Google has beaten Go, what could it do to bridge?

[broze]

Honestly I found it pretty funny that this happened so quickly after this announcement from Facebook.

[Fluffy]

Hearthstone AI?

 

Nah, still bridge related :), Hearthstone is too complicated, but I was once interested in AI for urban rivals.

[rhm]

I think it has already happened. 15 years ago, Jack played at about "meesterklasse" level (highest level of the Dutch competition, comprising about 100 players out of an organized bridge population of about 150,000). I don't know how much it has improved since then but the increased CPU speed since year 2000 alone might be enough for it to win the Bermuda Bowl.

 

It is not easy to measure, though. In Go or Chess, just a handful of games against a World champ is enough to give you an idea of the relative strength, and those games are easy to set up. In bridge, it is a lot of work to make the computer understand the human bidding and carding system, and then you need a sample size of several hundred boards to make a reasonably robust verdict.

As explained above much of the skill required in bridge is partnership oriented.

What you can do on your own is only one ingredient of the overall picture. What really matters is what you can accomplish in a partnership.

I would claim that at high levels of Bridge the majority of poor and good results have a lot to do with partnership communication and understanding (or lack of it).

Do you seriously believe if you would play with Jack as partner it would make you competitive several levels higher than what you are currently capable of, say qualifying and taking part in the Bermuda Bowl, because Jack is so much better on certain aspects than the majority of bridge players?

 

I seriously doubt your claim.

Letting Jack play with Jack is mickey mouse in comparison. The real test is what BBO does with its robots.

Show me the partnership of such a robot with a human or an independently developed robot beating anyone else.

 

Rainer Herrmann

[helene_t]

As explained above much of the skill required in bridge is partnership oriented.

What you can do on your own is only one ingredient of the overall picture. What really matters is what you can accomplish in a partnership.

I would claim that at high levels of Bridge the majority of poor and good results have a lot to do with partnership communication and understanding.

Do you seriously believe if you would play with Jack as partner it would make you competitive several levels higher than what you are currently capable of, say qualifying and taking part in the Bermuda Bowl, because Jack is so much better on certain aspects than the majority of bridge players?

 

I seriously doubt your claim and I am not aware of as single experiment where this has even been tried at all for good reasons.

Letting Jack play with Jack is mickey mouse in comparison.

 

Rainer Herrmann

I concede that Jack+Jack vs two humans is slightly unfair because each Jack will be perfectly tuned to this partner while the human opps (unless we are talking of very stable partnerships like deWijs/Muller, the Hackett twins or Meckwell) will have less than perfect partnership understanding. But maybe your point is more that Jack lacks the skill to develop a distinct understanding with each partner.

 

Jack actually can do some of this since if multiple users use the same Jack install, they have to log in with seperate IDs because Jack analyses their style and slowly builds up a knowledge base of each human's preempt style, overcall style, falsecarding style (not sure about the latter, maybe it is restricted to bidding). But of course, Jack is not human-like in this respect.

 

You could play an indy, then, where 99 expert humans (unknown to each other) participated together with one Jack. My guess is that Jack would be very close to, if not above, the World elite in such an experiment. I might be wrong today but then I will probably be right within the next 5-10 years.

[barmar]

As explained above much of the skill required in bridge is partnership oriented.

There are also other psychological features that are hard to program. During the play of the hand, a significant aspect is trying to infer what's in the unseen hands based on how those players are playing. One of the most unique human abilities is called "theory of mind" -- this is when you imagine what someone else is thinking, based on their actions. We do it all the time, mostly without thinking, even as infants. It's generally considered fundamental to our language ability.

 

I've thought on a number of occasions how we might add this to a program like GIB. When GIB does its simulations, it mostly uses information from the auction, and the known cards that have been played, to calculate likely types of hands of the other players. To make better inferences, it would have to go back through all the previous plays of the other players, perform simulations at each step, and determine which hands are consistent with the actual plays.

 

Ginsberg has told us that he tried something like this in GIB, but the computational expense was overwhelming. If he limited the number of simulations so the time was acceptable, the results were not very helpful.

 

If you play against the BBO robots now, you can generally tell when they're doing a complicated simulation, there's a very noticeable hesitation. Imagine multiplying that by 10 or 100 for almost every play, which is probably what would happen if we tried something like that.

 

This is where neural networks would probably do better. Rather than having to explicitly simulate all the possibilities, the network would learn to recognize common patterns and deduce the implications.

[etha]

I doubt Jack would win the BB. It is ok but I doubt it would win a JEC match for example but it would be competitive.

You can get some idea of current robot strength from Wbridge which runs daily duplicate match point events free with 3 players being robots and one live player. Wbridge plays some insane methods which is a drawback, jack is at least changeable enough to can probably find a cc you can stand to play with it.

 

Might be interesting to see how jack or wbridge does as a robot/human partnership, if anyone wants to try let me know I can run wbridge on bbo reasonably easily as it has a one hand mode unlike jack.

 

Jack's bidding is worse than top players, its declarer play is its best ability which is on a par with pretty much the best I claim.

[gwnn]

I wish at some point people replaced the arguments starting with "Computers will never be able to do X because (...)." with "Human intelligence can do X because (...) and this impossible to replicate with computers because (...)." For example, "Computers will never be able to play Go because the possibilities are endless." -> "Human intelligence can bridge over the possibilities of Go because they have pattern recognition and aggressive pruning, and computers will never be able to have those because (???because what???)." This is not targeted at anyone in particular in this thread, just these Luddite arguments about what computers could and could never do seem to woefully fall short in my eyes. I recently saw someone comment on Facebook that a computer will never beat a human at go in the next 500 years (!).

[etha]

Yes I was never very convinced by Roger Penrose's arguements about what machines can theorectically ever do. Of course he had a vested interest in them not being able to do some things human brain's can do given he thinks souls exist.

 

Assuming the brain really is only made up of the stuff we can see without inventing something there is obviously no reason to suppose a computer can't do everything a human brain can.

[StevenG]

Assuming the brain really is only made up of the stuff we can see without inventing something there is obviously no reason to suppose a computer can't do everything a human brain can.

Absolutely. All we need to do now is let the computer evolve naturally over millions of years. Perhaps someone should start a breeding programme.

[etha]

Will it be illegal to keep them in a dark cupboard?

[barmar]

Absolutely. All we need to do now is let the computer evolve naturally over millions of years. Perhaps someone should start a breeding programme.

If you think you're being sarcastic, you're not. There's actually a method of programming called "genetic algorithms". It starts with numerous versions of a program with random differences, and they each tackle a problem. Then the versions that did the best are "mated" and give birth to children, which are new versions of the program with random parts of each parent swapped (analogous to the crossover in biological sexual reproduction) and occasional random mutations as well. This is repeated over and over, which mimics natural selection.

 

It doesn't take millions of years, effective results can be obtained in hours.

[lycier]

http://i66.tinypic.com/33b1laq.jpg

 

 

 

 

According to the TeachWeb,the date of challenge match are 9,10,12,13 and 15 this March.

It is said that AlphaGo owns the ability of self learning at present and with the passage of time, AI can become more clever, so some experts of our country think the first game is key.

 

Google spent $400 million for the DeepMind, there were only 20 people then, now more than 200 people in total, and they invested funds even at all cost.Playing against world Go champion,AI will exhibit its great progress with glorious future. I think " AlphaBridge" will beat world top players in the bridge game in the future.

[psyck]

If you think you're being sarcastic, you're not. There's actually a method of programming called "genetic algorithms". It starts with numerous versions of a program with random differences, and they each tackle a problem. Then the versions that did the best are "mated" and give birth to children, which are new versions of the program with random parts of each parent swapped (analogous to the crossover in biological sexual reproduction) and occasional random mutations as well. This is repeated over and over, which mimics natural selection.

 

It doesn't take millions of years, effective results can be obtained in hours.

 

 

We can play with this kind of stuff ourselves here:

http://rednuht.org/genetic_cars_2/

[hrothgar]

Lee Se-Dol is looking forward to play against AlphaGo.

After determining the date of the challenge, Lee said it is his great pleasure for him to play against artificial intelligence, " Whatever the outcome will be, it would be very meaningful event in the history of Go.I heard that artificial intelligence is unexpectedly strong, but I have confidence to win this time at least."

Many readers of course support Lee. they think the biggest differences between artificial intelligence and human are every step of computer calculation is the best choice, and the layout of human is not necessarily best, but the human is able to set a trap. Of course,many readers strongly support AlphaGo, "Don't look down on artificial intelligence, AI owns super computing power, human can do?".

I will vote for Lee Se-Dol, I think it is impossible for AlphaGo to beat human at present, even AlphaGo have beaten European Go Champion, but compared to Go champions from China,Japan and South Korea, he's too weak.

 

So, looks like Lee Se-Dol lost the first game.

 

Better hope that he wins three out of the next four...

[mgoetze]

So, looks like Lee Se-Dol lost the first game.

 

Better hope that he wins three out of the next four...

After watching the first game, I don't think that's going to happen. To my amateur eye, it seemed like AlphaGo was playing a bit slack in the endgame because it was confident of winning, which is to say that it was not really a close game.

[lycier]

So, looks like Lee Se-Dol lost the first game.

 

Better hope that he wins three out of the next four...

 

After watching the first game, I don't think that's going to happen. To my amateur eye, it seemed like AlphaGo was playing a bit slack in the endgame because it was confident of winning, which is to say that it was not really a close game.

 

 

 

It's miraculous! AI unexpectedly won victory over him twice at present !

After Li se-dol lost the game ,China's Go world champion KeJie said, " even if AlphaGo can overcome lee se-dol, could not overcome him ".

In another word, if AlphaGo only wins Ke Jie, Human beings will throw in the towel.Why didn't AlphaGo choose China's Go champion Ke Jie?

[lycier]

People's Daily(China) reviews that the dignity of thought belongs only to human beings.

[Zelandakh]

Why didn't AlphaGo choose China's Go champion Ke Jie?

I am sure the people behind AlphaGo would love to take him on. Ke Jie himself might be a little more reluctant having more to lose from such an encounter. You can be sure he is studying the computer's play very closely looking for weaknesses.

[jogs]

Chess is suited to computers. Go also but is more difficult.

The problem with bridge is you don't know where all the cards are. Whereas is chess and go you know where all the pieces are.

Bidding has improved considerably but still has a lot to be desired. When chess programs were starting to challenge masters, bridge programs were no better than beginners at bidding.

Chess and go are deterministic/perfect information games. The location of pieces are known.

 

Bridge is a non-deterministic/imperfect information game. Bridge first need to develop a better method to evaluate the fit of the partnership hands. How many tricks can the partnership make in each strain?

[billw55]

I see no reason to expect alphago to achieve a different result against Ke Jie than against Lee Sedol.

 

It is true that bridge includes some aspects that chess and go do not. Most notably, incomplete information and partnership communication. Against this, the game space is just much smaller: fewer decisions overall, fewer legal alternatives per decision, and fewer reasonable alternatives per decision.

 

I expect that bridge could be conquered by computers as soon as someone is willing to invest enough to accomplish it. The funny thing is, in bridge it will be harder to diagnose success. There is substantial variance/randomness, and even a theoretically perfect team will lose often to a very strong, but imperfect, team. Even hundreds of boards as at the BB might not resolve the difference consistently.

[barmar]

I expect that bridge could be conquered by computers as soon as someone is willing to invest enough to accomplish it. The funny thing is, in bridge it will be harder to diagnose success. There is substantial variance/randomness, and even a theoretically perfect team will lose often to a very strong, but imperfect, team. Even hundreds of boards as at the BB might not resolve the difference consistently.

If it were to win a long match by a small amount, that would not be considered decisive. But if it wins a 128-board match by 30 IMPs or more, I think that would be quite significant.

 

The best way to test it would be to have a team of computers competing in a full week-long event like BB or Spingold. So it wouldn't have to win just one long match, it would have to win several of them. There are occasional unexpected upsets in a few matches of these events, but any team that makes it as far as the QF is almost always made up of players who are all considered the cream of the crop. There are rarely any surprising teams there.

 

So I think the way to find out if bridge has been "beaten" is for them to play one of these premiere events from start to finish and win it.

[jogs]

If it were to win a long match by a small amount, that would not be considered decisive. But if it wins a 128-board match by 30 IMPs or more, I think that would be quite significant.

 

 

Do you appreciate the amount luck plays in bridge results? The standard deviation of a board is around 5 imps depending on relative vulnerability. In a 128 board match the s.d. is about 55 imps. Assuming the world champs are 1 imp per board better, the opponent team has about a 2-3% chance of winning providing they don't choke. Computers don't choke.

[lycier]

http://i66.tinypic.com/5nt01i.png

 

Today AlphaGo beated Lee se-dol third time!

 

If AI can beat Go game thoroughly, I am afraid that the next target whould be the bridge.

let's think about the future of the bridge :

 

1- AI team always be a world champion.

 

2- If the AI team doesn't take part in the competition, the human can only hold world bridge second championship.

 

3- All the bridge Champions will shut up, AI master will teach them " How To Play The Bridge". They will be living in poverty since AI will rob their bread and butter.

 

4- All the books of the bridge champions will become a pile of waste paper. The books of AI bidding system writtern by AI will become a best-seller.

 

5- No director. No cheater. AI have great ability to solve almost of the problems on the bridge game.

 

6- If AI come to BBO,I believe AI will design a " self design " function " - all the self design of BBOers will adopt the manner of self research, self design and self debugging, do as BBOers like, and never need BBO to design. On the BBO forums, AI will become moderators, it can teach all the BBFers " What is the correct auction".

 

If AI be SkyNet in the future, where will we go? How terrible! Would you still like AI?

[Antrax]

3. Teaching bridge isn't about telling you the right answers to everything. We're a long way from systems that can explain themselves in a human-understandable format.

4. I'm pretty sure some HUM methods are more efficient and are already banned. No reason to expect the AI to generate something that will be ACBL-legal. Also, knowing how to play a system != knowing how to design a system, though I'm sure you could solve the second one with today's technology too.

5. ??? Do you expect AIs to come up with some revolutionary way to handle revokes, mechanical errors and UI?

6. Computers can't even tell you whether a computer program will run forever or not. Why do you expect them to be able to design themselves?

[sakuragi]

BBO should implement 4 humans vs 4 bots team match, when we are yet capable of winning, before it is too late ... lol