Commentary on GIB-GIB vs Bobby Goldman-Tony Reus, 18 September 1997 by David desJardins [Note that I did not observe this match, unlike the Meckstroth-Rodwell match, so these impressions are based solely on the hand records posted by Matt Ginsberg.] Board 497: Tuna's double is aggressive, especially with EW not vulnerable, and may well be motivated by GIB's reputation. But this time GIB has its bid. In the play, GIB makes the bizarre play of a heart to the king---an outright blunder, which should lead to down two. Tuna misdefends in the end position (discarding C5 rather than SJ) to endplay himself and let GIB escape for down one. Tuna should definitely be able to count partner for the club king and get this one right. GIB does get good marks for correctly judging the endplay at the end, but still a lucky escape for an average board. Board 498: Tuna bids wildly to an unmakable 3NT. GIB disregards the wisdom of the ages (fourth from longest and strongest), which allows 3NT to make. After the club ace lead, the hand is over. Certainly GIB's analysis is understandable: if the queen of clubs had been in the North, then the club ace lead would have done no harm, and if South had held Qx instead of Qxx, the lead would have been a great success. Nevertheless, if GIB wants to prove that it can judge better than humans in this type of situation, the jury is still out. (Board 499 is missing, for unknown reasons. Was it fouled, perhaps?) Board 500: Preemption works. East steps way out (5C would go down 2000), but escapes with a great result. I'm surprised that Tuna didn't double 5C, and I'm surprised that Goldman didn't bid 6D at the end. But hindsight is wonderful, and admittedly South doesn't know that the club king is working. While it worked here, I don't think much of East's bidding strategy, and I don't think it will lead to good results in future matches. Board 501: East-West bid only slightly oddly to a normal 3NT contract which is difficult but makeable. GIB's play of the heart jack at trick two is pointless. The second round of diamonds at trick three is ok. Then GIB guesses to attack clubs. As it turns out, playing spades would have worked out better, but this is not unreasonable. At trick seven GIB can still make the contract by continuing clubs, and this would seem to me the normal line. But GIB instead switches to spades and goes down. No terrible decisions by declarer, just a tough hand which it fails to get right. There are lots of ways to go down, and GIB found one of them. Board 502: Tuna's delayed save on this board is even worse than GIB's delayed save on board 500 (because he does it in a dead rather than a live auction), and the result is a well-deserved disaster. GIB gets no particular credit here: East-West missed a good slam, and then were given a fielder's choice of +1100 or +1370. Board 503: Flat board. Everyone bids normally, nothing much to the play. East-West may be lucky not to find their diamond fit, as 3D rather unluckily goes down 200 (or 500 if doubled). It is interesting that GIB doubles on this hand but overcalls 1NT on board 497. I suppose it likes the second stopper on the earlier board. Many humans would probably bid similarly, although I would prefer 1NT on both. Board 504: GIB's bid of 1S (instead of 2H) is unusual, but the auction works out well enough. GIB's heart ace at trick two is quite odd, but does no particular harm, since GIB later guesses correctly on the second round of hearts. Tuna's duck of the spade ace is very odd, and should allow GIB to make an overtrick by the normal approach of finessing in clubs. GIB then plays clubs from hand instead, and the normal par result is achieved, in a rather unusual way. Neither side seems to distinguish itself here. Board 505: GIB2's super-passive diamond lead against 3NT is unusual---a club would seem normal. But it works well enough, as Goldman guesses to play diamonds immediately and is headed for defeat. (Declarer must play hearts from hand, instead. That wouldn't work so well if clubs were 4-4, though.) Then GIB1 makes a ghastly blunder by covering the jack of hearts. Suddenly, down one turns into making four. I think a human, if unsure about covering, would reason that declarer would have played hearts from hand with AKxx(x), and get this one right. I don't know if or how GIB is capable of this kind of reasoning. Conclusions: GIB's bidding is obviously vastly improved over what it displayed in the "Hall of Champions". I think it is equally clear that it is not yet at an expert level. Referring back to my comments on that match, Ginsberg has addressed many of the "minor hurdles" that it had yet to overcome, and is now up against some of the truly hard problems of bidding. Even on those hands where it reaches the correct contract (like board 504) it's not clear that it is doing so in the best way. It still has a problem with jumping into auctions at a high level, but as this match shows, so do some humans. GIB's level of card play seems adequate but not phenomenal. If combined with really solid bidding, it may be enough to compete at high levels, but at this stage in its development, I would expect a top pair to beat it consistently in a pure test of card play. On the other hand, I also think that it is reasonable to expect further improvement. It should be possible to do something about "waiting plays" (cashing off winners for no particular reason) like the jack of hearts on board 501. It should also be possible to do something about GIB's tendency to play high in second or third hand as declarer, like king of hearts on board 497, and the ace of hearts on board 504. Finding a better lead on board 498 might require some rule-based knowledge (i.e., "fourth from your longest and strongest"). Of course, it's always possible that GIB is "right" in that case and we humans are "wrong"; the one bad result isn't enough to judge. The play error that seems the hardest to "fix" is covering the jack of hearts on board 505. This may require a technique which actually gives GIB more bridge knowledge, not just tuning of the program.