[ACT-R-users] RE: [soar-group] ACT-R/Soar on a robot

[Nick Cassimatis]

Troy,

Many of your specific examples involve decay.  Many AI people see
memory decay as something you would NOT want to give a system, but it
has at least one useful function that I have found in my modeling of
infant physical reasoning.  It keeps less-relevant memories and
knowledge from impeding on your current thought.  For example, while
tracking an object move from point A to B, you perceive and form
memories of intermediate motion events from A to p1, p1 to p2, p3 to
p4, .... pn to B.  For many tasks, those memories of intermediate
motion events are irrelevant and having them inhibited helps.  So, in
AI terminology, decay can behave as a useful heuristic.  The reason to
not completely eliminate the memories of intermediate events (from a
normative AI perspective) is that new information can make the
relevant so you do not want to discard them.

> skip over rules, to speed up the procedure.  But really, 
> psychologically, it seems as if the symbolic level rules are being
>“rewritten” or “re-compiled” as subsymbolic representations.  On our 

Maybe it just seems like this because the increase in speed makes them
less accessible to what self-reflection, what ever that is.  So
introspective difference might not be accounted for by a qualitiive
change in the rule matching, but instead by the limits of
introspection.
 
> 7) Is cognition sensory specific?  Much of our code right now for 

You can structure a system so its productions match against buffers
that represent "mental images" that are to some degree isomorphic to
your actual sensor input.  So, on the one hand you have the supposed
flexibility that comes form reacting to new stimuli, but on the other
hand you have the added flexibility and power of reacting to/thinking
about occluded, past and future parts of the world.

Finally, many of your questions initially sounded dangerously vague
and philosophical to me, "when is a memory a memory?", but you made
sense of what you meant by referring to specific design features of
models.  I think avoiding potential confusions and pointless debates
such as this could have been is itself a good reason for doing
cognitive modeling.

Best,
-Nick