Beauty Is In The Eye Of The Artificial Intelligence System

‘Beauty is in the eye of the beholder’. But artificial intelligence now ‘thinks’ it has a definition.

Using deep learning techniques, data scientists from Warwick Business School trained a computer system on 200,000 images from the website Scenic-or-Not, where members of the public vote on how beautiful a British scene is. These include Loch Scavaig on the Isle of Skye… and Newbury Road roundabout.

Unsurprisingly, images containing natural features such as valley and coast scored highly.

The project was linked to earlier studies by the same team from Warwick’s Data Science Lab that showed a direct correlation between residing in a scenic location and good health. If the AI could recognise beauty like a human, city planning for wellbeing could potentially be automated.

Deep learning essentially involves inundating a powerful system with labelled information, and waiting for it to make connections, categorise and sort data. In theory, it can then contextualise new scenes using that information. Such a model is designed to replicate, to a lesser extent, the connectivity of the human brain.

The Warwick wanted an objective take on what makes a scene beautiful, and used the MIT Places Convolutional Neural Network to run the task. While going through the Scenic-or-Not images, the system – which has already been trained on 2.5 million labelled images – duly began to label everything it could recognise, from grass to horizon, hills and sky. In doing this, it also noted whether or not they were attributes found in a highly rated scene.

Unsurprisingly, images containing natural features such as valley and coast scored highly, with historical architecture also upping scores. Open expanses such as athletic fields lowered the score.

“Flat and uninteresting green spaces are not necessarily beautiful, while characterful buildings and stunning architectural features can improve the beauty of a scene,” wrote Chanuki Seresinhe, co-author on the research. When tested on 200,000 photos of London, the AI identified famous locations such as Hampstead Heath and Big Ben as beautiful.

The work carried out at Warwick is similar to work carried out by Google’s X Lab in 2012 that trained a neural network made up of 16,000 computer processors to watch YouTube videos and find images of cats.

Google has since trained an AI capable of building its own encryption and creating its own language. Facebook has gone one dystopian step further – when its Artificial Intelligence Research (FAIR) group trained bots to negotiate, they all learned to lie.

The findings from Warwick, so far, have not been revolutionary: uninteresting is not beautiful, ancient churches are. But the study is in its early stages, and points to the system at least being able to understand what many city planners may not have grasped – adding a patch of grass does not necessarily transform a space.

These subtleties in what makes a scene beautiful have “clear relevance for planning decisions which aim to improve the wellbeing of local inhabitants” said Suzy Moat, co-director of the Data Science Lab. But there are other obvious drawbacks to the study, mainly that Scenic-or-Not does not hide within its data the secret to how we perceive beauty.