COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	%lectur[w89,jmc]		Notes for European trip lectures
C00003 00003	Mathematical Logic and Turing's Ideas about AI
C00008 00004	The common sense physics informatic situation
C00010 ENDMK
C⊗;
%lectur[w89,jmc]		Notes for European trip lectures

Edinburgh - contexts

Sunderland - naive physics

Edinburgh - Turing's program today

Moscow-1 - Philosophy and AI

Moscow-2 - Formalizing common sense knowledge and ability
Mathematical Logic and Turing's Ideas about AI

Abstract: Alan Turing's 1950 article, ``Computing Machinery and
Intelligence'', contains the first statement that the problem of
making an intelligent machine is the same as that of making an
intelligent computer program.  He also posed the problem of artificial
intelligence in a concrete form---playing the imitation game.
He thought this could be done by the end of the century.
The present lecture will discuss the problem as posed by Turing
and the prospects for solving it.  We will emphasize the possibilities
offered by programs that use mathematical logic, a subject to
which Turing made important contributions but scarcely referred to
in his article.  If enough can be found out, the lecture will also
discuss the intellectual atmosphere in which Turing's article was
written.

notes:
The imitation game is a good idea to smoke out people who
won't admit that machines can be intelligent no matter what they do.
A current example is John Searle.
It's a bad idea as a specific criterion for AI.
ambiguous, holistic

Turing begins by saying that he is not going to attempt definitions of
``machine'' and ``think''.  He rejects the idea of basing definitions
on common usage, and I agree with that.  With regard to ``machine'',
we are in good shape, thanks mainly to Turing.  Turing machines can
be defined explicitly, as he did in his 1936 paper.  Computation machines
in general can be defined to be any structure that is computationally
equally powerful.  Enough kinds of computational processes have been
proved equivalent to Turing machines, so that the historically motivated
choice of Turing machines doesn't restrict generality.

``Intelligence'' is harder.  There still isn't a technical
definition of intelligent behavior.  The ``Turing test'' might be
taken as a sufficient condition for intelligence, but it is
ambiguous in several ways and it certainly isn't necessary that a
machine be able to imitate a human to be considered intelligent.

Hilary Putnam's notion of ``natural kind'' permits proceeding as
follows.  Intelligence is a natural kind about which we have only
partial knowledge.  For those not familiar with natural kinds,
Putnam's favorite examples are taken from nature, e.g. a lemon.
We all can recognize lemons in a store.  We can even tell a child
how to find the lemons in a store and tell them from the other
fruits.  However, our success in instructing the child depends
on the fact that there aren't other fruits in groceries so like
lemons that the distinction is difficult.  We recognize that
we don't know all the properties of lemons.  There could be
a biologically different fruit that I could not distinguish
from lemons.  Fortunately, for me there isn't.

Let's try to apply this natural kind idea to intelligence.  

AI as computer science and AI as biology

Dennett, the intentional stance, the logic level and ascribing mental
qualities.

informatic situations
The common sense physics informatic situation

	The object of this lecture is to express a point of view
about physics in the common sense world.

1. Examples of such physics.  Dropping and throwing objects.
Spilling pitchers of water.  Walking, running, climbing and jumping.

2. We are dealing with some of the same phenomena as scientific physics.

3. The reason common sense physics can't be just a replay of
scientific physics is entirely informatic.  These informatic
considerations will apply to robots with gigaflop computation
rates almost as much as they apply to humans.

4. The relevant scientific physics is mainly classical mechanics.
Classical mechanics works best when there are known differential
equations involving the forces among objects