Right, and then you could argue that people who actually think of themselves as "smart" will know of such experiments, and will thus rate themselves as "average," knowing that such a label actually indicates that they really are smart.

So basically this kind of survey says very little at all.