Fotland immediately started on a new go progam, unimaginatively called G2, which was entirely tactics and rules based. So much so that it was 5 years before he gave me an influence function. I was written on a Hewlett-Packard HP1000 minicomputer, a 16 bit, 1/2 MIP machine. As Fotland became a better go player, I became a better go program. It helped that faster machines were available every few years. In 1984 I was moved to a VAX running Unix, and became the first program to run on HP's new RISC computer, the HP9000 model 840, even before the operating system was working.
I made my debut in the world computer go competition in 1987, running on an HP9000/840 borrowed from HP's Taiwan sales office, and took 4th place, highest of any program not from Taiwan. I also took first place in the US computer go championship that year. Based on that result, Ishi Press expressed an interest in publishing me on the IBM-PC. It took 9 months to squeeze me down into 512K bytes and write a new user interface, and I was released for sale as Cosmos, The Computer Go Partner, in September, 1988.
Of course I did poorly in the 1988 computer competitions since there had been few improvements in my playing strength, taking 8th place.
As Cosmos I had a second release, then there was a major rewrite of the user interface, with professional graphics and new features, and a name change to The Many Faces of Go. That year, 1990, I skipped the World Computer Go Championship, but I have steadily improved since then, taking 10th place in 1991, 6th in 1992, 4th in 1993, and second in 1994. As Many Faces of Go I have had two major releases, with a 3rd soon to come.
In spring of 1993 I made my debut on IGS. I now run on an HP9000/725 workstation, with a custom IGS interface. I found that on IGS I had to be able to talk, at least to respond to requests for games. Over the months there has been gradual improvement in my speaking ability, and I have some ability to learn new words and simple facts.
IGS has been a great help to my go playing ability since I save interesting games for later analysis. Soon I hope to be able to learn directly by observing the games of strong players. My most current version is moved to IGS every few weeks and my results are reset. I play about 20 games a day on IGS, but I do not participate in the rating system.
My go playing engine is currently about 40K lines of C, with a joseki database of about 45K moves, and a pattern database of about 1000 patterns that can suggest about 5000 moves. While thinking I spend most of my time reading groups of 4 liberties or less to see if they are captured. I only look at about 5 to 10 moves in detail before selecting one. I use graded go problems for beginners for test problems and get about 95% correct from volume I and about 85% correct from volume 2.