Cognitive Architecture

To a large extent, implementing a Cognitive Architecture with CAs is the big problem.
A complete architecture would solve the Turing test and thus the big problem.
However, I need a more graded approach.
1. CAs as categorisers. A good basis, though still a lot to do.
2. Variable Binding. Current problem. This enables rules.
3. Together, categorisers and variable binding give a rule based system that can be a reasonable Cognitive Architecture.
4. Fill in details. Timing, layering, range of neurons, neuro-transmitters, CA size, etc. See back burner
5. Explore subsystems, and better systems. Just started thinking about this.
How would this help someone design an interface?
Clearly we're a long way from this now, but it could be used to simulate learning and reactions to the interface.
How could this help the designer now or shortly?
Perhaps a model of recognition based on CAs would give a better answer than current answers. This could also provide support for the architecture.