ARTIFICIAL INTELLIGENCE (AI)

To play a hand 'Double Dummy' is to play a hand looking at all the cards in all the hands. We have built a Double Dummy Solver that is able to determine the best line of play, if everyone at the table plays his cards perfectly at each turn to play.

At the opening lead, the number of possible combinations of plays is umpteen gazillions. There are 13 initial plays, then for each of those for second hand play there are 2 or 3 or 4 choices, then for each of those for third hand there is another few, then for fourth hands, then for the round two lead there is a choice of 12 plays, then .... We have used some very clever AI to prune this immense tree of possibilities from gazillions to a number which is reasonable to handle. For most kinds of deals, with a fast computer we can find the perfect card(s) to play on opening lead in fractions of a second.

In subsequent rounds, play becomes faster and faster, and typically we can make a full double dummy analysis in tenths or hundredths of a second.

Now  for some of you it wouldn't be much fun to play against an opponent who actually 'peeked' at your actual cards and played perfectly, so the AI routine goes like this: 

  • Internally, Buff deals an 'appropriate' hand for each hidden hand, making sure to deal suitable suitlengths and hand strengths (things are 'known' about the hand if it has previously bid, or disclosed information during previous plays)
  • Buff then sends that hand to the Solver, which returns a play that is 'best' for that contrived hand
  • Buff then repeats steps 1. and 2. a number of times, and collects a number of 'best' cards
  • At the point at which Buff stops thinking, Buff determines the BEST card of all the 'best' cards, and plays it.

You choose how  many deals that Buff looks at, and how long you want it to look.A faster computer produces smarter plays for the same intelligence level, because more shuffling/solving loops are possible, producing more 'best' cards, and therefore a greater likelihood that the BEST card is a very good one.

You can also elect the option to 'keep thinking until I stop you'. Buff continues the shuffler/solver loop until you stop it (actually it stops anyway after 400 loops, more than this is kind of redundant).

Plus there are a cumber of 'short cuts' you can elect. For example it's kind of annoying to see the program think about whether to play the '2' or the '4' or the '6' from a holding of '642', when for all practical purposes it cannot make any difference (except possibly to signal partner), so you can ask Buff NOT to think in this obvious kind of situation.  Other common don't-think situations are when your best card cannot beat dummy's lowest card, and when you cannot beat the highest card played-to-date. Since the longest thinking is on the opening lead, you can ask that that be an instantaneous 'rules-based' lead instead of an AI lead.