everything in NLHE is a math problem as well..

yes, but when you speaking of ranges, implied odds...., it's based anyway on a subjective perception.

All he's saying is that computers do a lot better analyzing games of complete information than incomplete information, assuming comparable complexity. Which is obviously true.

ty ...

Bill,

What percentage would you give yourself in a match that was short - less then an hour?

More importantly, have you heard the one about the cider in the ear? More than one person is claiming to be able to beat Snowie, so I'm guessing that they are not wrong...

I personally have seen people do the exploit of Snowie 4 that atrifix is talking about and it is beyond doubt doable.

To really talk about who is better, bots or humans, we need to agree on how to compare skill. Personally I think that the clostest we get to "perfect" backgammon at the moment is heavy rollouts. Let's say we agreed upon that the "true equity" of a position is the average equity of equivalent rollouts on Snowie, GnuBG and XG.

My statement is that no human outplays any of the bots playing at Snowie 4 3-ply or equivalent settings. Interpret outplay as loosing less "true equity". This goes for both match and money game.

In the older snowie a top pro could break even or even slightly beat it over a short session. Human's lose much of their cognitive abilities quickly. After let's say an hour, snowie would be better.

I assume the newest version of snowie is superior. Therefore, I think it would be impossible for a human to beat it long term.

I am interested in Bill's thoughts on short sessions vs the latest version of snowie.

i can beat any software;but i`d need software aid and a fast machine for it.
whoever claims that he has an edge vs BG software errs.or more probable
is setting up a scam.

I'm surprised there's no record on this. Haven't there been any recent public matches between experts and bots?

If not, the posters here should set one up. It seems like we've got a horse race - there are plenty of claims and counter-claims on both sides.

2$ ON DA ROBOTS

If the cube were hard-capped at 16, could any person win? With the 4096 or whatever gnubg allows, you can get blown up an awful lot of times trying to get to one exploitable position where it happily caps the cube with you.

I think Snowie 4 is a favorite over any human in short sessions and a bigger favorite in long sessions, for the reasons you stated.

I'm using "favorite" here in the sense that Hallberg did, namely that Snowie's error rate is lower than any human's in normal backgammon.

There's a separate issue being discussed here, namely whether there are trick positions which, if you can get Snowie into them, will enable you to get a very big cube where you're a huge favorite. That was easily doable against early versions of Jellyfish in positions where one side got all his men back, hit a checker in the bearoff, and then built an outer-board prime. JF evaluated his position as being a 90% favorite when he was in fact about an 85-15 underdog, so the cube would reach his limit (128) very quickly. You could reach these positions about 1 time in 15 or 20, giving you a huge edge overall. The JF people made an ad hoc fix in JF 3.0 by not letting the cube get above 4 or 8 in such positions.

I've looked at a bunch of such positions in Snowie 4 and it seems to evaluate them more or less correctly. However, at least a couple of people here claim that such positions exist, and I assume they wouldn't be willing to wager serious money unless they knew what they were talking about. But as of now I don't know what these positions look like. If I find out, I'll post one here.

They do exist. Generally they involve creating a prime starting around the 18 to 16 point and trapping a single checker behind it. Snowie will beaver these type of positions.

IMHO:

I believe so with a cap of 16.

I am not sure with a cap of 8. Probably not.

Definitely not with a cap of 4.

If you are talking about "normal" backgammon my guess is strong humans tend to play about as well after an hour as when they started playing. They do eventually get tired but it takes like at least 8 hours.

Snowie, GNUBG, XG, are all favorites vs. a human in a short or long session. Long session is better for the bots of course since it gets closer to the long run and the bots never get tired.

This offer would not have been made unless the candidate - who has worked with Snowie constantly - had found it relatively easy to engineer the deep backgame/snake situations that is the basis of the killbot strategy. This is not a general backgammon proposition.

I've looked into the Snowie problem a little more and created a new thread to handle the discussion, "Beating Snowie". Anyone who's interested can look over there.