> It's helpful to discuss though as a background to AI in even
> a basic bot. If you want to give the bot some smarts at some
> point you have to stop with the engineering and take a look
> at imprinting it with some physiological behaviour

After the little I have come to know about the world I live in most
attempts by humans to endow their robotic creations with 'intelligence'
(of the kind that allows it to cross the street or buy groceries) will
likely fail. I think that bug-style simple logic and the capability to
learn (carrot or stick button and some form of evolution capable AI
system, like a neural network), is what is needed for a start, and then
many, many, many thousands or tens of thousands of real world interactions
to build a workable set of rules. Once you have a bunch of working rules
it is easy to transfer them to other robots. Of course someone would have
to pay for this, and then no-one would insure any kind of robot based on a
heuristic system that cannot be proved to be 'safe' (because it has
learned what it can do in the same way as people have, who *are* life and
accident insured by the same companies). Yet I believe that this is the
only way to gather information about what those rules should really be.

F.ex. Asimov's robot laws would not allow a robot autopilot or defense
system to save 1 million people by killing one, for example a terrorist
who has stolen an aircraft with nuclear weapons on board, en route for a
major city. Okay this is extreme, but there are other scenarios. Like
three unconscious people in a submarine after an accident, one each in an
automatically sealed compartment, with oxygen on board just enough for two
to live until rescue is sheduled to come. Can the robot decide who is to
live? Will he kill all three by not killing one? Will he be 'blamed' if he
does? If he does not?

Peter

--
http://www.piclist.com#nomail Going offline? Don't AutoReply us!
email listserv@mitvma.mit.edu with SET PICList DIGEST in the body