Hi all,

> Why is computer chess so much more popular than computer backgammon?
one things could be the smaller community. I don’t know how it is in other 
countries but in the German BG Federation are around 400+ members (and maybe 
1000 people in total  playing on a tournament every now and then) and about 
90,000 chess players.

> 1. A Connection Protocol
> 
> Chess has UCI and CECP. Both are ugly and not very nicely documented. It 
> should be really easy to come up with something better.
I completely agree. Some might know that BGBlitz can use GnuBG as a complete AI 
replacement from 3.2 (if installed). I did this as a proof of concept that an 
external AI works.
Naturally parsing cli output is not very elegant ;) so after release of BGBlitz 
3.4 (soon) I’ll will do a more performant and elegant API so other AIs can 
connected to BGBlitz as well. The API is naturally open, with no dependencies 
to BGBlitz. I’m confident that at least one AI will plug in. I hope that helps 
that the BG-AI scene is not that anaemic. 

>  
> Apart from that, nothing else looks like a massive challenge to me, neither 
> technically nor financially, at least for a team of motivated developers. 
> Point 6 is probably the biggest challenge but my impression is that it should 
> feasible for a couple of developers with a solid DevOps background.

I agree with most of your points, but you haven’t mentioned the biggest 
obstacle next to the missing UCI equivalent. 

I feel there are 3 common believes:
1. XG is the best bot. It solves the game and is always better than any other 
bot
2. if XG errs, it is only by a small margin and in this rare case
3. a rollout fixes this.

All 3 are wrong. 1 is probably true when you add the words “on average” and 
even then it is very close (GnuBG and with the new BGBlitz AI as well. A more 
reasonable measure would be where bot A is better than bot B)
2 and 3 are just plain wrong. XG has no idea of outfield primes, it can’t 
handle containment games (mitigated partly by gigantic search filter) and it 
rejects sometimes correct moves in situations with very many moves.

But as long as most people in the community believe in 1-3, I don’t see an 
incentive to develop something new.
Unless BG explodes as in the 70ies commercial companies will be reluctant to 
invest at least 1-2 million. You have galaxy, and we (at least I) have no idea 
what Travis plans with XG (and if it succeeds. People write SW, not money). 
Will it be a desktop? Will it be a server with a monthly fee?
And as an individual you must be completely nuts. Even when you write the 
coolest AI which might take between 6 month and 2 years, invest at least 
another 5 years to build the rest around?

I hope the API will help a bit, but I’m not optimistic.

ciao
Frank
 


Reply via email to