[nopinsight]

There are many reasons to use a representation that is "similar" in some way to the real thing. These include sharing inference mechanisms to predict characteristics like physical movement, affordance, etc among a number of entities when they are represented consistently. (e.g., snakes are represented with an elongated shape; eggs, circular.)

In theory, it could be possible to devise a system of representations that have such desired properties despite their differences from reality. But I suspect that in practice the most efficient choice would be to mimic certain essential features of reality.

[camelNotation]

You can suspect that, but Hoffman's research doesn't support it.