Information Technology Reference
In-Depth Information
12
Can Computers Really Think?
This is a book about thinking. Looking at a wide range of behaviors that appear to
depend on thinking, it showed how they could be realized in computer programs:
solving recreational puzzles, interpreting visual scenes, understanding English
expressions, planning courses of action, playing games, and some forms of explana-
tion and learning. Only simple versions of these activities were considered, of course,
but in each case, thinking—bringing knowledge to bear on the activity—could be
understood as a form of computation.
What can be learned from this? Perhaps the following:
It is possible to get computers to do fairly impressive things even with small
programs. (All the programs in this topic are well under one hundred lines long.)
What is needed is a programming language that allows stating what needs to be
known to perform the task.
It is indeed possible to interpret at least some forms of thinking as com-
putation. What needs to be known can be represented as a collection of
(atomic and conditional) sentences, from which conclusions can be drawn using
back-chaining.
The real issue is not so much whether intelligent behavior can be produced at
all but
how well
. Almost all the tasks considered (puzzles, language, planning,
games, and so on) worked in simple versions, but as the problems got bigger
and more complex, the basic methods did not scale well.
Not surprisingly, the bulk of the research in artificial intelligence (AI) is concerned
with how to achieve intelligent behavior in complex situations where simple tech-
niques do not work (like playing chess and Go, rather than tic-tac-toe, for instance).
The topic showed some of the ways of dealing with this complexity:
how to do better than basic generate-and-test;
how to represent action and change for large worlds;
how to play games without searching the entire game tree;
. . .
