Kids are a hodgepodge of brilliant insight and cloudcuckooland leaps in logic. I think it's more that they're utterly unfettered by constraints on what is "normal" and still determining how all this seemingly random data they're being given fits together. So they try to fit the data into what models they have to work with. Often they get it wrong, but it makes sense within the framework of what they know.
My 4-year old daughter still hides behind her chair when there's a cartoon monster on the TV screen (like, an utterly unscary cartoon monster -- she's scared because the characters are scared). At the same time, she was coloring a picture of Mickey Mouse and her babysitter said, "Oh, hi Mickey! What do *you* think we should have for a snack?" and my daughter gave her a cold stare like she was an idiot and said flatly, "That's a picture. It doesn't talk."
So seemingly logically inconsistent (realizing that one thing is make-believe but not another) but as I said, I don't think it's that she's scared of the monster in the cartoon, she's scared because she sees others being scared. I guess my point is that children dont necessarily learn things in the same way as adults. Instead of using prior knowledge as a foundation for logical thought, they sorta plug new information at random into prior models. If they get what seems like a good fit, they go with it. If they're corrected or discover that their model isn't consistent, they either:
1. Switch to a different model (often seamlessly, and in a matter of seconds)
2. Ask for additional information ("Why?")
3. Handwave the inconsistencies away ("That's because he's a MAGIC robot. Duh.")
She also announced that she wants to be an eyeball for Halloween, and doesn't understand why her mother and I think that is the most awesome thing we've ever heard.