I gave this scenario in the talk: imagine you were a paleolithic craftsman making handaxes. Now, you have a competitor making them handaxes too. We'll assume you make a better hand axe - sharper, cleaner lines and more wedge-shaped than the competitors. This leads to a higher "usefulness" than your competitor's handaxe.
You consistently make better handaxes than your competitor. You get better mates than your competitor. You have a higher probability of spreading your genes. Your "good" handaxes has become a proxy signal for your skill in making handaxes, which itself is a signal of your intelligence.
Over time, if the population uses only the "goodness" of the handaxe as a deciding factor in deciding mating partners, the population needs to be able to discern "good" handaxes from "bad" handaxes. As the lines get smoother, we become more and more tuned towards this standard of "beauty", and we put more and more reverence towards the makers of such items of beauty.
In short, humans see math as beautiful because it is a signa for intelligence, which is a deciding factor of fitness in our evolution. Indeed, our evolution seem to value intelligence over strength anyway.
Now, all the above is personal deduction of course. But I believe if we do go on and figure out the evolution of the medial orbito-frontal cortex of the brain, we'd see it's tied to intelligence.
[0]: https://speakerdeck.com/chewxy/beautiful-python - not very useful slides without the talk. The 4 criteria can also be translated to: usefulness, simplicity, averageness (I used the example: if the man on the street can read your python code and understand it, you did a good job), cleverness
[1]: The Art Instinct by Dennis Dutton
[2]: http://www.ted.com/talks/denis_dutton_a_darwinian_theory_of_...