Did you start playing chess in the 19th century? It’s been true for a very long time that common positions had been analyzed to death. Until you get to Master level and beyond, the winner of most games is simply the person who’s studied more chess. It’s a complete myth that it’s whoever is most geniuser.
Most games at Grandmaster level finish in a draw
Speaking of bluffing…
In keeping with this thread, today I taught someone how to play chess. We got three games in before he quit due to mental exhaustion and for a rank beginner who had to have the pieces identified and moves explained to him prior to play he did pretty good.
No, I don’t care if computers are better than people who play chess because 1) I don’t play computers and 2) I play more as a social activity (i.e. excuse to spend time with someone I like) than with the notion of winning tournaments.
Have we lost all concept of “leisure activity”? You don’t have to be an expert to enjoy doing something as a hobby or for entertainment.
The thing is, though, at the level I play at (probably about a 1500-1600 level these days; I used to be in the 1700s, but rust never sleeps), it’s still a strategy game.
Sure, a computer can brute-force its way to a win - not just against me, but against anyone - but it doesn’t even know it’s playing chess. Fuck the computer. Is it still fun to play chess? For me, the answer is: yes, it is.
It’s probably been close to 50 years that computers have been better than me at chess, and if it didn’t spoil the game for me then, why should it have spoiled the game for me when Deep Blue beat Kasparov?
I’ve never really understood this objection. Sure, the computer doesn’t know that it’s playing chess, but it would be really, really easy to program it so that it would (much easier than programming it to play chess in the first place). The only reason nobody ever does program them that way is that there’s no reason to.
And to stave off the next objection: AlphaGo, the computer that’s now the best Go player in the world, wasn’t even programmed to play Go. It was programmed to learn how to play games, and then learned Go. Presumably, it could also learn to play chess or other games, and the smart folks at Google are probably training it on those other games right now.
I just cheat. If I make a horrible move, I just go back fix it.
I have not played go, but I read the human opponent lost 4-1.
I have played chess. Right now the top computer programs have the edge, but I suspect the best way to beat a computer is to play for an initial draw opening. Sometimes in chess not making a move is best.
In the mid game offer the computer two fronts of attack. One that looks a little weaker but is not, and the other that looks stronger. The key would be to position both knights, the hardest piece to predict in chess in such a way where they can tip the balance to one side or the other when the tables are momentarily turned.
Just a theory.
I would be interesting to see two top computer player each other.
I’m decent at Stratego, and there is some luck in the game as you don’t know the value of many pieces until they are revealed, and navigation of your pieces, once the board is set up, can be problematic. Its a rather crowded game, and one with slow moves and less computation in comparison to chess, as you really can’t plan ahead several moves until you know the value of the pieces.
Essentially if your spy takes out the Field Marshal ( #1 ) , you should be able to win the majority of the time, as you only need to keep your Marshal back and use the General ( #2 ) in all out attack mode backed up by one of your Colonel’s ( One of two #3 ) until you find the spy or take out the other general on a trade.
I have beaten the computer this way on the highest level setting over 100 times to just a few losses.
Modern AI’s exist and compete in various tournaments including the Computer Stratego World Championship, but are currently no better than an intermediate level human player, so HA!
Chess if you think of it is really a perfect math game, but the world we live in is never prefect.
You might be looking at the old matches. In the latest matches in May, the top player in the world lost 3-0. The current version of AlphaGo (called Master) is undefeated against humans. AlphaGo beats Ke Jie again to wrap up three-part match - The Verge
Google also released many games of AlphaGo playing itself. The pros are analysing these games, but some of them are very difficult to understand.
Maybe. But you’ve dropped a pretty big open “if” in the middle of your post.
How do you ensure your spy takes out the computer’s Field Marshal before his spy takes out yours? What strategy do you use that the computer is unable to duplicate?
Locate his #1 Do not move yours.
Place your #2 piece on the front line with the spy directly behind him. Attack with the #2 piece. Once the computer moves toward #2 piece and shadow him with the spy and flank him with a mid-level piece ( say a #4 ) to test what’s out there before attacking.
The computer in most cases will attack your #2 piece with their #1, as its, a logical move, not knowing the spy is in striking range behind him.
You can also retreat with your #2 if the computer chases him ( as its a #1 in most cases ), and have a spy wherever ready to strike.
On occasions where the computer does not move his #1 Marshall, just wreck havoc with your #2 piece on that side of the board with the spy next to him.
This works most of the time.
Alternatively, if you locate the computer’s #2 piece you can attack with your #3 on the opposite side of the board and shadow him with a spy and mid-level piece. Usually, the computer’s #1 piece will go after him.
Stratego isn’t all math, it’s strategic thinking, deception ( as you don’t know the value of the pieces unless they have attacked or been attacked ) bluffing, and intuition. A good initial set up can swing an outcome of a game. It’s not chess where all sets ups are equal and viewable. In Stratego, some fronts are better than others.
No computer program really can understand all parts of the above, but it would clean house if all pieces were discovered and revealed. But Stratego just doesn’t work that way, and with several bombs and an unknown value of the pieces.
An experienced player never moves some pieces, keeping the opponent guessing if it’s a bomb or not.
Like I said, a seasoned mid-level player beats the best computer programs at this game.
A few other tips for those who like to play.
Keep your #7’s behind some bombs.
Place the flag in a back row flanked by bombs in front and on the side.
Never ever move the back corner or two pieces touching it on either side, unless needed as the back corner piece is the best spot for the flag…I think.
Put the Spy on the opposite side of the field, next to a #2 or #3’s
Put the Marshall away from the #2 and Spy, but next to #4’s
OK, and if a human player used that strategy against you, how would you defend against it? It looks to me like you have a fairly weak computer program that you habitually play against, and you’ve gotten very good at beating that one specific opponent, without developing much skill in the game in general.
Every single one of those parts, computers are already better at than humans are.
I would think it’s a matter of not much computing research being thrown at developing a good Stratego AI, despite the existence of a Computer Stratego championship. It does seem to be the case, from what I can tell, that computers haven’t learned or been taught to play it particularly well yet.
More than that. AlphaGo and machine learning can basically learn any game nowadays, given enough training matches.
I’m decent at the game. I should be I played it over 1,000 times. Like I said a mid-level Stratego player is better than a computer and it has been proven.
Vs Human players, if they use that strategy on me I would not take the bait. There’s a lot of bluffing and chasing of pieces.
The Stratego is an imperfect game where you don’t know lots of values. This is why a computer, who’s only good with known values will lose a best of 5 series to the better human players.
This is absolutely not true. There are numerous algorithms that deal with uncertainty. It might very well be true that there’s no world class Stratego AI, I have no idea, but it isn’t because of that.
So I did some poking around. There’s about 1/10000th the number of papers on Stratego AI than Chess and Go. So I read a few of them and it seems that the poker is the preferred game when doing research on decision making under uncertainty. That makes sense, poker is simply more well known than Stratego so if you’re going to talk about something at a conference or publish something to a journal it is helpful to have a well-known context. Now, is this lack of interest definitely why there’s no world-class Stratego AI? I can’t say that, but it seems at least possible. Keep in mind, that this is based on about 10 minutes of searching and skimming through articles (mainly introductions and references) so there’s at least a moderate chance I’m wrong about why Stratego AI is not at the same level as other games. But I can say with certainty, that it has nothing to do with an AIs inability to deal with decision making under conditions of uncertainty.
It’s not an objection, just a statement of fact. The computer doesn’t know it’s playing chess.
You could program the computer to tell you that it’s playing chess, but the computer still wouldn’t know that it was playing chess. And that will remain the case unless and until ‘true AI’ comes to pass.
You could program a computer to recognize a variety of different tasks, all of which it is capable of undertaking, and to tell you which of those tasks it’s currently working on. If there’s any meaningful difference between that and what a human does, I’m not sure what it would be.
This is really going to come down to how you want to define “know”. There’s two ways to think about what it means to know something, one is centered around awareness and the other around recognition. If we’re thinking about this in an awareness sense, I might tend to agree with you mainly because awareness is so enigmatic. If we’re talking in the sense of recognition, then Chronos is correct. Recognition would be not too difficult, and I’m not talking about hard coding it, I mean you could train an AI to recognize and differentiate certain tasks, with various caveats of course.