AylerKupp: <<Sally Simpson> The real tragedy being if Black had played 34...Kg7 Duchess (White) would not have seen 35.Qf8+. At 9 ply it was too deep for it to see.>Even if what the article said was true, that 9 ply was beyond Duchess, that probably meant that <on the average> 9 ply was beyond Duchess. But by this time the concept of search <extensions> was well known; Shannon in his classic 1949 paper http://vision.unipv.it/IA1/Programm... described his Type-B strategy involving what came to be called <quiescent positions>:
"From these remarks it appears that to improve the speed and strength of play the machine must:
(1) Examine forceful variations out as far as possible and evaluate only at reasonable positions, where some quasi-stability has been established.
(2) Select the variations to be explored by some process so that the machine does not waste its time in totally pointless variations."
He further indicated that "It is not difficult to construct programs incorporating these features" and even described how it could be done.
Duchess was a strong engine at the time, playing in 7 ACM North American Computer Chess Championships from 1974 to 1981. It also finished 2nd in the Second World Computer Chess Championship and tied for first in the Eighth North American Computer Chess Championship, both in 1977. So it was likely that in 1977 it incorporated the concept of quiescent search, particularly since it also incorporated the concept of iterative search deepening. See https://chessprogramming.wikispaces.... The clincher as to whether the Duchess programmers were familiar with Shannon's concepts is a photo on the page indicated by the link of Claude Shannon himself sitting on the side of Duchess in the game Duchess vs. Belle, although the photo is dated 1980 (and Duchess lost that particular game).
The longest sequence, as you pointed out, is 34.Qa8+ Kg7 35.Qf8+ Kxf8 36.Bh6+ Bg7 37.Rc8+ Qd8 38.Rxd8+ Re8 39.Rxe8#; 11 plies. But this includes at least 2 captures, 35...Kxf8, 38.Rxd8+ and possibly 3 (39.Rxe8#) depending on how Duchess counted captures. And if Duchess also incorporated forced moves into its quiescent search categorization (per Shannon), then that also includes 37...Qd8 and 38...Re8. So that's at least 4 plies out of 11 when Duchess would nave needed to evaluate only 1 or 2 positions per ply, reducing the number of nodes to be evaluated signifcantly. Given that Duchess was able to evaluate about 200,000 nodes in its search tree per 3-minute move (same wikispaces link), it seems to me that Duchess was quite capable of identifying that 11-ply Principal Variation even without search tree pruning heuristics, given that alpha-beta pruning was probably also used, since that principle has been known since 1956 (https://en.wikipedia.org/wiki/Alpha...), particularly since it would have an overwhelming material advantage (up a queen) by the 9th ply after 38.Rxd8+ so it would have been predisposed to select this line as its PV. And we certainly know that Kaissa was able to do that since the horizon effect kicked in and it played 34...Re8 to delay the inevitable mate as long as it could.
But, if Duchess was indeed unable to see the entire longest mating sequence after 33...Qxd6 due to its 9 search ply limit, then I can formulate AylerKupp's 2nd corollary to Murphy's Law (see my forum's user profile header for the first):
"If you use your engine to analyze a position to a search depth=N, your killer move (the move that leads to a forced win) will be found at search depth=N+1, regardless of the value you choose for N, and your engine will miss it."
Kibitzing
