Screenshot of Xerox Aesthetic Search

Xerox’s quest to strip all originality from Photography

A couple of months ago I came across an article on how Xerox have developed a new algorithm to sort photos based on their aesthetic qualities, you can try the free online demo here. At the time I thought “That’s cool!”. Then I gave it some more thoughts and realised that this is a cool technology, but not how they are using it.

On the surface the way Xerox are using this technology looks great, “what a great method to sort the wheat from the chaff”. Just try the demo and it looks impressive. I am certain the intentions of the research team at Xerox are good, however here is a concern with this technology. I have posted about my thoughts on Algorithms before. The problem with algorithms like this is that they rely on rules and rules have to be based on textbook or conventional wisdom. So in photographic terms these are rules like the rule of thirds, the conventional belief that shallow depth of field = better that colours should not be dull.

Great, but also a load of BS (nonsense). The best images break rules. So when sites like Flickr licence this technology  and incorporate it into their Explore algorithm, the selected images that are output are even more homogenised and boring that they currently are.

But wait, there is a solution

The main issue with these algorithms at the moment is that they are based on rules designed to please the masses, but rules don’t have to be fixed. What these algorithms need, is to be engineered in a way that they are personal to you, the viewer. They need to learn each time you like an image, establishing a ruleset that is dynamic and changes over time. Also important is that they adapt to ones changing tastes and also throw in the odd curve ball that breaks your own rules, allowing you to explore new boundaries.

We do need to find ways to identify the good images though all the uploaded noise, but blanket filtering like this is the wrong approach. What we need is personalised discovery algorithms to help us find new refreshing images that we, personally, like. Even then the computer is a long way off from identifying the emotional or intellectual qualities of an image. So for now the human element will not be replaced.