twixt bot?

i would like to play a twixt bot if there is one. thanks

The twixt bot is named TwixtBot. I think you can click on his user page and send an invitation to start a game.

Hi all,

it seems that "TwixtBot" is the baby of

Janik Steinhauer - and was developed in

his Master Thesis project at Maastricht

University.

The thesis can be found

here.

The bot is based on Monte-Carlo tree search.

Ingo.

Interesting. In this thread the TwixtBot claimed he was written by Jordan Lampe.

But can a robot really know such a thing definitively??

You can play T1j off line if you have Java. It's not very strong, but row handicapping is implemented.

I have to excuse for my wrong "information".

TwixtBot is NOT a program by Janik Steinhauer.

In Steinhauer's thesis he mentioned a short

test match between his bot and T1j. This match

was won by T1j by 3-1.

The reason for my misinformation was a google search.

I had entered

"ai twixt player" 8.0

The ouput were exactly 3 hits: two on the LG bot,

the other one for the Steinhauer thesis.

Sorry again, especially to Jordan Lampe,

Ingo.

Ingo, maybe TwixtBot was made to _think_ it was programmed by someone else? Hmmm.

Seriously, I'm sure we'd all love to hear more from the author on the algorithms used.

And by the way, TwixtBot did eventually accept my test invitation, so that _does_ work.

thank you

i sent invitation and twixtbot accepted pretty quick. thanks again for your replies.

I was just wondering, is anyone using a program to help them play twixt? I'm ok with using stats from past games (like Alan's done, especially since he made them public) but I don't think I like the idea of someone using a computer to help them determine the probabilities involved with each possible move selection. How do others feel about this? Also, would someone admit it if they were playing with assistance?

Everyone is just using Alan's Commentator to play through some moves. There might even be some player's using data bases. However it is very unlikely a program exists that has any chance against an intermediate human player. You can't be sure nobody is using a program, but if they are using a program it won't help much. So TwixT is one of the safest games. If you are interested in game AI's you may also have noticed that there was never a TwixT tournament at the computer olympiad, although they accepted TwixT programs. I think TwixTBot's performance is already very impressive, still he has no chance against an intermediate player.

In short words: you may fear 'man in the middle', but to fear computer programs in TwixT is just paranoid.

Thank you for the come back (trucker lingo) maraca, but I was just curious (not paranoid), it seems to me that most good twixt players are also very computer literate if not complete wizards and I'm not, I just like playing games. So, it stands to reason (from my point of view) that someone could be using a program that might give them some advantage. So, I guess I just hope it isn't true.

Yes you are right that many TwixT players know how to use a computer. I think it's mainly because the requirements needed for computer science are similar to those needed for TwixT (pattern recognition, logic thinking, good imagination,...) But that doesn't mean people coming from other fields have to be less successful.

You're probably right, maraca, that there's no program that can outsmart an intermediate player, I assume because you can not ever know for sure what your opponent will do, but couldn't a program work based on what would be the best move or moves your opponent could make? This assumes a program can calculate the probabilities involved with any move and I'm not sure that's possible, but if it was, my concerns would be valid?

PS. This may be premature, but congratulations! I look forward to playing you somtime (maybe soon).

CSM, if a program could accurately assign winning probabilities to moves then your concern would certainly be valid, and a very strong computer player could be written.

The very large search space saves us from that. Consider if there were only 64 possibilities for the next move. To search just 7 moves ahead would require trillions of board positions to be evaluated. In twixt there are many more possibilities for each move, and usually many moves until a full win.

A practical twixt computer player must instead rely on other techniques, such as start and end game databases, heuristics that recognize patterns in the game, and random techniques such as monte carlo search, where random games are played from each position, and a success probability calculated from the percent of games won.

We haven't heard yet what techniques the TwixtBot uses.

I agree with jds.

You could write a computer program that evaluates local situations perfectly, but to know the global value of a move is very hard. Don't underestimate the human brain here, it's extremely strong in pattern recognition. If you look at a picture, then you can easily reconstruct the 3D scene in your head, for a computer program this is a very very hard job. So a computer program has to consider many more moves than a human, because it doesn't know which ones are worth considering and can therefore not evaluate a position deep enough. You could also maybe detect winnings some moves before they actually happen, still I think the human will be better at this in the near future.

Monte Carlo has some problems too in TwixT, I tried it myself and found that if you do standard Monte Carlo 1. the probabilities were in a really narrow range and lot of fluctuations 2. Normally the spots for the shortest connection (ignoring the opponent) were getting the highest probabilities. Maybe this could be fine tuned and it's just the limitation of my capabilities, but for me it is enough to say: programs won't have a chance in the near future.

What you also have to consider: Given someone writes a program stronger than humans, would he...

A: beat the crap out of the poor online players

B: make a scientific publication

C: try to sell the program

D: try to get attention (for example to get a good job)

E: try to make money in some other way

Looking at this I conclude that there's a very high probability we will recognize if there exists a program stronger than humans.

You should not feel too sure that there

are no strong bot-helpers around in TwixT.

Remember 9x9-Go:

Starting in 2002, there is a strong 9x9 Go

"player" here on LittleGolem: sm9

In those days the go bots were very poor, but

sm9 (being a human+bot-team) was able to win

the second 9x9 championship:

http://www.littlegolem.net/jsp/tournament/tournament.jsp?trnid=go9.ch.2.1.1

Ingo.

You can't compare 9x9 to 24x24 ;-)

The reason is that the problem of finding a winning move is NP hard I guess. This means making the board only 1 bigger leads to an exponential or even worse increase of time.

Thank you jds and maraca, I will sleep better tonight. :)

twixt bot does not accept draws? i have a game that is drawn, i have offered a draw, but so far the board keeps filling up with more moves.

game #1271751

Neat game! I see twixtbot has two draws, both under 100 moves, so it must accept a draw at some point. did you try to send a message? maybe the bot's master needs to be aware of the draw and can step in and stop the game...

As far as I know the bot looks if both player can't win and then it's a draw. In your game the result isn't 100% clear yet.

Sorry you are right, I missed a link, it is 100% drawn game.

the game has become a mess of random moves just to make a move, every move i make i offer a draw. i was hoping the game would come to an end after 150 made moves, but its now gone past that.

Maybe the Twixtbot will accept a draw once Black's potential WIN is ruled out - using links to connect M12 and O10. Right now, that's a winning possibility.

Oh - hey, don't get so down on the bot, guys. It's beaten me 2 out of 3 games! It is a useful learning tool, and certainly better than T1j (which I also play - and can also beat me, but not so often).

(meow)

Hi Aaron,

I have a proposal which may save your day, although

it goes slightly off-topic.

Several years ago, I had designed a board-game named Documenta,

a derivate of Domineering. I wrote a rule set on it for

"Zilliions of Games". Unfortunately, Zillions (especially Zillions I)

was a very weak player at Documenta. So, when playing against Zillions,

I tried to reach two goals: to win and to create a final position

that looks like good modern art.

Look here for an example:

br>href="http://www.zillionsofgames.com/cgi-bin/zilligames/submissions.cgi?do=show;id=312">Documenta - Fine Art

Perhaps, you can try the same in your game against twixtbot:

not only getting the draw, but creating a final position

that looks nice.

Ingo,

wishing all the best for 2011 to everyone here on LG.

Sorry, somehow the link did not show up as intended.

New try:

Documenta - Fine Art

@MisterCat: we haven't said anything bad about the TwixtBot yet, I already said I think its performance is very impressive for a bot. M12 can't connect to O12 if I see it right, exactly in the same area I was missing a link too before.

@Ingo: Nice idea, but I guess in this case without intervention the board fills up completely.

@ the current situation: My guess is that the bot uses some divide and conquer algorithm which might think black can still win, because the chain of white is divided in more than 2 sub parts. I would just write a message to the bot and if those don't get forwarded to the programmer, write a message to the programmer.

maraca wrote:

> @Ingo: Nice idea, but I guess in this case without

> intervention the board fills up completely.

Yes, it may fill completely. But you - or more precisely

Aaron C - decides about the black-white-pixel picture

emerging from that fill-up.

Ingo.

What I write here, seems of small immediate

value for AaronC and his "eternal" game.

But, I got antoher idea: Connect6 and TwixT

are similar games in the sense that draws

can occur but do so only very rarely.

In the Connect6 community a serious proposal

was made to eliminate draws: players are allowed

to make pass moves. When at the end (full board)

no normal winner has been found then the player

who passed first is declared winner. (If both

continue placing stones until the very end then

the player who can not place because the board is

filled is winner.)

Ana analogous rule may be added to Twixt:

Pass moves are allowed.

In case of a game end without side connections

the player who passed first is winner.

Ingo.

PS: Perhaps such a rule had been proposed already

earlier.

This latest eternal game was due to a bug with me not updating the list of "impossible links" correctly after the swap move, hopefully fixed now. I also have a slight improvement in strength, so I updated to version 8.1.

@Ingo,

With your rule, imagine a situation where player A must make move X to draw, otherwise he loses the game. So A plays X, and now B passes, and A ends up losing based on this rule. Is that desired?

Hello Twixtbot,

first of all thanks for your nice bot,

and for letting it play on LG.

> Is that desired?

Yes, that is fully intended.

In this situation player A had

to make the/a last defensive move.

So, he had the weaker position.

Cheers, Ingo.

That would change the object to preventing your opponent from completing a path. Your own path is irrelevant. Compared to that, I prefer the possibility of a draw.

Dvd Avins wrote:

> That would change the object to preventing your opponent

> from completing a path. Your own path is irrelevant.

Formally, yes.

And one of the advantages of Twixt (and much more of Havannah)

over Hex is in my opinion that creating own "structure" and

preventing "opponent's strucuture" are not equivalent.

> Compared to that, I prefer the possibility of a draw.

I am not sure, what I would prefer in the long run.

However, I would like to see a skilled Twixt player where

preventing opponent's connection makes a rather different

game compared with "his" normal style.

Ingo.

And how would this "case of a game without side connections" be defined, exactly? A large part of the appeal of Twixt, IMO, is the simplicity of its rules.

DJ Bush wrote:

> And how would this "case of a game without side connections" be

> defined, exactly? A large part of the appeal of Twixt, IMO, is

> the simplicity of its rules.

It would be the traditional rules of Twixt, plus one

additional sentence: In case of a game without connecting

winner the one who has passed first, is winner.

This sentence might imply a lot, but the rules -

as a text - would not have changed much.

ingo.

At the beginning of the game, there is no connecting winner, as far as I can tell. So the first move is pass, and the first player wins. You don't need to convince me this is silly. You need to define what you mean by "the case of a game without connecting winner."

I don't like the proposed rule, but I can explicate it: When it is a player's turn, the player may pass. If neither player wins the game as already defined, the player who first passed will be declared the winner when either both players pass consecutively or all the holes have been filled with pegs.

That is different, and at least somewhat more strategic, than an alternate pass-to-win rule. The alternate says that a player may pass, ending the normal sequence of the game. At that point, the other player may place pegs in any empty holes and place any bridges that are legal based on all the pegs he then has on the board. If that other player then has a winning path, that player wins. If there is no winning path, the player who passed wins.

The first rule allows a player to pass and ultimately win if s/he can afford to give up a tempo, even if there are still threats on the board that must be responded to should the opponent pursue them.

Dvd, I understand your second proposal, which I believe is closer to what Ingo had in mind. The move "pass" means you make no more moves in the game. If the other player can now win with unlimited moves, that player wins, otherwise the player who passed wins. This might be a relatively simple mod to the rules, but it still seems like a lot of bother for questionable gain.

I apologize for misunderstanding what you meant, Ingo. And I apologize to all for wandering off topic.