xix Heuristic search procedures require algorithms, called "heuristics," for estimating the values
of nodes in the state space being searched... The development of good heuristic-estimate functions remains a key
problem for AI research: experience indicates that this is an area where the problem of problem representation remains central
p.68 A paradigm is a general model of something that is found to be useful for investigating that
thing.
p.95 Thus, "heuristic programming" refers to computer programs that employ procedures not necessarily
proved to be correct, but which seem to be plausible. Most problems that have been considered by AI researchers are
of the sort where no one knows any practical, completely correct procedures to solve them; therefore, a certain amount of
proficiency in using hunches and partially verified search procedures is necessary to design programs that can solve them.
So, by a heuristic is meant some rule of thumb that usually reduces the work required to solve a problem. (Again,
it may be possible to prove that the heuristic will always supply solutions to some set of problems, i.e., that it
is algorithmic.) Clearly, much of the conscious thinking that people do is based upon the use of heuristics
that have not been shown to be algorithms. The realization of this fact and its incorporation in the design of computer programs
was an important step in the development of artificial intelligence, signifying a recognition by AI researchers that intelligence
is often exhibited in situations where one's understanding and knowledge are incomplete.
p.100 An evaluation function is some procedure that can be applied to the finite description of a node in
a state-space problem and which will produce an estimate of the "value" of that node (the likelihood that the node lies on
a path to a goal node)... Samuel (1959, 1967) used an evaluation function that examined the important "features" possessed
by a board configuration in checkers, to produce an estimate of the "strategic value" of the configuration"
p.119 In general, a strategy is any set of rules that tells a player what choices he should
make for all situations that might arise during the course of a game.
p.121 The strategic "problematic" aspect of strategy arises from the definition of a payment function,
which specifies that certain paths through the state space of such a game will yield payments to the players.
p.128 If a program could be designed to generate only those successors that were "reasonable," that could
do a minimax analysis on the resulting reasonable game-tree, and that could select the alternative below its current
situation with the highest reasonable evaluation, it would still be able to play a very good game.
p.129 A static evaluation function is a method for estimating the value of a node which is not
dependent on the values of the successors to that node. A good static evaluation function is one that tends to give estimates
that agree with the true, theoretical values of the nodes in a game tree... For our purposes, a static evaluation function
is necessarily a computational procedure that can be applied by a computer to its description for any given situation that
might occur during a play of the game. The function should yield for the situation a numerical value approximating that which
would be obtained by analyzing the game completely.
p.134 However, the worth of the technique [JLJ - alpha beta tree search] is greatly dependent upon the order
in which the nodes of the tree are taken for examination... In using the alpha-beta technique, it is desirable to have some
method that will make it likely the best nodes are evaluated first.
p.164 Alternatively, one can view a game as a problem in which the solution is a tree, rather than
a sequence, of operators.
p.274 A language is a set of sentences that may be used as signals to convey semantic
information... for a computer that uses a sentence, "understanding" may be corresponded to making internal data structures
(vectors, lists, graphs, programs, etc.) that model these elements of the meaning of the sentence.
p.336 Throughout, this chapter adopts the idea that "understanding," whether human or mechanical,
is a process that involves "model making."
