We tend to think of aesthetic judgment as something particularly human, but a group of Xerox engineers are building an algorithm for finding good photographs
Xerox has unveiled an early version of a program that can discern the content of a photo ("no text tags required or used!" the site says) and sorts photos into "good" and "bad" -- subjective judgments we normally leave up to humans.
The program isn't perfect. Some images are not categorized accurately -- pictures of "Clouds_Sky" includes a photo of a woman dancing in a club, partially obscured by the clouds from a fog machine, and "Beach" includes a paper crane sitting on top of sheet music. Additionally, the "good" photos all look quite similar, the parameters perhaps being set a bit narrowly (surely some good portraits have light backgrounds?). But because there's disagreement among humans about aesthetic quality, perfect is an elusive measure, even if the program were completely accurate. Despite its flaws, on balance the program's picks for "good" images are more dramatic and more interesting (if a bit stock-artsy) than the photos it defines as "bad." (Above, take a look at some samples from the program and the judgment the computer rendered. The images come from thumbnails, so apologies for the low resolution.)
A computer may not know good from bad, but if the factors that set good and bad apart are quantifiable (contrast, clarity, composition), then a computer can imitate that judgment.
Image: Xerox.
We want to hear what you think about this article. Submit a letter to the editor or write to letters@theatlantic.com.
