AI beats top players at Bridge in two-day tournament AI algorithms crushed eight world champions playing the card game Bridge, marking another milestone in machine learning systems becoming better than humans at specific games. Top Bridge players were invited to play against NooK, AI software developed by French startup NuukAI, in a tournament over two days in Paris. They …
Also, I wonder what other parts were left out.
Left out: All parts that a statistical inference machine cannot do. So, basically anything that resembles intelligence was left out.
The reason for building this is probably the headline you can create for the media. Computer beats Humans at bridge is surely sensational news. Especially sensational when it are top players you can put in the headline. It all is to attract "investors" to sink more money down the money pit called AI. Nothing new here.
I don't entirely disagree, but you might want to be a bit more circumspect about making such a sharp distinction between "statistical inference" and "intelligence" (whatever that means to you, which may or may not be the same as it means to someone else). Modern theories of cognition such as predictive coding and "active inference" (and by theories I do not mean "just theories" - they are increasingly testable, and being tested) suggest that statistical inference may be a key aspect of cognition and intelligence in humans and other organisms.
So while this bridge system sounds pretty underwhelming, I am somewhat more impressed with, say, AlphaGo; there, both human and machine players appear to use "inscrutable" black-box heuristics to strategise. It seems to me that in that case it's plausible that what the human player is doing - at some subconscious neural level - may be comparable (in kind rather than in mechanistic detail) with what the machine is doing. (The human, it's worth remarking, had the benefit of aeons of evolutionary time to "train" its spatial strategising skills.)
The point is that "intelligence" is not some magic sauce - however it's implemented, there will be underlying (neural) mechanisms - and some of those mechanisms may look a lot like statistical inference.
The bidding is the most complex part! Also if they left the bidding out then the card play will change as you decide a lot of your play based on how the bidding went.
It is like saying a robot beat professionals at cricket, parts of the game such as fielding, and batting were left out. That's just not cricket.
Also the statement that bidding " involves trying to trick your components by using various strategies concocted with your teammates." is wrong. Bidding is a system where you communicate complex ideas with very limited options on how to communicate. If your opponents ask your partner what they understand by your bid you have to answer honestly. You can psyche and bid something out of system to confuse everyone, but it is cheating if you and your partner secretly concocted it.
Yes, the bidding is a tricky part, but the play can be based on statistical analysis. I remember one pair who were always messing me around. Anyway, one evening at the Bridge club, I was playing a hand against them, I had a 5 card heart suit and 10 points, so I opened 1H, pass, 2H, pass, pass, pass. Two off. My partner was angry with me until I said I had psyched, and we'd got a top for that hand as the opposition should have been making game in spades. Revenge can be sweet.
I tried to play Bridge for a few years socially, I'm generally good at cards, but Bridge is on the next level, the bidding conventions alone were enough for a two week course, let alone which one does what...
When the machine can beat my friends, I'll put down the newspaper and look interested...
Bridge without bidding is almost like 500, isn't it?
I used to play rather a lot of 500 at uni. Well, it was either that or go to lectures.
I once (bid &) took 10 hearts without either bower.* My partner (a serious bridge player slumming it with us plebs) was apoplectic when they both came out on my joker a coupla tricks in ("#%@!&$% I AM NEVER PLAYING WITH YOU AGAIN!!"), and my eventual win rendered me utterly beyond the pale. He stormed out, didn't speak to me for weeks.
Bridge without bidding!? I reckon I could 'ave that AI!!
* = 100% of all tricks, while missing 2 of the 3 top trumps. Required "a bit" of finesse.
As has been said before, the bidding is the skillful bit. Poor excuse to say that everyone has their own rules, but I believe when you play Bridge at a club, it is expected that the conventions used by the different participants are declared before play starts. On that basis, you tell the AI that the rest of the table are using xyz conventions, which is a doddle* for an AI system to implement. If play proves that someone doesn't properly apply a convention then that is a similar situation to someone who doesn't put down the rational card when playing, tricking the AI (or human player) into making wrong assumptions.
Question: Does AI have a record of the bidding before play starts? If not then that puts it at a severe disadvantage. The first card to be played can be the deciding factor in a game. If yes then the AI needs to know all the conventions in use anyway, as if the opponents are using say, Blackwood to work out slam bidding then the bidding will be grossly misleading without knowledge of the procedure. For example a 5 diamonds bid doesn't mean that they are strong in diamonds, it means that the bidder has one Ace.
On the surface such conventions might come across as completely irrational, but it is down to minimising downsides if cards are not in the correct place, and as such, it is unambiguously signalling the content of the bidder's hand.
It could be that AI was successful where the convention side of the bidding appeared to the layman to be rational and not muddied by analysis. In other words, hands where cards (points) are evenly distributed among the hands, and the overall advantage of any hand over another is merely a King or Ace. Bland, unexciting games.
IMHO the absolute knockout task for AI would be to devise a perfect Convention which would cope with all bidding eventualities assuming everyone uses that synthetic convention. Now come back when you've written an app that does that...
*Ok I'm being facetious here.
As Bridge is played in teams each player is only as good as their partner.
If the AI is teamed with a human it's hard to disentangle if the AI or the meatsack contributed most to the win (or loss).
If the AI partners another AI then it's just a matter of working out hundreds of rules for the bidding process - don't even need "AI" for that, a huge logic tree would probably do the job. No pair of humans could ever hope to memorise a fraction of the quantity of discrete rules a computer could manage.
"As Bridge is played in teams each player is only as good as their partner."
There is more to it than that.
When playing Hearts at a long-ago job, my favorite partner was J___. She was probably the least smart person playing, but I always knew *exactly* what her bid meant, and when she played a card I always knew *exactly* what she had left. As opposed to K___. He was probably the most smart person playing, but he used a little too much imagination when bidding and playing.
Similar situation when playing tandem chess -- that's where a team alternates moves on their side without conversation. It's more important to play a move that your partner will grasp the followup than to play the absolute "best" move. It was always amusing when on the other team player A would make a move, player B would take it back, player A would make the same move *again*, and a fist-fight would ensue.
If AI is "better" than people playing Bridge are we going to just abandon all Bridge tournaments for people and just setup BridgeToc for all future games? I see AI as being helpful to improve the way that people see things but replacing people with AI would be a disaster.
In 40,000 years will there be a post on Facebook saying that a recent Archeologist algorithm suggests that humans appear to have been replaced by AI when the last Climate Change occurred, almost a repeat of the Neanderthals disappearing too?
So Machine Learning (I cringe at calling it "Artificial Intelligence") can beat us all at Bridge. But can it play Cribbage? Or cook a cheese omelette?
It seems to me that each ML system that's developed is good at ONE THING ONLY. When someone creates an ML that is good at EVERYTHING, then we'll need to worry.
Competition bridge is also duplicate bridge. Here the hands not just bid and played at one table but are played by all tables. A typical game may have 24 hands played at 6 tables. The hands are dealt into carriers at the start of the game and these are shared out among the tables (4 to a table in this case). A typical 'movement' means that each table plays four hands, then sends the hands to the next higher table while the East-West pair move to the next lower table. The scoring scheme for this is quite complex and your bidding and playing strategy will be influenced by factors such as which 'seat' you are relative to the 'dealer' (designated on the carrier for that hand of cards), who the opponents are at your table and so on. What you won't be able to do is 'trick' your opponent while bidduing -- competitive players have what's called a "Convention Card" which lists the responses and conventions in use and in the US, at least, you're required to "Alert" your opponents if you're using an unconventional bid.
I'm pretty sure that a decent AI system can play a hand of bridge -- actually, as users of "Bridge Base Online" will know their Robots do a very credible job, especially at play -- but I think these programmers are going to need some hands on experience rather than a quick flip through "Bridge for Beginners" to get truly competitive.
Never mind all that "you look like you're about to top yourself" malarkey ; how about bloody restoring the ability to search within a date range!!
Since they removed it, essentially anything non-landmark is wiped from rediscoverability within a coupla months. Save all URLs separately or they're gone forever. Web-wise, you're living in an Eternal Now, a rolling memory hole.
I only rarely use Google, but their peer mindshare in searchengine land is so strong that all the others followed suit. Even duckduckgo!
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2023
