pursuits. Researchers in the field of natural language processing are interested
in developing machines that can manipulate natural language and therefore
lean in the engineering direction. Thus, linguists operate in simulationoriented
mode—building systems whose goals are to test theories. In contrast,
researchers in natural language processing operate in performance-oriented
mode—building systems to perform tasks. Systems produced in this latter
mode (such as document translators and systems by which machines
respond to verbal commands) rely heavily on knowledge gained by linguists
but often apply “shortcuts” that happen to work in the restricted environment
of the particular system.
As an elementary example, consider the task of developing a shell for an
operating system that receives instructions from the outside world through verbal
English commands. In this case, the shell (an agent) does not need to worry
about the entire English language. More precisely, the shell does not need to
distinguish between the various meanings of the word copy. (Is it a noun or a
verb? Should it carry the connotation of plagiarism?) Instead, the shell needs
merely to distinguish the word copy from other commands such as rename and
delete. Thus the shell could perform its task just by matching its inputs to predetermined
audio patterns. The performance of such a system may be satisfactory
to an engineer but the way it is obtained would not be aesthetically pleasing to
a theoretician.
The Turing Test
In the past the Turing test (proposed by Alan Turing in 1950) has served as a
benchmark in measuring progress in the field of artificial intelligence. Today
the significance of the Turing test has faded although it remains an important
part of the artificial intelligence folklore. Turing’s proposal was to allow a
human, whom we call the interrogator, to communicate with a test subject by
means of a typewriter system without being told whether the test subject was a
human or a machine. In this environment, a machine would be declared to
behave intelligently if the interrogator was not able to distinguish it from a
human. Turing predicted that by the year 2000 machines would have a 30 percent
chance of passing a five-minute Turing test—a conjecture that turned out
to be surprisingly accurate.
The Origins of Artificial Intelligence
The quest to build machines that mimic human behavior has a long history, but many
would agree that the modern field of artificial intelligence had its origins in 1950.
This was the year that Alan Turing published the article “Computing Machinery and
Intelligence” in which he proposed that machines could be programmed to exhibit
intelligent behavior. The name of the field—artificial intelligence—was coined a few
years later in the now legendary proposal written by John McCarthy who suggested
that a “study of artificial intelligence be carried out during the summer of 1956 at
Dartmouth College” to explore “the conjecture that every aspect of learning or any
other feature of intelligence can in principle be so precisely described that a machine
can be made to simulate it.”