G.Joly@cs.ucl.ac.uk (Gordon Joly) (06/03/91)
If we conjecture that the Turing Test will always fail, since we could
always, in principle and given enough time, "see through" an
artificial mind, the "time to discovery" could form the basis of an
alternative. Compare this approach with the way in which Eliza
(Doctor) did - "you can fool some of the people some of the time."
The Turing Test Quotient (TTQ) is a metric based on the average amount
of time before n people can spot that a given AI system is not real
intelligence. In we had a sample of n=1000, that would be the TTQ of
order 1000. The (time) unit would be the average of all subjects, with
an upper time limit of exposure (not boundless, see above) calculated
from a trial sample, of perhaps n/10. The TTQ will be a log scale
measure. An average time of 6 mins is 1 unit on the scale, with an
average of 60 mins will be 2 units, 600 mins give 3 units and so on.
I guess a 3 to 4 unit system, of order n=1000, might be of some real
use, as a general purpose, common sense, AI system. The experiments
should be done double blind.
____
Gordon Joly +44 71 387 7050 ext 3716
Internet: G.Joly@cs.ucl.ac.uk UUCP: ...!{uunet,ukc}!ucl-cs!G.Joly
Computer Science, University College London, Gower Street, LONDON WC1E 6BT
Drop a utensil.