No sex please ...
> "Sentient Ascend uses the same concepts to “breed” together the individual changes in a design layout in a “million potential combinations across multiple pages” to determine which is best."
Ok, so they're using genetic algorithms. Been there, done that - the easy bit.
> "Poor combinations don’t make it through the selection process."
Ah, now there's the hard bit. Imagine you're using a genetic algorithm to generate a system to control the velocity and course of a payload so that it docks with the ISS in a specified time. It's a finite problem, you can exactly specify the starting conditions* and you can build a simulator that the selected programs can be tested on. After a lot of iterations, you'll start seeing programs emerge that can control the docking procedure. Given long enough, you may find that the selected programs can deal with a larger range of starting conditions, than the human generated programs that the ISS collaboration uses.
However, in the case of website design, we don't know the right answer. There probably isn't one correct solution and we certainly don't know how to specify it and select for it. I'd guess that they've just coded a system that uses expert recommendations to generate the selection rules. As long as the experts are HCI experts rather than pointy haired bosses, the resulting website may be an improvement.
So my guess is genetic algorithms selected by an expert system.
The pointy haired boss will, of course love it, use it to generate a website, then use a team of developers to tweak it so that it "looks and feels right". Oh well, at least we keep our jobs until AI can replace the pointy haired bosses.
*Newtonian physics and engineering data will suffice.
Icon: let's leave sex to the experts.