• The Ing Chang-Ki Weiqi Educational Foundation, and its representative Mr. Yang, for sponsoring the event
• Hendon College, for allowing the use of their up-to-date facilities
• All of those who helped to organise the event:
  • Many of the staff of Hendon College
  • Nicholas Fortescue
  • Kirsty Healey
  • David Keeble
  • Martin Smith
  • Bill Streeten

The 1998 Ing Computer Goe Congress, or "Ing Cup", was held on November 21st and 22nd in Hendon College, north London. The College generously allowed us to use their facilities, including 32 new PCs with 266MHz processors, throughout the weekend. It was organised by Nick Wedd.

Hendon College is near to Colindale underground station, and I had expected participants to find their way there easily. However London Transport chose this weekend to close the entire Northern Line north of Golders Green, and provide a bus service instead. Despite this, all the participants managed to arrive on time, except for one who overslept.

The proceedings started with a dinner in a Chinese restaurant in Hendon. This restaurant charges customers by height - those under 1.38 metres pay half-price. Two of the three inseis thereby qualified for half-price dinners, the third was just over 1.38 metres.

There were seventeen entrants. The idea of running the tournament as a self-scheduled all-play-all was considered. In theory, each program ought to be able to play two or even three games at once, using the Go Modem Protocol. However, we decided to use a Swiss system, and to fit in as many rounds as possible.

This decision was fortunate, in the light of what I now know about the Go Modem Protocol. This protocol is not nearly as robust as it should be; or at least, the programs which claim to support it do not support it robustly. Tournament Organisers should not assume that the Go Modem Protocol will speed things up. In one case, two programs both supported the Protocol, and were able to use it to play against each other, but it took their operators more than half an hour to persuade them to do this.

The Competitors

1. Silver Igo
2. Hamlet (also known as Hit)
3. Goemate (formerly known as handtalk)
4. Go4++

David Fotland's Many Faces of Go has, in my opinion, long been the best buy for an english-speaking Go-player wanting a program which plays Go and also contains useful teaching material. However the version that you can buy does not play as well as the version which competed in the Ing Cup. This version beat its three leading rivals, Goemate, Go4++ and Wulu on the first day of the event, and confirmed its lead by winning its fifth-round game against Gogol, scheduled for the morning of the Sunday, on Saturday evening.

Indeed, had the Ing Cup been limited to six rounds, which was all that could be guaranteed within the timetable, this would have guaranteed the first place to Many Faces. On the Sunday morning, there was some discussion about whether to fit in a seventh round. David Fotland finally, and generously, agreed to there being a seventh round, thereby risking losing his first place.

This risk became more serious when Many Faces lost its sixth-round game to Go Intellect. It bacame more serious still when it was losing its final-round game to Jimmy, and David must have been regretting his sporting behaviour at this point. However his program managed to pull itself together and outplay Jimmy in the yose, thus achieving a score of 6/7 and beating Wulu on the tie-break (SoDoS).

Wulu is a new program, written by a team that works in Chen Zhixing's house, and includes his daughter Chen Guobao. It won all its games except against the overall winner Many Faces, and was placed second. It was operated by another of its authors, Lu Jinqiang.

Go4++, written by Londoner Michael Reiss, beat its long-standing rival Goemate, but lost to both Many Faces and Wulu. Its score of 5/7 was also achieved by Goemate, but Go4++ scored better on the tiebreak my the smallest possible margin, gaving it third place.

Although Chen Zhixing's Goemate is a rewritten version of Handtalk, and Handtalk has won many past tournaments, Goemate seems to be weaker than Handtalk. It lost its games against Go4++ and against David Fotland's Many Faces of Go. I do not know whether these games were unrepresentative - a single game is not sufficient to compare the strengths of two programs. But Goemate is clearly a different beast from Handtalk, and plays more calmly, rather than trying to kill implausibly large groups.

Jimmy's author Yan-Shi Jim, and TeamGo's author Gary Boos were unable to attend, but the organisers had agreed that any program whose author could not attend would still be allowed to enter, and would be operated by the organisers. David Keeble generously spent his weekend operating Jimmy, and saw it almost beat the eventual winner in the final round. Bill Streeten also deserves thanks for operating TeamGo.

Alpha 1.1 was accompanied by a team of three: its author Arkadij Frolov, his manager Valeriy Emelyanov and sponsor Michail Burlak. However it was missing a vital code module. This absence, unfortunately, caused it to crash frequently, and to lose all its games on time, except when its opponent crashed first and could not be reloaded.

361's author David Skidmore will be familiar to historians of computer Go. His program competed in the first ever Computer Go Tournament, the Acornsoft Computer Go Championship held in January 1984.

The Prizewinners

The Insei Games

They all defeated it convincingly. If it had beaten two of them it would have won a prize of 550,000 Taiwanese dollars, that is over £10,000.

Results, by round

Insei 9-stone games v. Many Faces of Go

Details of the entrants

Another representation of the results

Some Comments

• Urge all entrants to read Advice on Writing a Go-playing Program.
• Use shorter time limits. My impression is that most programs play almost as well at ten minutes per game as at an hour per game. It is certainly easier to debug and improve a process which takes a few seconds, than one which takes a few minutes. I believe that encouraging programmers to make use of a full hour on a pentium 450 processor is not the best way to encourage productive new ideas.
• Require all entries to be submitted to the organisers, along with their source code, a month before the tournament - see footnote below.

Footnote

I put these accusations to those accused. In one case, the accused authors denied any plagiarism, refused me permission to send their program to an expert for examination, and withdrew it from the event for technical reasons. In the other case, it proved very difficult to contact the accused, and I have still had no direct communication from them. Eventually I disqualified their program on the grounds that I had not been received it before the deadline which I imposed on all entrants.

I understand that FOST, the organiser of the other major Computer Go event, is also deeply concerned about the issue of plagiarism. I have also heard that the two allegations of plagiarism which were put to me are not the only ones to have been made recently in the field of Computer Go [I now (2000-11-10) understand that any other allegations that may have been made have subsequently been withdrawn. NSW].

Because of this, it is possible that there may be no more serious Computer Go tournaments with prize money. A sponsor (FOST, Ing Foundation, or any other) has a choice between setting aside large sums of money to pay for programming experts and legal fees, and giving itself the power to disqualify entrants arbitrarily. I would not be surprised if they prefer to withdraw from the field altogether.

Nick Wedd

Last updated: 2000-11-10.