[galeforcewinds]

I agree that the number of data points needed for study and the number of variables involved are both key factors. The one area that comes to mind where a approach might be easiest is closed platforms like the hardware Kindle reader -- for example, measuring read times by a large population of the same text, with varying fonts, on the same hardware. Improved typography might reduce read times some modest percentage over baseline due to improved readability.

[brianpan]

Sure you can use ML and test subjects to maximize readability, but that's only one part of typography. Akin to Google beta testing shades of blue. http://www.zeldman.com/2009/03/20/41-shades-of-blue/