then a day later he finally defeated the strongest version (Intermediate Level).

Trice was only the 5th person at the time to defeat Chinook on this the highest setting of the web-based version of the program (see the image below, and note the list should be read bottom up based on the dates).

It was shortly after this that Trice was sought after constantly by checkers players on the Microsoft Gaming Zone (most were aware of Chinook and pining for the opportunity to defeat the program and have their own names on the website). Eventually Trice crossed swords with many of the great checkers players from Kentucky, including the world's #2 player, the late Don Lafferty.

The idea of writing a checkers program was too much for Trice to resist. He was eventually put in touch with Gil Dodgen of California through an email introduction by Jonathan Schaeffer. Gil had already written an incredibly strong checkers program that was ranked #6 in the world overall, including humans and other programs, since 1992. At the top of the list was Marion Tinsely (who had been deceased a little longer than a year back in 1996), followed by Chinook, so there was no shame in holding slot #6 behind these two titans of checkers.

Gil had recently partnered with a company called "Cornell Systems" who published his program under the title "Cornell Checkers", but they went out of business. Gil put a great deal of effort into his program, playing it in two National Championships and having it do quite well in each. He retired the program after the Cornell debacle, and he was reluctant to "dive back into" checkers programming. During his first phone call with Ed, Gil jokingly said "If you can beat the program, I'll think about it." Jonathan mentioned to Gil that Ed was an "interested programmer" looking to work on the game of checkers, but he left out the part where Ed had already defeated Chinook several times.

Gil played the Cornell Checkers program against Ed via the telephone, describing where the program wanted to play, and Ed told Gil where he would move in response. After over an hour, Ed finally was able to win the game. Gil was one part shocked and two parts impressed, so they immediately began collaborating on the next incarnation of the program. See the image below on the left.

Released in August of 1997 on both the Macintosh and the PC under the name of World Championship Checkers, it quickly became a very popular piece of software. The name of "World Championship Checkers" was not a claim on any title, as that would be "World Champion of Checkers", so there was some confusion among different groups of checkers enthusiasts. With 2.5 billion endgame positions probed in RAM as it searched, coupled with Ed Trice's personal opening lines that would defeat players of all strengths in the Microsoft Gaming Zone, and some unpublished play belonging to Marion Tinsley given to Ed by Don Lafferty, WCC had no challengers to the claim it was the "strongest piece of software" money could buy.

Former World Champion Don Lafferty penned this testimonial:

Dear Ed,

I wish to thank you and Gil Dodgen for producing the computer checkers program, World Championship Checkers (WCC). I am very much enjoying using the many wonderful features of WCC which make this program by far the best available. The new: extensive library, the size piece data base, the marvelous graphics, the analysis feature, screen capture, much improved cross board play etc truly embrace our grand old game. I let WCC randomly select five 2-move openings - we had 10 fairly quick draws with WCC set on 30 sec per move. The lines of play were sharp, solid & the best I know of. I felt like I was playing Marion Tinsley. All of WCC's features are very easy to use. I highly recommend WCC for all levels of players. Thanks again for this most excellent product.

Don Lafferty
Eastview, Kentucky

After moving back to Pennsylvania in 1998, Ed Trice began looking more at the game of chess again. This time, he recalled that many years before the former World Champion Jose Capablanca wanted to add some pieces to the chessboard to make the game more interesting. Capablanca's board had 80 squares, with two new piece types. One was called an Archbishop, and it could move like a Knight or a Bishop. The other was called a Chancellor, and it could move like a Knight or a Rook. These pieces bring to chess what is has been missing. Namely, the set of all possible combinations involving two pieces fused into one. (Recall the Queen is the fusion of the Rook and Bishop).

Starting out with just what was called the "frying pan set" (extra chess pieces were melted together in a frying pan to make the new pieces)and an 80-square board he had printed at an oversized printer's shop, Ed started touring some local chess clubs and soliciting comments about Capablanca's Chess.

The initial feedback indicated there was "some interest". The fascination with the new pieces soon wore off after it was demonstrated that the player who moved first could build up an unassailable advantage, no matter how stoutly black defended. Ed had to "go back to the drawing board" and see if the fault was in the configuration Capablanca had chosen, or if the game of chess was somehow impossible to extended to an 80-square arrangement.

After testing many different ideas with business partner Andrew Kirkpatrick, it seemed that none of the piece layouts were really that appealing. Taking a week off from work, Ed was determined to exhaustively explore other setups. The hard work paid off, and he was finally left with this arrangement as "the best":

The game was named Gothic Chess to distinguish it from Capablanca Chess. Even to this day, there are some very unintelligent people on the internet who believe the two games are identical. In fact, there are very different.

Ed Trice and Andy Kirkpatrick formed the Gothic Chess Association on October 16, 1998 and later renamed it the Gothic Chess Federation. Soon after producing their own boards and pieces, Gothic Chess began to take off. Tournaments were being held for prize money, and in 1999, there was already one event with a $5,000 payoff advertised in Chess Life magazine.

While the Gothic Chess Association was a growing and thriving business, Ed Trice and checkers programmer Gil Dodgen decided to revisit some of their old source code and embark on another interesting project. This time, they would attempt to solve some very difficult checkers endings by building a large database of "Perfect Play". This would involve computing what the best possible move was in as many positions as possible. A few thousand lines of code and two weeks of continuous processing later, Trice and Dodgen were able to prove that the hardest position to win in the game of checkers with only 5 pieces on the board still required 147 "perfect" moves. While the databases were being computed on Trice's computer the media took interest, especially since a checkers tournament for computers was being discussed.

The topic of what was considered "Perfect Play" had been discussed on a few discussion boards. Three programmers of other checkers software, Martin Fierz, Roberto Waldteufel, and Ed Gilbert, were a little skeptical about the "benefit" of having so-called "Perfect Play" available to a program.

After Ed Trice and Gil Dodgen computed the 5-piece Perfect Play database, Martin Fierz agreed to participate in a test. Martin agreed to take the winning side of the most difficult 5-piece win with his checkers program, named Cake. Ed Trice agreed to play the new Macintosh version of WCC against it, defending the losing position.

The conjecture: WCC might be able to postpone the loss indefinitely using its "Perfect Play" database to continually move into the longest loss. Since Cake would have to search for the best move on every move, it might not be able to find the move that converges on the win.

The position that was used to test WCC and Cake is shown below (graphics from the Macintosh version of the new WCC, which was later named Only Perfect Checkers).

The piece that is highlighted is the red piece that must move in order to complete this win in 147 plies (74 moves for the winner, 73 moves for the loser). Now you might think that 147 plies is a depth that is "impossible" to search, and you would be right. The belief was that this extremely long win would allow Cake to select a move that would not be best, allowing WCC to continually postpone the loss.

There was just one problem with Trice's conjecture: Payne's Draw.

It turns out that a checkers author named William Payne wrote a book entitled Introduction to the Game of Draughts, published in London in 1756. "Draughts", which is pronounced "drafts", is what the English call "checkers". Payne was a sharp player, and he was able to describe a position where with only 5 pieces on the board, the side with the 2 kings can force the side with the 2 kings + 1 checker into a draw.

In the position above, the strong side can just barely win, and the losing side comes close to being able to set up Payne's Draw, but can't quite do it. It turns out that a program can win this very long ending without needing any information about what the best move is on any given turn. All that is required to win is knowledge of which moves lead to Payne's Draw, and by avoiding those, the winning side always gets closer and closer to the win. There are many paths to the draw and few that win. All modern checker programs have every permutation of Payne's Draw stored in memory, so they will never wander into it with 5-pieces on the board if they are in a winning position.

Cake did not win as fast as was possible, but it did eventually win the 5-piece version of the test.

Now the broader question was: Can checker programs always win positions that are theoretical wins? Since the 1960's and the days of the ancient Arthur Samuel program, this has been the operating assumption.

But the answer to this question would have to wait a while.

By late 2001, Gil and Ed also computed the "Win-Loss-Draw" databases for up to 8-pieces of the "non-lopsided" checkers databases (those with no more than one piece difference; for example, four against four, four against three, etc.) This was about 132 billion endgame positions in total. These endgames were much easier to compute than the Perfect Play databases which captured the Distance To Win.

Dr. Jonathan Schaeffer compared the data obtained by the Chinook Programming team with the Dodgen/Trice data. Of the 132 billion endgames compared, there were 89 positions with different results. This meant that one of the teams (or possibly both) had incorrect results stored. The question was, who?

Dr. Schaeffer wrote about this in his updated book, One Jump Ahead. As a strange side note, there was a faction of users on the internet who insisted Trice never was of assistance to Dr. Schaeffer. They consistently posted to discussion boards that Trice was also not mentioned in Schaeffer's book.

Shown above is the index, and clearly Trice is mentioned on several pages of the book. The picture below shows the discrepancy alluded to above, and it also proves that Gil Dodgen and Ed Trice were of some assistance to Dr. Jonathan Schaeffer in his quest to solve the game of checkers.

Dr. Schaeffer has always given credit where credit is due. Even on his website, at this link:

Database Verification

Ed Gilbert, Gilbert Dodgen, and Ed Trice are all identified as people who helped to verify the Chinook databases.

Some of the posters on the internet still believed the remarks of Schaeffer above were too implicit. However, there was no way to deny that Ed Trice was of assistance to Dr. Jonathan Schaeffer after reading this in Schaeffer's book:

Dr. Schaeffer writes:

"With hindsight, this whole incident was amazingly fortuitous. I had just embarked on a massive effort to compute the nine- and ten-piece databases. Gil and Ed's computations discovered an error in my data at the start of the computations, at a time when it was easy for me to fix tings. What if the error hadn't been uncovered until a year or two down the road? Then it would have had a devastating effect on me and my dream of solving checkers."

So Ed Trice did help Dr. Jonathan Schaeffer solve the game of checkers (as did quite a few other people). All of those internet discussion board posters were proven wrong.

In the meantime, Gothic Chess was considered a unique enough creation to be awarded U.S. Patent #6,481,716 by the United States Patent & Trademark Office on November 19, 2002. This effort was the culmination of over 18 months of filings between the law offices of Morgan, Lewis & Bockius and Drinker, Biddle, & Reath.

With the patent secured, the Gothic Chess Federation could invest more time and money into promoting the game without worry that another business entity could come along and compete directly with them. The quarterly newsletter, Gothic Chess Review, was soon tom come out six times per year instead of four. Membership increased from 2,000 to over 9,000 in less than one year. Gothic Chess sets were selling more since the price dropped from $30 to $20 based on buying more at a time.

Tournaments were being run almost every month somewhere, with tournament directors taking their 20% cut of the profits in exchange for all of their hard work.

Checkers programming did not stop during this period. Trice and Dodgen extended their "Perfect Play" databases. After computing the 6-piece perfect play database, Trice reported that the longest win had only grown by about 20 plies. The jump from the 2.5 billion positions in the 6-piece database to the 19 billion in the 7-piece database did pay a much higher dividend. It took over a month to compute the Perfect Play for 7 pieces. The longest win was now 253 plies, 127 for the winner, 126 for the loser. This grabbed more media attention (see above) and this time Ed Gilbert and Roberto Waldteufel were willing to participate.

Roberto Waldteufel had what was considered the 2nd weakest checkers engine of anyone that was selling a program back then. The only one that was weaker was Sage Draughts by Adrian Millet. Ed Gilbert's program was very near the top of the list. Just a few years later, it had become the best program by virtue of computing over 8.5 trillion positions in the 10-piece database and by building an auto-generated computer book of opening moves numbering in the millions.

Ed Gilbert's program is named Kingsrow. Roberto Waldteufel's program is named Wyllie. Each of these programs was given the most difficult 7-piece position to win, and WCC had to defend the losing side.

The longest 7-piece database win was much harder to execute for a variety of reasons. The winning side has many moves, almost all of which win in each position. This means that a search needs to explore more options, and the game tree would be much larger. It also meant that without the benefit of a "Payne's Draw" type of scenario, there wasn't much information to go on. All moves looked about equally as good. The fact that the win was so very long also was a great help.

In the end, Wyllie and Kingsrow were unable to win the game. This was remarkable. Checkers was supposed to be an "easy" game, yet with only 7-pieces on the board, the win was difficult enough to defy what is called a "strong solution."

Ed Trice and Gil Dodgen wrote a paper about this experiment, and the International Computer Games Association published their findings.

Click here to download "The 7-Piece Perfect Play Database for the Game of Checkers"

Computing larger "Perfect Play" databases for personal computers ran into a brick wall in going beyond 7 pieces. Since the longest win for 7 piece requires 253 plies, and a byte of memory can hold only 256 unique entries, each position for 8 (or more) pieces would require twice as much RAM (2 bytes per entry).

The useful portion of the 8-piece Perfect Play database would require 264 gigabytes of storage and a machine with at least 40 GB of RAM. This had to be set aside for a while.

With that on hold, Ed Trice heard that a hardback book was coming out in the field of Computing Science, and they wanted to republish the paper he and Gil Dodgen had written.

The book was called Advances in Computer Games 10: Many Games, Many Challenges which is now also available on (although it is somewhat expensive).

Gothic Chess was again in the media spotlight as the playing population grew to support larger and larger tournaments. Of special interest was the College Scholarship Tournament that was held each year.

For an entrance fee as low as $100 (early registration) any High School student who turned in a 12-0 result would have a fully paid 4-year ride to the college of their choice. Students already in college with perfect scores would have their previous tuition payment reimbursed immediately and any future tuition (based on a 4-year schedule) would be paid by the Gothic Chess Federation. Partial scholarships and cash prizes were also available to 2nd place through 16th place finishers. It became a logistical nightmare to coordinate these events. With thousands of participants, security problems (exceeding the lawful occupancy nearly cancelling one event, people forcing their way through areas delimited by personnel barriers, etc) and several injuries at the Kansas City playing site, the structure of the event was forced to change dramatically. Many smaller local and regional events were substituted, with winners advancing to subsequent rounds and locations potentially further and further from their homes. It was impossible to satisfy the blackout dates of all candidates, despite months of pre-planning, so it was eventually curtailed altogether.

In June of 2004, Ed Trice published another paper with the International Computer Games Association Journal:

You can download this paper as a PDF file by clicking this link.

In it, Trice compares the games of Gothic Chess to Capablanca Chess and Bird's Chess. All are chess variants comprised of 80-square boards that are 10 files wide, with the same 8 ranks that are found in regular chess. In one section of the paper, Trice demostrates mathematical formulae to compute the "true value" of any chess piece on a board of any size. The mathematician Henry Taylor derived similar formulae in 1876 for chess boards that were square, but these were of no use in rectangular geometries. Trice's results were more complex from the perspective of the number of terms per formula, and, in the case of square geometries, they were able to compute the same exact values as the Taylor formulae.

In another part of the paper, Trice shows how Capablanca's Chess was derived from Bird's Chess. Trice also is able to demonstrate why Capablanca changed Bird's configuration in a very logical fashion. So many discussion boards that deal with chess variants have incorrect notions about Capablanca's chess. The most egregious error is the underlying assumption that Capablanca's Chess was created "stand alone" with no prior model on which it was based. The second most flagrant error on those same discussion boards was misnomer that "Trice just changed the pieces a little" from the Capablanca configuration. These same people ignore that Capablanca (1924) did that very same thing to the Bird configuration (1874). If you look at all three boards, you can see that Gothic Chess (Trice) is more similar to Bird's Chess than Capablanca's Chess. Those otherwise intelligent people from chess variant sites are showcasing their own ignorance and shallow mindedness.

In December of 2004, Trice was invited to submit another paper describing the results of the 2004 Gothic Chess Computer World Championship. By this time, Gothic Chess was being played in many different countries. Programmers fielded entries from Germany, The Netherlands, Poland, England, and the United States. There were programs playing that were created decades apart, and therefore represented Artificial Intelligence epochs. CapaGNU chess was an 80-square variant from the 1980's that could be modified to play Gothic Chess. Zillions-of-Games was an engine with a technoligcal foundation in the 1990's. Gothic Vortex, SMIRF, ChessVand TSCP Gothic were all from decade one of the 21st century. Of interest is that Gothic Vortex and TSCP Gothic were both programs ported from existing 8x8 chess engines.

This paper was later published online at the website.

This tournament served to kill several myths in the chess variant community.

The first to die was the belief that the Zillions-of-Games engine was "strong". Again, mostly due to the bias of the discussion boards such as, the strength of Zillions was often exaggerated. The program is impressive in that it is able to play any game at all by merely parsing a "rules file" that conforms to a scripting protocol. This same engine can play chess, checkers, Capablanca Chess, Gothic Chess, a chess variant that is 40x40 with many pieces, etc! But, with a score of 6 wins, 7 losses, and 1 draw, Zillions-of-Games finished in 5th out of 8 places.

The next myth that was killed concerned the strength of the ChessV program, written by Greg Strong. Greg Strong wrote this program to "bridge the gap" between Zillions-of-Games and a dedicated chess variant program. The ChessV program was a dedicated engine that could play many variants also, but it was not an interpreter like Zillions-of-Games. Each of the chess variants played by ChessV were programmed by Gregory Strong. Because the ChessV program was able to manhandle the Zillions-of-Games engine, the same discussion board users exaggerated the strength of that program as well.

While the creation of Greg Strong finished in a respectable 3rd place out of 8, the weak play against the top 2 finishers left no doubt in anyone's mind that it was at least a class or two behind. ChessV scored 0-4 against the top programs, Gothic Vortex and TSCP Gothic.

ChessV did even worse in the 2005 and 2006 Gothic Chess Championships, and it was not entered in later events. Rather than try to improve his program, Greg Strong has since taken to vitriolic attacks of the game of Gothic Chess and Ed Trice in particular.

The final metric that this tournament helped establish was the strength associated with CapaGNU. It was a DOS-based program created in the 1980's running on much faster hardware. CapaGNU was handicapped in two regards. First, it was playing Gothic Chess, and not Capablanca's Chess, therefore any game-specific heuristics were essentially not going to be encountered by its evaluation function. Second, the software advances in searching had been considerable in the intervening 20 years, and it was not able to benefit by this by being frozen in time.

This was the first tournament where such technological epochs were thrown into the mix together. The results demonstrated that more advanced, highly tuned evaluations and hand crafted approaches are able to outperform generic algorithms and outdated software searching strategies.

From 2005 to 2007, Ed Trice tried to bring chess titans Anatoly Karpov and Bobby Fischer to play a 12-game match of Gothic Chess. After raising the sum of $15,000,000 for the prize fund, the match groundwork was seriously underway.

Anatoly Karpov signed the contract to play. All that was left to do was get Fischer to sign on the dotted line.

No comments:

Post a Comment