Think about the memory-prediction framework: it's all layers over layers over layers of simple predictor elements.

What a simple predictor does? Gets input from a lower layer, recognizes spatial and temporal patterns and passes up a new signal.

I assert that the passed up signal is a simpler/ more compact representation of the received (and recognized) pattern.

This makes the predictor a simple (or not that simple?) compressor.

And makes the whole memory-prediction framework a huge complex compressor (except for the other details such as prediction).

And what we know about compression? It works on non-random data.

Then the memory-prediction framework can be viewed as a complex compressor.

And if the brain works anything like it, then the brain is a complex compressor.

And the data it compresses is the input received from the world. Which is not random as the signals are governed by the regularities and rules found in the world.

So the brain compresses the world. :-)

I wonder how compression relates to understanding, intelligence, thinking.

Ideas to check out:
  • Jason Hutchens and his predictor
  • Information Theory (entropy, information content, surprise)