Greg Detre
Wednesday, October 30, 2002
editing
book � main philosophy question of it: pass the THAT reliably � better than
chance
did he say
it was sufficient??? he did say it wasn�t necessary, e.g. I fail the Turing test
in Italy
philosophers: is the test criterial for
thinking?
Chomsky: who cares if it�s criterial
Dennett: y, it�s criterial, i.e. sufficient
everyone else: no, it�s not � but for different reasons
Moore: it�s not criterial, bu it�s evidential
how do you judge how good a test is, if you�ve declared the original
question �can machines think?� meaningless?
what does it mean to �replace� the question with the test?
what�s the
pertinent property (e.g. length when comparing metre-sticks) of intelligence?
Turing says verbal behaviour
Block:
knockdown argument that the Turing test is inadequate
one
(non-knockdown argument) is:
Fred is lying in bed, pretending to be ill to
skip school
Fred and Luigi are conversing in Italian, but
Fred�s not fluent in Italian, he�s just pretending to be to be eligible for an
Italian scholarship
deniers
think that you can have the symptoms without the germ
you have to exhibit something that has the symptoms without the germ
random
character generation � infinitesimal chance of passing
this has the capacity to pass, but is unconvincing
Block: the
Aunt Bertha machine
enumerate all the possible things that AB would say given a
conversational path through an hour
Parkes: �as intelligent as AB for that hour�
number of nodes > atoms in universe
is it logically sufficient?
main objection � from exponentials
table lookup doesn�t feel to Shieber like an adequate structure for
intelligence
why should we care about the internal structure?
general capacity???
you have to do it in a way that doesn�t run afoul of exponentials
how it happens vs what happens
Block: richness of information processing
�general capacity� is a psychologistic notion (which he supports,
right???)
can the Turing test reveal a general capacity that doesn�t require exponentials?
think of the Turing test as an interactive proof(???)
Turing test
as exponentially strong evidence
space-physical realisabiilty, time-realtime
play 2
grandmasters, 1 black, 1 white, play one GM against the other via your arms
proof in the form of a conversaation
graph isomorphism � are these two graphs the same, except the vertices relablelled and the srambled on the page
NP-complete problem
given an algorithm for graph isomorphism, show it�s correct by:
make a copy of one, scramble it up yourself, give it to the other person who says he has the algorithm, and if he can tell it�s isomorphic, then you know he has the algorithm
certification of the proper functioning of an algorithm that you don�t have access to
interactive proofs can be done for anything that can be computed in polynomial space
related to zero knowledge proofs but more general
questioner is a mortal/Arthur, verifier is God/Merlin
all the questions will help you equally
calculate the bounds of: if Merlin is wrong, the probability that you�ll find it
see comic on University of Chicago webpage
the whole
concept of a huge look-up table is so far beyond our intuition, even though
we�re relying on it
grade A
machine
thinking vs
simulated thinking???
machine as
judge in Turing test (inverse Turing test) � can be shown to be equivalent to
the Turing test
�thinking�
is operationally defined though, I think???
Antony�s
question: what about if you pause for 10 seconds � it should ask you what�s
wrong, but the AB machine wouldn�t
you could have (absent) keystrokes as a kind of time-keeper. failing
that, then you just have to add a clock to the AB machine (very unsatisfying
solution)
table
look-up doesn�t have flexibility � is that why he doesn�t like it???
is
consciousness supposed to be the �germ�???
natural
kinds
intelligence
vs life??? both have the same kind of fuzzy, human-defined definitions
interactive
proof � graph isomorphism??? see above
capacity vs complexity???