I have a rather off-beat (but hopefully reasonable) question to put to
you all for comment/discussion regarding a kind of thought experiment I
have been forming.
What I have been considering is a modified Turing Test. Rather than
dealing with the 'traditional' test, I am trying to imagine a virtual
universe in which agents interact: both human and artificial.
Firstly, imagine that this virtual universe is a perfect replica of our
own. An artificial agent that was indistinguishable from a human agent
(measured by observation of action and interaction) would, in my
opinion, pass the Turing Test. Obviously though, I have expanded the
Turning Test to include observing actions and to not just rely on
communication (and this may be rather harsh).
Now, what I would like to ask you all is can you suggest a broad set of
constraints that could be placed on a virtual universe such that the
current technology of artificial agents would be indistinguishable from
human agents in this constrained universe.
For example, I could constrain the Universe to a set of mathematical
problems. However, a well programmed agent might easily beat a human and
hence be picked for the computational genius it is (ie, and NOT HUMAN!)
What I am looking for is a very simplified model of our world with
constrained actions and interactions that would make it impossible to
discern between the artificial and the biological but while maintaining
a high degree of variety and complexity of action/interaction.
Thanks in advance,
Tim Wilkin
--
____________________________________________________
_--_|\ [EMAIL PROTECTED]
Tim Wilkin / \ Ph: +61 3 9905 9672
Aerosonde Project \_.--._/ Fax: +61 3 9905 9689
________________________v___________________________
CRC for Southern Hemisphere Meteorology, Monash Univ
School of Computer Sci. & Software Eng., Monash Univ
Bureau of Meteorology Research Centre
----------------------------------------------------
