| |
BGonline.org Forums
New XG
Posted By: Jake Jacobs
Date: Monday, 28 April 2025, at 8:14 p.m.
I made a jocular post about the takeover of XG, but here's a more thoughtful one.
Don't expect too much.
There have been a number of bots used by the backgammon community. I sense from some comments that many are thinking: "Wow, we are getting AI! It will be like DeepBlue or AlphaGo!" But remember, we have had AI, or at least neural net technology, for thirty-five years.
There are at least five factors which make for a good bot:
1) Price
2) Ease of use
3) Number of features
4) Speed
5) Playing ability
Thought experiment: How many copies of XG have been sold? I assume it is the bestselling of the bots. How many people own a copy? I would be amazed if as many as two thousand people in the US own XG. Maybe I am undercounting the number of players here, but based on live tournament attendance, I would think that the odds of fewer than a thousand copies sold here is much greater than the odds of three thousand. If I am wrong I would be thrilled to learn of my error. The USA is a large country, the most populous country where backgammon is popular. There may be countries with more players, despite smaller populations, but I would still be stunned if XG has sold more than 20,000 copies worldwide.
This would imply that, at $50 a copy, Xavier made at most a million dollars, gross, from XG. How much did it cost to develop AlphaGo or DeepBlue? A search indicates that the latter cost a billion or two, while the former cost something like $35 million. At least prices were trending down. Unless Kalanick is planning to donate development costs, either XG will no longer be cheap, it won't see all that much improvement, or both.
When it comes to items 2-3 on my list, there is probably room for improvement, but XG has already gotten high marks in both areas. It's hard to imagine much more than a tweak here, and a tweak there.
Item 4, speed? The experts can weigh in, but my understanding was that Xavier wrote the software in assembler to make it fast. One's hardware is the key here. Otherwise, adding features is likely to make it slower.
Which brings us to #5, the one most people are thinking about. How much better might it play? Let's assume Kalanick spends fifty million, with no expectation of making money, out of love for the game. How much stronger will it play? And will it radically change the way we play?
I'd be surprised if New XG outplays XG Classic by more than PR 1. The only way that I could foresee a radical "improvement" (quotes around it for a reason) is if New XG develops a package which is really good at the sort of backgames traditional bots have been weakest at playing, AND develops a method of targeting and exploiting that weakness in XG Classic. In other words, it is designed to specifically outplay XG Classic, which is not the same thing as outplaying humans.
Meanwhile, let's suppose that New XG actually does seem to be much better than the best human players, as well as XG Classic, and we see some amazing new moves. For instance, it plays opening 63: 24/21, 13/7. Should you start playing your opening that way? I wouldn't. New XG might know how to play after making that opening, but would you? Probably not.
Anyway, those are some of my thoughts. Possibly all wrong. If so, I trust the rebuttal will be edifying.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.