Its an assumption of the IQ Test. The test itself assumes a Gaussian distribution.
IQ Test was a quick-and-dirty testing methodology made in the dark ages of psychology. It turned out to be a decent measurement of general intelligence, but its theoretical fundamentals are... kind of shoddy. I mean, its better than a lot of early 1900s psychology... but its still from an age when psychology was more folk-science than real science.
Fortunately, the IQ Test actually seems to measure something... and due to its age there are a large number of results that span over a century. And the test has been relatively standard throughout the time (with exception of its increasing difficulty, as our population has grown smarter on the average)
The normal distribution simply implies that there is some sort of "average" or middle area, and that the further you go from it both sides (smaller/larger, faster/slower, etc), it becomes harder and harder to find instances of that measurement.
It's just the way it is, and I'm probably not smart enough to explain it to you in a concise-enough way, but that's the gist of it from my understanding.
I'd expect intelligence to be more of an asymmetric distribution, with the vast majority of people being above the threshold of, say, catatonia; most people being under "genius" level, and a blip for geniuses that have really changed humanity's capabilities as being pretty far to the upper bounds, but with incredible impact on the group. Something like a "6 sigma" intelligence on a normal distribution would probably show up a lot more often on the skew distribution I have in mind.
Although, this might be measuring something like E[f(intelligence)] where f is a convex impact-measuring function of intelligence