Week 1: Too large to represent
The lecture sets the constraint and nothing else. A finite system inside a larger world cannot hold that world, so it holds a compression, and compressions lose things.
Two formal results do the work, both older than the cognitive science that needs them. Ashby’s law of requisite variety (1956) says a regulator needs at least as much variety as the disturbances it must counter — so you cannot regulate what you cannot represent. Conant and Ashby (1970) go further: any successful regulator must contain a model of what it regulates.
Shannon supplies the other half. Rate-distortion theory is the mathematics of deliberate loss: given a fixed capacity, what is the least damage you can do? That is the question every system in this course is answering, whether or not it knows it.
I close on the evidence-grading notation, which is the only piece of housekeeping in the lecture programme and the one you will use most.
Teaching team
- Halvard Emms — convenor