Speaking of other bots. I was approached by one guy asking me if I had any plans in this direction. I think it takes a lot of skill and dedication to develop another backgammon bot and I think I am already quite overstretched by maintaining the live database, Heroes and developing Backgammon Studio 2. Not sure I have the skills for it to begin with either! While AI/machine learning is an interesting subject I haven't worked on anything in that area.
I think one of the reasons why there aren't many new one coming along is the financial aspect. How to make good money on the effort you put into it? I think that is one of the reasons why we don't see XG3 and might not do so for quite some time. The market is just too small. I'd love to be wrong on this one and we'll see an XG3 announcement tomorrow! :-)
What I think is the only viable option is a subscription model like what I do for Heroes. Can this be done for backgammon bots too? Xavier is already doing this with his XG analysis support for Galaxy. Are there other opportunities? One possibility is perhaps to provide analysis to an online position database server...
On Backgammon Studio you can edit a position and then ask for a gnu analysis of the modified position. This analysis could be done with a different bot if one was available.
Here I made a small error and in hindsight I see that I end up with only stripped points which I may have to clear in the next roll possibly leaving a shot.
Then I dragged one spare from the 4 point to the mid point and now the gnu analysis shows that making the 10 point is correct.
How much would you be willing to pay to have XG analyze the position? Yes you can do this already by ctrl-c on Studio and then ctrl-v in XG but that is a bit more tedious. $10 per year? $20? What if you could effectively use XG3 analysis (assuming it exists on some level)? $30/$40? To me that sounds perfectly reasonable. If other bots came along that could also be interesting to integrate with and you could get the opinion from more than one bot and see if they differ much in evaluation.
Why not just use XG directly? You can but you don't have a database of 8 million positions you can search through. There are also other cool stuff like one click on the scorecard length area and you get evaluation of the position at various scores. 1 click. How much effort is it to get the same information in XG?
And a huge amount of other sweet features which doesn't exist in any other product to my knowledge. While I am showing images from Studio 2 most of this is also available on the old server albeit in a bit more awkward manner... It is called progress! hehe
As a pure research project without any financial motivation there could be some possibilities. I did come across an interesting thread about AlphaZero and backgammon some time ago: https://github.com/deepmind/open_spiel/issues/774
Best regards,
TP