[ACT-R-users] computational basis of act-r

[Phil Pavlik]

It seems to me that it is a mistake to expect ACT-R to compete with
oscillation theory (connectionism) by asking for a close correspondence
to brain functioning.  While some correspondence is desirable, a close
correspondence will make our efforts at modeling high level processes as
difficult as the connectionist's efforts.

The beauty of ACT-R is that it represents neural processes (even at the
subsymbolic level) in the aggregate rather than try to describe the
oscillations of individual neurons.  It's strength is that emergent
properties of the brain do seem to be well represented by these
aggregate models we construct.

ACT-R, in my mind, complements connectionism, and only competes because
it is human nature to compete.  I see ACT-R being more effective and
explanatory for high level modeling for the foreseeable future...

I have read the first few chapters of O'Reilly and Munakata's book and
several papers by McClelland, and I fail to see that in 2 or 3 years
they will be modeling tasks like CMU-ASP.  When their models are complex
enough to handles such tasks, I wonder whether they will still be
tractable enough to be suited to training applications.  

Philip I. Pavlik Jr.
Psychology Department
Carnegie Mellon University
Pittsburgh, PA 15213
 
ppavlik at andrew.cmu.edu
http://www.andrew.cmu.edu/user/ppavlik/


-----Original Message-----
From: act-r-users-admin at act-r.psy.cmu.edu
[mailto:act-r-users-admin at act-r.psy.cmu.edu] On Behalf Of Hongbin Wang
Sent: Tuesday, January 14, 2003 11:16 AM
To: act-r-users at act-r.psy.cmu.edu
Cc: Hongbin Wang
Subject: RE: [ACT-R-users] computational basis of act-r

Shadows of the mind

When the mind works, it leaves shadows that people can observe. Suppose
a
student S solved the equation "2x+5=9" and it turned out the response
time
was 2s. Both researchers JA and RO tried to understand what had happened
in
this 2s period. JA looked S' verbal protocol (e.g., "I move 5 to the
right,
and do the substraction. So I get 4, ...") and developed an Act-R model
to
simulate the underlying knowledge structure of S' behavior. RO looked S'
fMRI recording and found the medium temporal lobe is particularly
involved
in the process. Based on well-documented principles of how the brain
works,
RO developed a neural networks model to simulate S' behavior. While both
models fit the data well, the question is, which one is right?
Apparently, both models are valuable. While the Act-R model is
psychologically plausible the neural networks model may be more
biologically
realistic. Both models captured some characteristics of the mind's work,
though at different levels (i.e., symbolic and subsymbolic).
Importantly,
these different levels are all "real", and no one is more real than
another.
A cognitive architecture that permits psychologically plausible symbolic
modeling and makes nontrivial predictions will not be outdated. 
How to find the possible links among these different levels of
shadows/models is the hard (and right) question. Act-R is already moving
forward in this direction. Act-R5's symbolic components have been mapped
to
possible brain structures based on recent neuroscience results. Though
rudimentary, it may not be too far away in time for people to start link
Act-R to brain processes such as oscillation. The link would not be
straightforward and linear. But as long as they all are shadows of the
mind,
they should be linked.


Hongbin Wang
School of Health Information Sciences
University of Texas Health Science Center at Houston

-----Original Message-----
From: Chris Chatham [mailto:chatham at m-laboratories.net]
Sent: Tuesday, January 14, 2003 6:55 AM
To: act-r-users at act-r.psy.cmu.edu
Subject: [ACT-R-users] computational basis of act-r

A couple of days ago, I talked to a researcher at Penn's Institute for
Research in Cognitive Science who believes ACT-R will be outdated in the
next couple of years.

He believed that the fundamental method of computation in the brain is
"oscillation" and that because ACT-R has no computational similarity to
the
neurological structure of the brain, it will always be a poor modeling
architecture.

I asked whether ACT-R might be expanded at the subsymbolic level to
include
this type of modeling.  Any thoughts here from the group, or in regards
to
the IRCS researcher's opinion?

-Chris.


_______________________________________________
ACT-R-users mailing list
ACT-R-users at act-r.psy.cmu.edu
http://act-r.psy.cmu.edu/mailman/listinfo/act-r-users

_______________________________________________
ACT-R-users mailing list
ACT-R-users at act-r.psy.cmu.edu
http://act-r.psy.cmu.edu/mailman/listinfo/act-r-users