AI Lessons

Geographic Names: US Personal Names: Kasparov, Garry

The implications of IBM's Deep Blue system, which defeated Garry kasparov in a recent chess match, are discussed. Grandmaster Joel Benjamin and Murray Campbell of IBM claimed important advances during the year between matches with Kasparov in Deep Blue's chess knowledge, not merely its ability to examine more positions. Those improvements include: 1. a large opening book, 2. the ability to assign valuations to each of the leaf positions it reaches through look- ahead search, and 3. the ability to notice patterns of pieces distinguishing one kind of position from another and to use different weights for features in evaluating positions of different character One implication for the future of computing is that speed alone cannot solve complex problems and must be supplemented with knowledge. Capabilities must be developed for learning patterns autonomously from information about the task environment

. . .
ain circumstances, it conducts very deep but narrow selective searches. Thus there are three directions-beyond increased speed-along which Deep Blue could have improved during the year, and its play gave every indication that it improved in all three. Its greater strength over 1996 is much more convincingly explained in these terms than by the modest increase in its look-ahead ability. Especially interesting is that these improvements correspond to three kinds of chess knowledge possessed by human chess players, enabling them to select good moves after relatively little search (almost certainly never more than 1,000 branches, except when the "book" is followed). Deep Blue's increased speed may be allocated, not to searching deeper but to computing and recognizing more sophisticated patterns, permitting more selective search. Deep Blue's essential elements can be summarized as: Enormous knowledge and databases for the relatively small target domain of chess played on an eight-square X eight- square board. The ratio of knowledge to size of domain is much higher than in most typical Al applications, such as expert systems. A basic strategy not meant to imitate a human but to combine rapid processing with knowledge. The mach
. . .

Some common words found in the essay are:
Deep Blue's, Deep Blue, Promising Applications, Sidebar Generic, Application Engineering, Arthur SamuePs, Blue IBM, Campbell IBM, IBM Research, Classification Codes, deep blue, deep blue's, chess knowledge, kasparov deep, chess match, ability notice patterns, leaf positions, computational power, assign valuations, notice patterns pieces, ability assign, ability examine positions, positions reaches, ability assign valuations, leaf positions reaches,
Approximate Word count = 1679
Approximate Pages = 7 (250 words per page)

More Essays on AI Lessons