I'm actually planning to start a series of articles on Norman's book, taken from a very explicit system and level design point of view. ( http://is.gd/1oKo ) I think I'll start with a principle other than affordances now, however.

This topic fascinates me, as I think there's a tie between affordances, simulation depth and the uncanny valley. The more options you allow, the more the small things you didn't include will stand out. Sort of along the 'simulating near the boundaries' thing Clint Hocking wrote about some time ago.

I'm all for more player freedoms and systems that can support exploration, but like you express, I think objects should be capable of the affordances they imply within the system constraints the player resides.

Interface is key here, and consoles definitely have a more focused set of inputs than a keyboard. There is the whole "context sensitive menu" thing, and I think that's superior for players building a system model in their heads than a blinking cursor. I'm also wondering if the whole 'motion plus' thing, coupled with a decent physics engine could give us a sufficiently granular player hand to interact with the world in a meaningful way, assuming there are some checks too keep out Trespasser-like frustration.