Based on my experience with AI/Expert Systems using Inference Engines I
think one of the problems is deciding what 'knowing' means.
Let's start with this axiom: Data (D), Information (I) and Knowledge
(K) are different things.
"451 deg F" is data (perhaps from a sensor)
"The temperature of my newspaper is 451 deg F." is information.
"When paper reaches the temperature of 451 deg F it auto-ignites." is
knowledge.
So I can infer that my newspaper is bursting into flames (new
knowledge). [Thank goodness for fire places.]
An expert system designed to capture this inference and draw the
conclusion that my newspaper is catching fire, maintains the D, I and K
in memory, somewhat symbolically. AI experts told us that the most
interesting things in life are symbolic rather than numeric (e.g who
your father is). I wouldn't say my PC running the example expert system
really 'knew' anything. Colloquially we'd often talk about the systems
knowing something but I think we all felt it an illusion - although a
useful one. Later we built systems that 'knew' when there was a major
incident requiring operator action such as an equipment freeze up.
To continue with this type of knowing, a roboteer can use many of the
AI/Expert System tools to 'know' stuff at this level, including where it
is and so on and even qualify that knowing with a degree of
uncertainty. The quality of the 'knowing' depending on the ability to
build the expert system, the representational system used within and the
quality and reliability of the sensors. We assumed all of the
components worked in order to draw the conclusion. Sometimes we deduced
that the sensors weren't reliable and therefore rejected the conclusions
that could be drawn from their readings - a sort of meta-knowledge.
Since doing this work I don't think there has been any real advance in
how computers know anything, so I hope this might answer your question.
Thanks
Robert C
Nicholas Thompson wrote:
all,
One of the reasons I came to Santa Fe and hitched up with FRIAM is
that I thought the people in this group were particularly well suited
to help me solve the problem of self knowledge. what it means to say
that an entity knows about itself. However, while we have had many
interesting discussions, I have never managed to quite get that
subject on the table. So here goes.
Let us imagine that we want to program a robot to do stuff .... many
of you have, I gather. Now, I assume that any robot worth it's salt,
will have a certain amount of self knowledge. It will know, for
instance, where it is. It will know the position of its effectors.
It may also know something about what it has done recently.
So how do roboteers provide their robots with such knowledge.? Now, I
assume, such knowledge gathering is accomplished through sensors. And
while the sensors gather information sufficient for the knowledge in
question given the context, the actual information that they supply is
much more limited. So, for instance, the sensor that senses "the
position of the forelimb" actually measures a current coming through a
resistor, attached to the joints in the limb and a small onboard
computer calculates the position of the limb based on a bunch of
reasonable assumptions about the shape of the robot and the
configuration of the world it is operating in. So even the knowledge
of robots is intentional, in the sense that it is incomplete, based on
assumptions, and from a definite point of view.
There are other questions I want to ask, but let me stop here for the
moment and see what The List has to say.
Nick
Nicholas S. Thompson
Emeritus Professor of Psychology and Ethology,
Clark University ([email protected] <mailto:[email protected]>)
http://home.earthlink.net/~nickthompson/naturaldesigns/
<http://home.earthlink.net/%7Enickthompson/naturaldesigns/>
------------------------------------------------------------------------
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org
