5

Computer programs have been produced for games such as Chess, Go, Poker, StarCraft 2, Dota. The best ones, Deep Blue and AlphaGo , AlphaZero, Pluribus,... are now considered better than the best human players. More to the point, the computers' game results have been influencing human play.

Apparently, computers are not yet better than human players in Bridge. There can be computer simulations of various hands and hypothetical opposing hands. But what progress have computers made in playing human players in tournaments? Have any new theories of bidding or play evolved as a result of computer-human interaction in Bridge?

2
  • 1
    I understand that to get the results we got for chess/go a lot of commitment/funding is required. Moreover, the prior result from chess/go are not easily translatable to something else, such as bridge. As a result, you might not see the same level of advancement in a bridge computer player until such time some one with the necessary prerequisites takes interest in it. This is not to say this is not possible with the current level of technology.
    – Andrew Savinykh
    Commented Mar 3, 2020 at 22:29
  • 2
    see en.wikipedia.org/wiki/Computer_bridge
    – Cohensius
    Commented Mar 4, 2020 at 11:06

1 Answer 1

Reset to default
4

There's a fundamental problem that the rules of bridge are not well-defined for computers. Specifically, the rules require that partnerships communicate their agreements - including implicit agreements arising from experience playing with their partner - to their opponents. No one has really figured out what it means for a computer to communicate its partnership agreements to other humans or to a different computer program.

(Note that it does happen in human bridge where one partnership explains their agreements to their opponents, there is misunderstanding, and a director or an appeals committee has to sort everything out, so these rules are not perfectly well-defined for humans either.)

This is less of a problem if the computer is playing a human designed bidding (and carding) system, but a machine-learning designed bidding system could quite possibly (in my opinion would very likely) be incomprehensible to humans.

2
  • 1
    Not just when there's partnerships, but even when it comes to multiple separate players, good AIs are harder to design - that's part of why all the examples given are 2 player games, there's less activity between the AIs turns to calculate based off of (and thus a lot less possible scenarios to evaluate)
    – Andrew
    Commented Dec 27, 2020 at 4:48
  • The first year Al Levy ran the computer bridge championship, Ginsberg entered GIB playing a wild transfer opening, relay strong club variant (which works better for computers than for humans as computers don't forget). It was very good straight up (for the time), but since the opponents couldn't "understand" it, it got lots of advantage from "incomprehensible bidding". After that (and even now), the allowed system is very restricted.
    – Mycroft
    Commented Jan 1, 2021 at 17:23

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .