[computer-go] The 50-Liter beer bet.

> The 50 Liter beer bet:
> I can write a 1-Dan 9x9 Go programm within 1 year.

>>But even using 19x19 human ratings, if you are the only programmer you
>>have 12 months. Taking into account you will use most of it inventing the
>>first really good FPGA approach, leaving little time for improving
>>evaluation functions and other go-specific knowledge, i'm confident i can
>>play white without komi. I'll offer you a belgian beer of your choice if
>>i lose ;-)
>>Jan (19x19 4dan)

You are right. E.g. my FPGA-Chess programm Brutus has almost nothing in common with my software programm Nimzo. In case of Brutus I got a 1 week intensive course (including some nice flight-instructions in his Cessna) by Ken Thompson. There were also the publications of Feng Hsu about Deep Thought/Blue.

I also think that one has to find new ways for an FPGA Go programm. In FPGAs a very fine grained parallelism is possible. But one has lost if one has to do things sequentielly. In this case one is hit by the >3GHz of the Pentium. The 1 year term is indeed very tight, but it is already difficult enough to interest for a 1 year project an investor. The programm will be indeed not very well tuned. But as a believer in search I assume that copying with the program tactics will be quite difficult for 1-Dan players. They will always reach a better position, but they will also be hit by tactical punches***. I have looked at several high level 9x9 games. These games are extremly tactical. This style should favour a tactical searcher. I think another point is that the humans have to learn how to deal with such a programm. Their evaluation will also be not well tuned against the FPGA-programm.

4 Dan is somewhat too strong. I propose that Ingo Althoeffer is arranging the details of the match and chooses the human team (but first the project has to start at all).

*** Current programms search too. But they search - at its best - in positions, where a human also does some reading (either explicitly, or by knowing the patterns). The main point of a full-search is, that such a programm finds also quite strange combinations a human would never look at. At least in chess even very strong players have serious mental problems to deal with this full-width search effect. E.g. Kasparov has made in the few games against machines more unbelievable blunders than he has probably made in is whole carrier against humans.

Chrilly