Specifically, how good is the tutor mode. As far as I can tell GNU set to grandmaster plays a phenomenal game, but sometimes GNU makes different plays than the tutor mode would recommend. The most glaring example is that GNU plays an opening roll of 64 by putting 2 checkers on the 2 point, but tutor mode recommends racing a back checker out to the 11 on the opening roll. IIRC it also plays either a roll of 21 or 41 differently than it recommends. There are other, less common, mid-game rolls where I suspect it does things differently than it recommends.

So is GNU good enough to learn from, or are the inconsistencies too great?

Hmm I wouldn't bother too much about the opening play of the rolls where there are several ways to play them (64, 21 and 32 mainly).

Does GNU recommend making the 2 point unequivocally or does this happen in match play, where the score might force you to play for gammon aggressively?

Anyway, making the 2 point with 64 has been established to be playable by Snowie, GNU and Co. Humans did not feel too good about it prior to that. It also took a while before the BG community accepted making the 3 point with 53 as the best move.

So even if GNU told me that making the 2 point with 64 is best by a slight margin, I still wouldn't do it in money games. I do prefer to play it 24-18 13-9 because it is far more flexible and easier to play than making a point deep in your board.

Apart from that, I think the tutor mode is very good. It is especially valuable to improve your cube handling. When I used to practise with GNU, I would switch the tutor off and let GNU analyse the session afterwards, marking my weak plays so that I could then look and learn where I screwed up (-:

Another technique would be to let the tutor complain but then play your intended move anyway.

Another thing to keep in mind is that some close situations might go either way if you let GNU roll it out. A roll out is always more exact than the tutors suggestion, which is based on a 1-3 ply search and its heuristic evaluation function.

Is this the program in this site?
http://www.gnubg.org/

Yes. It's a bit tricky to find how to install on Windows but this page under the above link has the info you need:

http://www.gnubg.org/index.php?itemid=56

~ Rick

GNU is a fine program, but it is a real pain in the *** to set up positions. Snowie is much better regarding this aspect, but on the other hand Snowie is quite expensive.

Expensive but worth every penny, in my view. Snowie is a really fabulous piece of software. Version 4 was the first one to really play backgames at a world class level and hence permit real backgame study. Its backgame ability, in turn, gives it more credibility in early positions that have the potential to head into proto-backgame situations. Version 4 finally recognized, for instance, that 13/11 6/5 was the best play with an opening 2-1.

Is there any concern that the clear availability of these kind of bots that at minimum represent very strong expert play cast doubt on the fairness of online games?

I would have to say yes.
Keep in mind I'm about a 1450-1550 avg. player, depending on site.
I used 2 play @ PartyGammon when that existed, Also Play65 for play-chippies.

[I still have sneaking suspicions that Play65 launders ducats for either the Mossad or al-quida. loooong story.]

I never really saw any really strong anti-bot policy regarding Backgammon @ the sites.
There wouldn't be any screen-saving involved, you'd just keep Snowie running
in a seperate window. Although i'm not sure if there's a mode in Snowie or GNU [either in the package or hackable] that allows you to manually input dice rolls.
Factoring in the huuuge cash-game activity that existed on both these sites,
and to a lesser extent Oddsmaker these days, yes, serious doubt is cast.

btw, i just noticed this here. Welcome to the show, Mr. Roberti.

$pike

Manual dice input is possible in both GNU and Snowie. However I don't think that it is easy to use either in order to cheat online because it would be quite difficult to handle.

And what would you want to achieve? You could hardly play more than one game simultaneously. So you would have to play high stakes and there you need a hell of a bankroll because the players you'd play there will be very strong, too, so that Snowie shouldn't have a crushing edge.

And if suspicions arouse, e.g. you ripped off a strong opponent, who is used to analyse his games with Snowie. So this player realizes that you played a couple of sessions 100% d'accord with Snowie. He complains and your account containing your hell of a bankroll is frozen. I simply don't believe it's worth a try. If somebody was criminal enough to contemplate cheating like that while having the bankroll, he would probably rather try sports betting accompanied by some match fixing, don't you think?

Also, at the ChessBase online chess server, they have some kind of automatic engine scanning. You're games are run through a test where the degree of congruence to some of the most popular engines is determined. If it is suspiciously high given your rating, they freeze your account right away, with no caution being issued beforehand. It might be easy for backgammon sites to have or at least implement something like this as well.

The funny thing with the ChessBase system is that you read notices of players being banned all the time. However, the culprits are usually weak players around ELO 1200, who I suspect just want to boost their rating so they would have something to brag about.

Wouldnt you just run it on a second laptop? If you were really concerned that someone would see your "perfect" play you could use GNU evaluator and sometimes go for a second (or worse) best move with a slight equity loss. I guess I am just really dubious of playing a perfect information game online for real stakes.

That said, I think learning the game would be extremely interesting, and this forum should be alot of fun for that.

Using the second laptop to enter dice and moves into Snowie and wait for an evaluation is what I consider difficult handling.

In a way you are right about only drawing help from Snowie in crucial spots. However, you must be a strong player in the first place to recognize the tough spots where you would want to have help.

But sure, this might be an issue. I don't know if you're much into chess, but in 2006 there was a big controversy regarding the championship match between Kramnik and Topalov where Topalov accused Kramnik of using engine support. Now, obviously a player like Kramnik is certainly capable of having 90% or more moves d'accord with Rybka or Fritz by his own.

Kramnik countered this by pointing out that in one game Topalov had an 87% accordance. Anyway it is clear that the symbiosis of a world class grandmaster and a top engine would be the ultimate nut to crack, as the grandmaster would know exactly when to use the engine (i.e. check the tactics) and otherwise play according to his superior strategy.

Cheating in online backgammon by using one of the bots is actually pretty easy to spot. The giveaway is the timing of obvious moves.

A good player will play obvious moves instantly, but slow down and think about the moves that are tough. (One way you can tell in live play if an unknown player is any good is to watch what he actually thinks about. If he pauses and thinks about the really interesting moves, he's almost certainly a good player.)

A player using a bot has to pause at all moves, because he's updating the position in his Snowie/GNU window, and he can't afford to get out of synch. That pause over obvious moves is a giveaway.

A few years ago a friend of mine who was playing on GamesGrid called and asked me to watch a match and see if I thought he was being cheated. Although they weren't playing for money, he was being cheated. His opponent was pausing over obvious plays, playing very slowly overall, and playing at a clearly world-class level. Interestingly, as soon as a non-contact position was reached, he would start playing very quickly, and begin making beginner's errors in the bearin and bearoff! He was subsequently banned from the site.

I played a few months on PartyGammon when it was still open to American players, and never saw any instance of cheating in any of the games I played. So I think cheating is possible, but very rare and easy to detect.

Thanks for the replies. As for your questions/comments, Viikatemies:

I wasn't worried so much about the opening rolls, it's just that those are the differences that are easiest to point out. Another discrepancy comes when one player is very far behind. Sometimes GNU will leave the last checker not in the home board on the 7 point instead of putting it in its home board, and will instead move forward checkers within the home board. It will do this even when the player has an outside chance at a gammon. I've never had tutor mode recommend a move like this for me when I'm far behind.

I only play money games against GNU. I wasn't even aware of the "analyze" portion of the software until you mentioned it. As an aside, I've never seen a piece of software more user-unfriendly than GNU. Anyway, the difference is that tutor mode recommends racing the checker out, whereas GNU itself always makes the 2.

Thanks again for the replies.

I think this may be due to tutor mode only working on a level of one ''ply'' whilst grandmaster works on 3 plies. A ply as far as I can tell is how far ahead the program looks ahead in its neuaral network when evaluating moves. The fewer the plies the faster the play. Try changing the tutor to 3 ply evaluation and there will be less discrepency possibly.

An interesting thing i read about w.r.t to the 6-4 opening move and other marginal ones is that some of the pros advocate making worse equity opening moves as it sets up a game which they think their opponent will play worse at throughout this negating this intitial loss. For example moving out one of your back checkers with 6-4 often leads to a racing game whilst taking/covering your 2 point can lead to a prime or holding game which bad players play relatively rubbish at.

EDIT: When you are using tutor mode, I recommend taking it off the ''identify doubtful'' option as this is not particularly accurate/useful at low ply levels and can be disheartening.

There are 2 scenarios coming to my mind where GNU would do this. The first and obvious is that losing a gammon is actually impossible, so that each move would have the same equity (-1.00). Therefore GNU wouldn't care which move it plays.

The second one is more interesting. There are some positions where leaving a checker on the 7-point is correct because it allows you to bear off a checker next roll no matter what the dice, whereas if you moved it into your home board, you could roll a particular quirky one that doesn't bear off. Usually we are talking about a hole on the 4-point here. An example that comes to my mind from scratch is if you have a pile of checkers on the 6,5,2 and 1-point but no checkers on the 4 and 3-point. If you would play the last checker from your 7 to your 6-point, you can't bear-off the next roll if you roll 43. If you have the checker on the 7-point, you can.

Oh yes. Using Snowie for the first time after ages of GNU is a revelation.

downloaded gnu. really enjoying it. enjoying backgammon as well. i played as a kid and had it classified in the Monopoly board game class (hey i was a kid!). never really thought about it as game that can be analyzed.

I did play a lot of backgammon but when snowie got more popular i lost interest in the game a little.
In my opinion and from my experience a lot of the players at higher limit online cash games use snowie or gnu while playing.I guess there are many ways of doing this without getting spotted,i.e. I heard that one guy was paying a student to sit beside him on a second computer and just quickly input all the moves.Another method is to just pause over very important decisions like doubling decisions-they dont update the position every move,they just set it up when they get doubled.It takes maybe 10 sec. to set it up if you are fast,so no problem to do it at all.
Dont get me wrong,I like backgammon a lot and I still find it fascinating and a wonderful game,I can just not recommend to play it online for cash.

If they can't be spotted, how do you know they're cheating? Perhaps they're just good players.

It takes at least 5 seconds to read-input-tell-input for every move which is quite slow for the obvious ones. Also, the length of time for each move will be roughly the same which makes it even fishier. Setting up a macro to auto input/output from the website to gnubg/snowie on an external hardrive would be much quicker and more feasible imo.

I know that there is a lot of cheating going on because I know a lot of the players that do it.And Bill,they are very good players already.
I am sure there are many honest people out there but I would go as far to say that it is considered 'standard' to use snowie from time to time for difficult decision.Thats at least how it is here in europe,it could be that it is different in the US where there seems to be a bit more ethics about this matter.

is GNU free?

Yes

According to this:
http://www.extremegammon.com/compare.asp
the Grandmaster setting is not as strong as 3-ply.

In a nutshell, which settings are the absolute maximum strength for GNU?
And has anyone found a way of saving these settings as the default? Every time I run the program, it goes back to 'Expert' and 0-ply.

Gnu is great and I don't think it's that unfriendly but I've never used anything else so I'm not qualified to comment in comparison to the other bots. Like almost all other programs every feature that you could want to get to is accessible from the top menu bar, if you thinks it's unfriendly go there and explore. It seems to me that all the top pros are using XG and if i was going to spend money on a program I'd pay $50 sooner than I'd play $350-$500.

As far as opening rolls; my version plays 64 in a cash game as 24/18 13/9, it's set to world class, 2-ply evaluation. I don't know why it wants to grab the two point on your version. I think I've seen it grab the 2 in a DMP situation but don't quote me.

Saving settings is easy in Gnu. Open the program, set up your configs and then go to menu bar>settings>save settings.

The world class setting plays a mean game IMO, I've tweaked my analysis to show any move that isn't top pick as doubtful and any 1% or more equity loser shows as very bad. menu bar>setting>analysis. in there you can tweak it to your liking...just remember to save your settings :-)

I'm right with Bill on the picking out a cheater, it's the timing. also, play on sites where you can export game data for analysis. I watched Petkol play a 13 point match yesterday. I recorded and analyzed the match in Gnu, he played with a 3.5% ER and his luck factor was "bad dice, man". That is an extremely impressive showing with those kind of dice. If I play with someone who consistently plays with an error rate under 1-1.5% I wouldn't play him anymore.

BTW- he won

Yeah, I've heard, that supposedly 3-ply on GNU might do some wonky things. I would guess, that supposedly 6 or 7 ply is the strongest setting, but I'm sure that's very very slow.

But really you will be fine with world class or supremo (both 2-ply, which equals XG or snowie 3-ply). Supremo is a bit stronger, but a bit slower.