Where machine and human preference collide: EyeEm's aesthetic algorithm

Or a consideration of what makes something beautiful

In July this year, image sharing (and selling) community EyeEm launched its smart new 'Discover' feed in its app. 'Discover' is a mechanism designed to help EyeEm's incredibly active community engage better with each other through a combination of inspiring images and editorial content.

The Discover feed provides both a selection of inspirational images and editorial content to motivate and inform its users. The editorial content aims to help people improve their photography–which has long been a cornerstone of EyeEm's philosophy–while the image selection is based on location data and by using EyeEm's aesthetic algorithm.

EyeEm is understandably proud of its aesthetic algorithm, and not just because it has been seven years in development, since it formed part of Appu Shaji's PhD research, but because it believes that the aesthetic algorithm does something that no other image recognition software does: identifies abstract concepts. 

EyeEm's aesthetic algorithm can tell you if a photo includes themes of happiness, or serenity, or excitement, among many others. This is a huge advancement on the image recognition software with which we are already familiar, which can identify people or objects, or even the application of photographic principles such as the rule of thirds or negative space. 

The aesthetic algorithm is valuable both as a creative and functional tool.

First, an image recognition programme than is able to accurately label photos beyond who or what appears in them has the potential to relieve anyone of the monotony of tagging and keywording their photos. Clearly this benefits anyone, or any organisation (such as EyeEm), that sells images and depends on them being accurately identifiable. If the process can be efficiently standardised, that's a huge boon. And even if you don't sell your photos, being able to locate them through the use of accurate keywords and tags is mightily useful.

Second, by using an algorithm that can identify images perceived as 'beautiful', there is a better chance of the Discover feed unearthing photos that deserve recognition but might otherwise go unnoticed. Not everyone is great at keywording, for a start. But furthermore, not everyone has a legion of followers to admire and appreciate their work. The aesthetic algorithm is, then, an effective means of negating the popularity contest of likes and upvotes that fuel other online sharing platforms, and with it the ensuing echo chamber of social media gratification and validation.

Anything that can relieve the monotony of image labelling and the isolation and clique-i-ness of online sharing is worthy of praise and investment.

But, the aesthetic algorithm does come with its own baggage. And for me that revolves around our humanity, and the conflict of using a machine to 'see', identify, and grade the product of a process that is inherently visceral. 

Haje has already articulated the first of my concerns in an article that he wrote in October 2014. Photographs are about telling stories, and it is this inherently human aspect to them which makes them valuable. Until a machine can identify and interpret a story within an image, then it falls woefully short in analysing the critical factor in the purpose of photography: narrative.

My second concern focuses on the notion of 'beauty', and to what extent an algorithm is capable of deviating from its own self-imposed notion of what is or isn't beautiful. Aside from the myriad variations that exist in people's tastes and preferences, which makes the same image both beautiful and ugly to two different people (and would somewhat compromise the purpose of using the aesthetic algorithm in the Discover feed), can an algorithm account for context, nuance, and variation? These subtle variations in interpretation, in storytelling, and in perception are what contribute to making some photos speak to someone, but not others. So can they speak to a machine?

Even if an algorithm is capable of learning what given people prefer (and I have no doubts that EyeEm's aesthetic algorithm can and does) and successfully deviates from the parameters of classical beauty, this presents another problem. Learning and development are dependent on challenge, even provocation. If we are constantly presented with a stream of images that an algorithm has identified as being to our taste, it restricts our ability to learn and grow. Why do we find particular combinations or scenes or ideas attractive? Why do we find others ugly? Without the opportunity to see the alternatives, we have no means to question our preferences, to consider alternatives, even to challenge dogma.

As uncomfortable as it might seem, it is good for us to confront what we don't like and to constantly assess what we find beautiful and why. There are many lives unseen, and we shouldn't be afraid of seeing them. Or to put it another way, we mustn't allow an algorithm to allow us to slide from the social media echo chamber of confirmation bias to a 'computer says yes' utopia of beauty.

The aesthetic algorithm is a wildly impressive development. And it can and should be refined and honed and used. But it comes with an enormous caveat. Photography is an essentially human practice. We take photos to make memories, to present our version of truth or beauty to the world, and to tell stories. If we allow an algorithm to determine what it is that we can and should find attractive or valuable in a photograph, then we rather undermine the process of making it in the first place. 

The benefits of the aesthetic algorithm need to harnessed and put to work, but in conjunction with actual human beings. I love the idea of being shown a stream of images that I might not otherwise see, which have been identified on the basis of something other than 274 other people liking them. But I can't become reliant on an algorithm turning up images that it thinks I will like, and neither should anyone else. Would I prefer a majority of images that make me go 'Oooh?' with the odd one that induces 'Ahh!' and the occasional 'Eww!'? Yes, I think so. 

Part of the pleasure of photography is the human connections that it records and fosters. That is what drives it. Photographers should never be afraid to step off the beaten path, both as producers and consumers.