Right. The thing is there are multiple different ways to represent a piece of information like 'red' that each have different advantages and disadvantages. Pixe

Right. The thing is there are multiple different ways to represent a piece of information like 'red' that each have different advantages and disadvantages. Pixel data in an image file, weightings in a neural net, a flag in a database and many many variations on each of these. An image classifying neural net doesn't just store 'red apple' as a colour value like an image file, but as relationships between all sorts of other aspects of images it has been trained on. It probably doesn't even store 'red' distinctly at all, the weightings that represent red in the classification of a red car may well be completely different from the weightings that identify red in images of apples. Likewise with human brains. So when Mary sees a red apple, that will evoke different responses than she would have to a red square or a red car. It's the relationships between attributes that matter at least as much, if not vastly more than the attribute itself in isolation.