Human vision is very good at reconstructing the 3D world from the rather limited information the eye sees, and it can also do that when presented with a 2D image of the 3D world. Artists and scientists have studied and exploited these human abilities.
Basically our brain can do edge detection. It can see that the side of the table cloth is a bit darker than the top of the table, therefore the edge of the table is where the two meet. The luminance component of an image is so much more important to us than the chrominance that we spent nearly a century being amazed at black and white photography and cinema. Even our high tech digital codecs use higher resolution for the luminance than the chrominance components.
In my map visualisations I have tried to exploit this the other way around: use the eye's edge detection ability not so much to find the edge, but to distinguish the light from the dark side. See for example the two maps above. Once we put a hard border line, it saturates our vision, we can no longer see if the East of England or Yorkshire are the darkest blue. This is much like walking in a dark park at night and having a cyclist with strong LED lights coming towards you. You can definitely see the cyclist's light but you can no longer see the path or indeed much else.
The map on the left on the other hand suddenly allows us to make the most of the limited dynamic range. We can now see that the East of England is the darkest blue. It is harder to see the border between areas that are the same colour, but this is a Tableau map, that's what interaction is for. After all quite often these maps are for people who know where the borders are, they are trying to see the borders suggested by the data, not the borders known a priori. Adding a hard border is much like a naive painter's approach: putting more effort into picturing what we know rather than what we see.