Advantages of a CA based Architecture

Turing Complete
1. Our experiments and others have shown that CAs are pretty good categorisers.
2. We're hoping to work on Variable Binding, but Sounge (2001) has already shown that plausible neural models can be used for variable binding.
3. We're also working on moving from the single attractor to new places in a biologically motivated way.
4. These things (along with an infinite number of neurons) are all that is needed to have a Turing Complete system.
Better Blackboard
1. Subsystems can be modularised (by limiting connections).
2. However, subsystems can also be combined in a limited fashion.
3. Hopefully, this can give us some data hiding, but also some good inter-module communication.
4. This gives us a better blackboard
Biological Guidance
1. We can use mammalian brains as models to solve problems.
2. E.g. if we need to solve Variable Binding, see how people (or mice) do it.
3. The search space for possible intelligent architectures is enormous.
4. CAs provide an intermediate level between neurons and full systems.