As you suggest "how attractive" is more nuanced than a 0-10 score on a linear scale. As I see it, this is no different than other recommendation systems. e.g. Netflix knows what you watch, knows what other people watch, and can make recommendations based on commonality in patterns. But each person is unique, and recommendations can take on this nature as well. Some users of the site may be relatively insensitive to physical appearance, while others might respond preferably to individuals with certain characteristics that can be quantified by a statistical model (whether this slice looks like Chris Brown, Pavarotti, or both). I'd also expect that this type of analysis will become increasingly common in dating sites, but elsewhere too. Anywhere that there are quantifiable measures that can be attached to people, and commonality established between people, there is the potential for "intelligent" recommendations to be made. Whether this works well in practice is another story, but time will tell. It's also the case that the devil is very much in the details for this type of work, so it's not as though this study means that this "problem" is solved.