Artificial Intelligence and Bridge
This article was published in 1984 but is still current:
indeed, no progress has been made since then with respect
to Artificial Intelligence (A.I.) as applied to Bridge in spite of the many partial attempts made.
To make things clear, several software were produced and marketed worldwide but strictly speaking, without clearly
referring to A.I. because using, for example, double dummy hands, or by systematically applying empirical methods
that were not always adequate, or by only working on certain well-chosen deals.
A real A.I. program must indeed faithfully reproduce human behavior, namely in this case:
- play all types of deals with closed hands
- being capable, on request, of explaining why each bid was made, or why each card was played.
What’s « understood » can be explained.
The reasoning set out by the machine indicates the quality of the game provided as well as whether it
is capable of proving its global apprehension of the problem raised.
In 2017 Will-Bridge, a french firm created in 1987 to exploit Philippe Pionchon’s work in the field of Artificial
Intelligence, remains the uncontested world leader in this field as it applies to Bridge.
What is Artificial Intelligence ?
Artificial Intelligence (A.I.) is a field whose vocation is to simulate human behavior as it applies to its perception,
understanding, or decision making activities.
It is felt in various areas: recognition of forms, recognition of words, analysis of language, resolution of problems,
With respect to the area of « resolution of problems » which is of interest to us here, A.I. is characterized by
a new approach in the data processing of problems which, contrary to classic systems which consist in describing
the resolution of the problem, for which therefore it is necessary to know the resolution algorithm,
but to put in place a structure enabling the computer to resolve the problem by itself:
the structure being in place, all that needs doing is to describe the problem, and no longer to have to resolve the problem.
That is the fundamental idea that has given birth to the “Expert Systems” intented to process problems known
as “knowledge problems”.
The Expert Systems are data systems in which the experts’ knowledge of a given area has been described, and who
know how to process this knowledge by using a computer’s calculating power to study existing correlations between
the elements of this knowledge and to draw conclusions, “inferences” which themselves complete this “Data Base”, etc.
In fact, in many areas, the problems of knowledge arise in terms of correlations and inferences between the objects and
the variables in this area.
The enrichment and the exploitation of this Knowledge will then consist in establishing new inferences.
In the processing of this type of non algorithmic problems, the computer must no longer be used as an
« intelligent slave » obeying procedural logic, but as a generator of inferences whose work is based on logic describing
And so, a real creation of Knowledge can then be produced by the machine within the limit of the rules of
knowledge and of the facts that have been provided to it.
Expert Systems (E.S.) exist in various areas such as Medicine where the machine is capable of creating a real medical
diagnostic, Geology where the work produced by the computer could be compared with what an expert geologist might produce,
The interest of Bridge
However, in the field of problem resolution, two very different types of problems must be distinguished:
the problems of « Knowledge » and the problems of « Thinking ».
If the E.S. are apt to deal with knowledge problems, they are in principal incapable of dealing with problems of « thinking ».
In this respect, it is interesting to emphasize to how great an extent Bridge, with its three very different phases, namely bidding,
the lead, and the play of the hand, really cover all of the aspects of the field of “problem resolution” and constitute a
special field of theoretical investigations.
Bidding and leads: Knowledge problems
These E.S. are totally able to deal with bidding, which is essentialy a problem of « knowledge ».
Also, they have an educational capacity and can easily answer questions such as:
“What bid should be made?” and “Why was this bid made ?”
As to the lead, it’s an algorithmical set of problems, therefore very easy to resolve.
However, it is not enough to deal satisfactorily with all of the bidding problems when educational objectives are set
and the difficulty begins when the subject of negative explanation, a variant of "rear chains":
“Why not that bid ?”.
The attempts which were made up to then on negative explanations in research institutions collided with the phenomenon
of combinatorial explosion and were abandoned.
In 1984, Philippe Pionchon made this technique progress by doubling the Expert System inference motor with a
second motor of « reverse inferences » enabling it to effectively deal with the problem of negative explication without
the inevitable combinatorial explosion.
Besides this exactly corresponds, and it isn’t surprising, to what happens at the bridge table:
each time it’s his turn, has to produce a bid (inference motor) and must decode his partner’s bids (reverse inference motor).
The outcome of this problem is very important because, besides the indispensible pedagogic aspect, it is thanks
to it that the computer will be able to “invent” bids such as the “3rd” or the “4th suit”:
by decoding every bid it could make in his situation, it realizes that neither is possible, for example,
because of questions of “forcing/not forcing”, an unprotected suit , strength of a hand or number of cards in a suit,
consequently, to make the adequate bid, there some data is missing, and it will itself take the initiative of “inventing”
the “4th suit”.
Exactly the same way a human player would do.
The problem of the bidding and the lead having been resolved, remains the problem of the playing of the hand, which is much
more complex and said to be impossible to solve.
Reflection problems: Strategic games
Historically data processing of strategic games, the first realization of A.I.(in the 70s) took place on the game of
Chess even though it was considered to be a much more difficult game than bridge.
Although nothing had been done for the game of Bridge, many small individual Chess machines could be found in shops that had
a quality of game play that was quite adequate, and the question that intrigued at the time was:
“Why do highly effective machines exist for Chess, a very complicated game, with none for Bridge,
a more simple game ?”.
The answer is: « Exactly... » which perfectly illustrates the « paradox of the difficulty ».
The paradox of the difficulty
There are several important differences between the set of problems posed by the game of Chess and those posed by the game
The main one is because Bridge has hidden elements whereas in Chess, everything is “on the table”.
To play Chess, a machine doesn’t need to be « intelligent », it’s enough for it to be able to calculate.
And there, a computer is unbeatable.
From a combinatorial point of view, Chess is much more vast than Bridge.
It is even so much so that one can say that no man will ever be able to master this game.
In other words, that for the complete and exhaustive analysis of a situation in Chess, there exists no expert in the world
who would be able to do it, and it would be more correct to say that for Chess, the computer is a lousy player,
but nobody notices it.
So the machine having decided on a certain move during the game, nobody will be able to say that there exists another
move superior to all the others.
Therefore, it is difficult to criticize the game produced by a machine.
All you can do is to note In Fine that the machine played either better, or less well than its opponent, and that’s all.
This impossibility of global apprehension is true to such an extent, that a player will choose a certain move simply because
"it is reputed to be good" or because "it allows a pleasant development":
It cannot make a complete analysis because there are too many subsequent developments.
Thus, - it’s paradoxical – it’s much easier to buid a machine that plays Chess since it amounts to submitting a problem
to it for which nobody knows the solution!
So, Bridge is much simpler
therefore its more complicated...
To play a good game of bridge with your cards, you must « think »…
You must be able to make a global analysis of the problem…
At the time, the problem was thought to be impossible to solve.
Furthermore, to deal with them, probabilities have to be taken into account, which is relatively easy for a machine,
but above all, you must work using modal logic, as the mathematicians say, it means that you must make use of the theories
of possibilities, fear, necessity, etc.
These are so many areas which are far from being mastered by Artificial Intelligence.
In Bridge, there is the « obligation to produce results »
In Bridge, at the end of the play, most of the hands, when you see the 4 of them, can easily be analysed, even by very
Therefore, it is easy to criticize the way a machine has played.
If at a given time a card and only this card must be played, every player will see it: therefore the machine must absolutely
For it, it is the “obligation to produce the result”.
In conclusion, and here we have a paradox, in Chess, since it is a very complicated game, when the machine plays badly,
nobody sees it whereas in Bridge, because the game is an easier one, it’s easy for everyone to notice.
A machine has the “obligation to produce a result” in a Bridge game which is not necessary when it plays Chess.
How should this obligation to produce a result be handled ?
The basic idea: the metaKnowledge
« MetaKnowledge » is the « Knowledge of Knowledge », as it were, the Knowledge you possess about Knowledge.
If for example you were asked if Mr. Smith was once president of the United States, this is a problem of knowledge.
If you possess this knowledge (if you know), namely if you possess the list of presidents, you can answer yes or no.
If now you were to be asked « Has Mrs. Smith ever been president of the United States? “, you would immediately answer
“no” even though you don’t possess this list, because you know that no woman has ever been president of the United States.
MetaKnowledge is enough, you have no need of knowledge to solve your problem.
This is typical of an important part of the set of problems that exist in how to play a hand of Bridge ,
and studying the theory of intelligent machines, Philippe Pionchon became interested in Bridge, a little bit by chance,
because he had a premonition of the advantage of being pertinent, easy to criticize and easy to design a model.
There was nothing like Bridge to put his theory to the test.
« To my knowledge, Bridge is the most scientific game in the world:
it uses every area of modal logic
and bridge players spend all their time, without realizing it,
reasonning in metaKnowledge”.
he declared before developing his basic idea:
elaborate a theory of intelligent machines based on an Expert System
Inferences based on metaKnowledge ,
isn’t that in fact the precise definition of intelligence,
my dear Watson ?
Possibilities, necessities, fears, this is, in fact is the daily life led by bridge players:
. « I don’t know who has the King of Clubs, but I know that East first to bid passed and had already showed he had 11 HCP:
so he couldn’t have the King of Clubs. »
. « I don’t know who has the King of Clubs, but if RHO has the King of Clubs, I can’t win. Since my objective is to win,
the hypothesis of necessity means that I have to put it in LHO’s hand.
I’ll play the hand as if I’m certain it is in LHO’s hand.
If it’s there, I win. If it isn’t, I’ve lost, but I couldn’t win: so I haven’t lost anything”.
. “ Only a 4-0 distribution of opponent’s trumps endangers my contract.
Therefore I’ll consider that they are 4-0 and I’ll try to find a strategy that wins whether they split 2-2 or 3-1.”
Of course if in many cases the use of metaknowledge makes it possible to intelligently deal with the problem of
hidden elements, it isn’t sufficient but can be made more complete with a simple calculation allowing the machine to
find all by itself the solution in other areas such as the handling of suits or the use of safety plays.
. « I don’t have the King of Clubs.
Is there a way to play a suit so that it will win wether the King of Clubs is located on your right or your left?”
The machine will put it both on the right and on the left, with the order to use it only once, and, if the solution exists,
it will find the adequate way to play the suit all by itself.
Once again, exactly how a human player would do...