From: Basto Jorge (jldcb199@ecs.soton.ac.uk)
Date: Sun Jun 03 2001 - 17:15:53 BST
It seems that there are some "defensive" mechanisms built in to avoid
some traps and some common sense tricks ("I'm a robot, not a
calculator"). So there could be some tests to turn around this like:
1. Does it always give the same response to the same question? What
would happen if the questions are repeated over and over? The same
response to the same questions would certainly be evidence of its
limitations.
2. Can it be taught? It claims it is not a calculator, but can it be
tought to perform some simple calculations and possible build on that?
Or more generally, can it be taught anything at all? For instance,
could he learn something about your job and later on show evidence that
it learnt?
3. This requires some time, but it is an application of the permanence
of intelligence issue of the life-long Turing test: for how long can it
convince you? Is there not a moment in time when you have certain
doubts that it is actually thinking at all?
4. Related to 3. Try to devise a way of creating a situation/plot that
leads to a frame-problem state and see how it behaves.
The problem I think is to devise THE one question that has it all, I
mean that can make the robot pass or fail and this can be tricky if
indeed the robot has some sensorimotor mechanisms and can learn some
more facts than those it had at creation time. I will think about this
after exams.
This archive was generated by hypermail 2.1.4 : Tue Sep 24 2002 - 18:37:31 BST