CAs as Categorisers

In a relatively simple simulated system a CA is a categoriser.
The environment is presented.
Input goes into the system, and the category that best matches the input is ignited.
That categorises the input as an element of that CA.
So, if I've got a system that takes pictures of buildings (Knoblauch 2007 Neurocomputing), I show it a building, and it's church CA ignites (the reverberating circuit becomes active), it has categorised that picture as a church.
I've built quite a few categorisers.
Typically, they use some sort of unsupervised Hebbian learning rule.