Chunks enter declarative memory either as popped goals, reflecting the solutions to past problems, or as perceptions from the environment.
The perceptions from the environment pieces is exciting and I think very important.
Productions are created from declarative chunks through a process called production compilation.
Can productions be created dynamically, and run-time? Seems cognitively plausible since people acquire not just new facts but new ways of doing things, tieing knots and so forth.
ACT-R also has a sub-symbolic level in which continuously varying quantities are processed… to produce much of the qualitative structure of human cognition.
I’m surprised that AI has had so much emphasis with relatively little attention paid to simulating human cognition. All the AI I’ve ever studied has a scant resemblance to cognition compared to this. Supervised learning, bayseian networks.
B sub i and S sub ji are quantities that participate in neural-like activation processes that determine the speed and success of access to chunks in declarative memory.
Similar quantities control access to production rules.
these quantities influence things like speed of retrieval of the number facts and choice among alternative strategies for doing multi-column addition.
Choice among alternative strategies = interesting
ACT-R also has a set of learning processes that can modify these sub symbolic quantities. These … serve to turn ACT-R to the statistical regularities in the environment.
Explained more in chapters 3 & 4
If the secret of science is carving nature at the joints, we propose ACT-R’s chunks and productions as defining the critical cognitive limbs.
This is very interesting. If not for CMN, I would not be thinking of HCI as all that cognitive.
Just as the identification of the atom
This is a great point to ponder. Modeling sounds a lot like ‘carving nature at the joints’. Is that a popular expression?
the ‘no-magic’ doctrine of ACT-R
What is this about?
Leave a Reply