Professional Documents
Culture Documents
A
lan Turing was a brilliant of Von Neumann. His 1936 “computational problem” and
mathematician who paper, On Computable Numbers revealed the limitations of the
showed that computers with an Application to the power of Turing machines to
could not completely prove Entscheidungsproblem, proved that handle problem solving.
mathematical assertions, extend- mathematics could not be com-
ing Gödel’s proof that logic pletely modeled by computers. In Hilbert, Gödel, and Church
could not completely model the early 1940s he developed a In 1900 Hilbert proposed that
mathematical truth. This con- computer model of German logic could completely prove the
nection between computers and cipher code that helped the Allies truth or falsity of mathematical
mathematics was later used to win World War II. In the late assertions, and listed 25
develop a mathematical founda- 1940s he developed computa- unproven mathematical asser-
tion for computer science, com- tional models of artificial intelli- tions that mathematicians should
parable to mathematical gence, chess, and the human try to prove. Russell and White-
foundations for physics and mind, suggesting that computers head’s Principia Mathematica
other sciences. could completely model human accepted Hilbert’s principle and
This column shows that Turing thought and would play chess provided an account of mathe-
machines are inappropriate as a better than humans before the matical logic as a universal model
universal foundation for compu- end of the century. of mathematical provability. The
tational problem solving, and that Although the 1936 paper was failure to achieve their goals led
computer science is a fundamen- primarily about the inability of to Gödel’s 1931 proof that logic
tally non-mathematical discipline. Turing machines to solve mathe- could not decide all mathemati-
Though interaction is not the matical problems, Turing cal theorems [3]. Gödel showed
only way to extend computation machines were adopted by theo- that the Entscheidungsproblem
beyond Turing machines, we retical computer scientists in the (“decision problem”) was in prin-
show that Turing, Milner, and 1960s as a mode of solving all ciple unsolvable by logic, and
others have used interaction for problems of computing. Here we this led to work by many mathe-
this purpose. examine the historical evolution maticians to further explain the
Born in 1912, Turing was of Turing’s model from mathe- theory and philosophy of mathe-
accepted by Cambridge Univer- matical weakness in the 1930s to matical unsolvability in terms of
sity in 1930 to study mathemat- computational strength in the logic or other models of mathe-
ics, and became a Fellow of Kings 1960s, and then to computational matics.
PAUL WATSON
College in 1934 at the age of 22, weakness in the 1990s as increases Gödel’s ideas were taken up by
completing a dissertation that in the applicability of computa- Church, who proved in 1935 that
extended the group theory models tion broadened our notion of the Entscheidungsproblem could
avoided the question whether the may no longer be fully appropriate References
computation of CCS and the to capture all features of present-day 1. Association for Computing Machinery. Cur-
riculum ‘68: Recommendations for academic
π-calculus went beyond Turing computing. programs in computer science. In ACM Cur-
machines and algorithms. Turing Our concept of interactive ricula Recommendations for Computer Science.
ACM, NY, 1968.
machines had been accepted as a models was questioned because 2. Brooks, R.A. Intelligence Without Reason. MIT
principal paradigm of complete we originally failed to provide a AI Lab Technical Report No. 1293, 1991.
computation, and it was prema- theoretical framework comparable 3. Gödel, K. On formally undecidable proposi-
tions of principia mathematica and related sys-
ture to openly challenge this view to that for Turing machines. tems. Monatshefte fur Mathematik und Physik,
in the late 1970s and the early However, complete models of 38 1931 (in German); English translation in
M. Davis, Ed., The Undecidable. Raven Press
1980s. In the last two decades, computation have often been 1965.
the computing technology has developed with no theoretical 4. Goldin, D. Smolka, S., and Wegner, P. Turing
shifted from mainframes and foundation or mathematical mod- machines, transition systems, and interaction.
In Proceedings of the 8th International Work-
microstations to networks and els. Even Turing presented c- shop on Expressiveness in Concurrency, Aarlborg,
wireless devices, with the corre- machines [7] and u-machines [8] Denmark, August 2001.
5. Milner, R. Elements of interaction (Turing
sponding shift in applications without a formal foundation. Award lecture). Commun. ACM 36, 1 (Jan.
from number crunching and data Though mathematics was 1993).
processing to embedded systems adopted as a goal for modeling 6. Siegelmann, H. Neural Networks and Analog
Computation: Beyond the Turing Limit.
and graphical user interfaces. We computers in the 1960s by anal- Birkhauser, 1999.
believe it is no longer premature ogy with models of physics, Gödel 7. Turing, A. On computable numbers with an
to encompass interaction as part had shown in 1931 that logic can- application to the Entscheidungsproblem. In
Proceedings of the London Math Society 2, 42,
of computation. A paradigm shift not model mathematics [3] and 1936.
is necessary in our notion of Turing showed that neither logic 8. Turing, A. Intelligent machinery. In D.C.
Ince, Ed., Mechanical Intelligence, North-Hol-
computational problem solving, nor algorithms can completely land, 1992.
so it can provide a complete model computing and human 9. van Leeuwen, J. and Wiedermann, J. The Tur-
model for the services of today’s thought. In addition to interac- ing machine paradigm in contemporary com-
puting. In B. Enquist and W. Schmidt, Eds.,
computing systems and software tion, other ways to extend com- Mathematics Unlimited—2001 and Beyond.
agents. putation beyond Turing LNCS, Springer-Verlag, 2000.
10. Wegner, P. Why interaction is more powerful
The model of interaction machines have been considered, than algorithms. Commun. ACM 40, 5 (May
machines as an extension of Tur- such as computing with real 1997).
ing machines was developed in numbers [6]. However, the
the late 1990s [10]; the theoretical assumption that all computation Peter Wegner (pw@cs.brown.edu) is
framework has been improved in can be algorithmically specified is Professor Emeritus in the Computer Science
[4]. Van Leeuwen, a Dutch expert still widely accepted. Interaction Department at Brown University, RI.
on the theory of computation, machines have been criticized as Dina Goldin (dqg@cse.uconn.edu) is an
assistant professor in the Computer Science and
wrote an article extending com- an unnecessary Kuhnian para- Engineering Department at the University of
puters beyond Turing machines digm shift. But Gödel, Church, Connecticut.
[9], which referred to these recent Turing, and more recently Mil-
models of interaction, admitting ner, Wegner, and Van Leeuwen
that: have argued that this is not the
the classical Turing paradigm case. c © 2003 ACM 0002-0782/03/0400 $5.00