[Martin Taylor 981227 17:37]
[From Rick Marken (981227.0940)]
Richard Kennaway (941226.2243 GMT)
What are beliefs, then?
Bruce Gregory (981226.1835 EDT)
Good question. Maybe Bill or Rick will enlighten us.
I think a belief is the reference state of an imagined
perception.
I don't agree with this. I think a belief is just a perception. You can
act on the environment to change a belief just as you can for any other
perception. You can have a reference state for a belief (a belief that
you would like to be able to hold) that is different from the current
state of your belief. I would like to believe I understand what people
are trying to do, but I don't believe I understand, so I act to try to
learn more. And so on and so forth.
Why use the two terms "perception" and "belief," then? The words have
different connotations. If I were to give PCT technical meaning to
"belief" I would take beliefs to be a subclass of perception. All
beliefs are perceptions, but not all perceptions are beliefs. Perceptions
that are not beliefs are primarily based on current sensory inputs.
Beliefs may be influenced by current sensory inputs, but they are more
dependent on "imagination" and memory.
To say this requires that "imagination" be clearly defined. In PCT, there
is an "imagination loop," but I don't remember a definition of what
imagination is. In some way, it must depend on modelling the environment.
The "imagination loop" provides the perception that would be produced
by some action if the action were to be performed. So, imagination must
involve the interactions of perceptions, but __not__ through the output
functions of the Elementary Control Units controlling those perceptions.
This doesn't define "imagination" but it suggests a characteristic that
imagination must have, if the term means anything in PCT.
If this is reasonable, it implies that beliefs depend substantially on the
interactions of perceptions, perhaps also on the interactions in imagination
of the control loops through which the interacting perceptions are
controlled.
In my Layered Protocol papers for the IJHCS special issue, I used the term
"belief" to mean a cluster of interacting perceptions, rather than a
single scalar-valued perception. In that usage, a belief is a complex,
vector-valued perception. I don't think that's an important difference
from saying that belief is just a subclass of perception, since any vector
can be part of an input function that produces a controllable scalar
perception.
I don't see why a belief has to refer to an __imagined__ perception,
whether it is a reference value or an actual value. I can change my beliefs
about states of the world by watching or reading the news, by observing
other people or the waving of tree branches, or whatever. These all may
create disturbances to my beliefs, if I am actually controlling them.
My version: a belief is a perception that has imagination as a major part
of its structure, whereas low-level perceptions depend more strongly on
current sensory values.
Martin