COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	%penros.2[w90,jmc]	Review of Penrose book for Reason
C00006 00003	Penrose has written a number of clever programs, and I
C00008 00004	\smallskip\centerline{Copyright \copyright\ 1990\ by John McCarthy}
C00009 ENDMK
C⊗;
%penros.2[w90,jmc]	Review of Penrose book for Reason
\input memo.tex[let,jmc]

\noindent {\it The Emperor's New Mind}, by Roger Penrose.
Oxford University Press, Oxford, New York, Melbourne,
1989, xiii + 466 pp., \$24.95. ISBN 0-19-851973-7

	The possibility of making machines as intelligent as people
has been controversial since before the first efforts at doing so
beginning in the early 1950s.  The people who believe it can be done
have mostly concentrated their attention on trying to discover
more and more intellectual mechanisms and program them for computers.
The people who think it can be done divide into two groups---those
who heckle the attempts to do it and those who ignore them and
concentrate on arguments in principle.  Penrose is in the latter
category.

	Those who make arguments in principle divide up into people
who consider certain performance as impossible for machines and
those who say that it doesn't matter what a machine's performance
is, it still shouldn't be counted as intelligent.
Penrose follows both of these tracks.

	Although he doesn't discuss any of the actual research in
artificial intelligence, this research is relevant to his claims,
and I'll begin with that.

	Some artificial intelligence research involves studying the
brain and trying to imitate it, but most involves studying the world
and trying to make programs that will achieve goals in the world.

	We humans express what we know about the world mostly in
terms of facts rather than procedures, and almost all of the
knowledge we communicate to each other takes the form of facts,
even when the goal is to tell the other person how to do
something.  Concentrating on facts also turns out to be most
fruitful when instructing a computer.  Therefore, the work
divides up into the task of writing programs that can generate
facts from observations and use facts to act in the world and the
task of supplying a large collection of relevant facts about the
world.  Both are hard, but most of the work goes into collecting
facts and expressing them in a form usable by the computer.

	Fortunately for artificial intelligence, a suitable
framework for expressing facts was invented by mathematical
logicians long before artificial intelligence was first
attempted.

Penrose has written a number of clever programs, and I
suppose this gives him a certain intuition about what
programs can and can't do.  Making a successful
Chinese room program will require far deeper understanding
of what knowledge is than the artificial intelligence
community has achieved so far, but that community is
far ahead of Penrose and Searle.

Penrose is fundamentally a sentimentalist.  We quote from
his Epilog.

There is no analysis of what is required to be able to answer
a question beginning ``What does it feel like to $\ldots$''.
In fact when an intrusive reporter asks such a question
of someone who has just suffered a tragedy, we Earthmen
often find it difficult to answer.

In the end, we'll just have to see how it turns out.
\smallskip\centerline{Copyright \copyright\ 1990\ by John McCarthy}
\smallskip\noindent{\fiverm This draft of PENROS.2[W90,JMC]\ TEXed on \jmcdate\ at \theTime}
%File originated on 23-Mar-90
\vfill\eject\end