Machine learning and natural language processing - The Computing Universe: A Journey through a Revolution - page 296

For the actual game, Friedman and the Jeopardy! team insisted that to coun-
ter Watson's advantage in electronically “pressing the buzzer,” IBM would have
to provide Watson with a mechanical finger to physically press a button. In
addition, IBM decided that Watson needed a graphical representation for the
virtual “face” of Watson. Ferrucci, mindful of the rogue AI computer HAL in
Stanley Kubrick's 2001: A Space Odyssey , suggested, “You probably want to avoid
that red-eye look because when it's pulsating, it looks like HAL.” 11 IBM had
just launched its “Smarter Planet” initiative, a campaign to explore how infor-
mation technology could promote economic growth, sustainable development,
and social progress. The icon for the campaign showed planet earth with five
bars, representing intelligence, radiating from it, and IBM decided that this
image would be the public face of Watson. To make the game more interesting,
an answer panel showed the audience - but not the other players, Watson's top
five candidate answers, and the confidence the machine assigned to each one.
According to Ferrucci, “This gives you a look into Watson's brain.” 12
The match took place at IBM's research center at Yorktown Heights in New
York. The producers recorded the show in January and swore the audience and
contestants to secrecy until after the match was broadcast in February. The match
consisted of two games, and after the first game Watson had the lead. Jennings
and Rutter did better in the second game, but Watson won the last Daily Double
and thus won the game. In his written answer to the Final Jeopardy clue, Jennings
added a postscript: “I, for one, welcome our new computer overlords.” 13
Watson won the game, but is Watson really intelligent? Ferrucci took the
view that “teaching a machine to answer complex questions on a broad range of
subjects would represent a notable advance, whatever the method.” 14 Prior to the
achievements of Deep Blue and Watson, most people would have said that activ-
ities like playing chess or competing on Jeopardy! required intelligence. However,
just as Deep Blue playing chess employed massive computational power to search
out the best chess moves, Watson used massively parallel processing to explore
and rank a huge number of possible answers to the questions. The IBM Watson
team did not try to model the architecture of the human brain, but instead used
a host of algorithms for natural language processing and machine learning that
gave the machine the ability to (mostly) correctly answer complicated questions.
It made no attempt to “understand” the questions as a human would. However,
from the point of view of an operational definition of intelligence as in the
Turing Test, one might say that both Deep Blue and Watson are intelligent. Most
experts, however, would say that Deep Blue and Watson are machines that just
simulate intelligence. Such systems, sometimes called weak AI , can match human
intelligence in a narrow field but not in broader ones. Are such systems a step on
the road to strong AI - machines that can really think, know, and learn - or are
they irrelevant to this goal? This is a question that has generated heated debate
among the philosophy and computer science communities.
In a famous thought experiment called the “Chinese Room” ( Fig. 14.14 ),
philosopher John Searle argues ( B.14.6 ) against the possibility of strong AI. In
the thought experiment, Searle imagines someone who does not know Chinese
sitting alone in a room, following directions for stringing together Chinese char-
acters so that people outside the room think that someone inside understands
and speaks Chinese. Searle explains:

Fig. 14.14. John Searle's “Chinese room”
thought experiment showed that a
human can follow instructions like a
computer and appear to external observ-
ers to understand Chinese, without the
human having any knowledge of the
language.

B.14.6. John Searle is a professor of
philosophy at Berkeley in California.
Within the computer science com-
munity he is best known for his con-
troversial “Chinese room” scenario
as an argument against “strong AI.”