Weather patterns and voting data are used to explore neural networks in the Serpentine Gallery’s latest Digital Commission, this time by artist James Bridle. Finding potential connections between weather satellite images and six years of UK polling figures, Bridle’s Cloud Index is inspired both by numerical weather prediction and the new wave of machine learning algorithms.
“I’ve been fascinating by the history of computation for a long time,” Bridle tells The Creators Project. “Like where did all these machines that we use everyday, that are now kind of invisible to us, emerge from? They mostly emerged from military technologies, but also from the desire to control and understand the weather. There was this long race to build better models of the world and weather is a perfect example of that: a vast global system that we’re desperately struggling to understand.”
Known for producing work around technology’s influence on human behavior, from drone shadows to going inside restricted zones with VR, Bridle’s Cloud Index is a metaphorical political mapping tool, illustrating weather based on voting indications surrounding the EU Referendum.
“So I took all the available data that I could find—polling intentions, opinion polling leading up to the referendum—and satellite images from weather satellites over the UK, putting all those things through a neural network,” says Bridle.
Understanding the function of Cloud Index’s neural network—called deep convolutional generative adversarial network (DCGAN), described in early 2016 by machine learning researchers—is what interests Bridle. Unlike the operation, which sees neural networks mimic human brains in order to predict outcomes (using facial recognition technology to determine faces, for instance), DCGAN works through two networks simultaneously, which learn and improve by bouncing off each other.
“It’s basically a kind of neural network that can associate data and images,” explains Bridle. “It understands how those images are put together and produces new images. The classic example is to train it on images of faces. To say, here’s 10,000 images of men smiling and here’s 10,000 images of women not smiling and then it will produce images of smiling women.”
Bridle's work, in turn, plays off the growing mistrust of the political system and democratic decision-making—amplified by the recent EU referendum results—to find new models of prediction and control. “Whether it’s people trading on the stock market, FB arranging the newsfeed, Google predicting your behavior online, surveillance agencies monitoring you, or self-driving cars, the promise of all these technologies is that if we gather enough information of the world then we can build some perfect model that will allow us to predict and control it,” says Bridle. “As committed to computer science as I am, there’s a lot of problems with that. It doesn’t really seem to hold. So I think we need other models of understanding the world that aren’t this purely empirical mode.”
While data collected by Cloud Index allows for multiple ways of predicting the past and future, greater accuracy requires substantially more data for guaranteeing results and thus, altering outcomes.
“These neural networks that we’re developing are kind agnostic,” says Bridle. “You can throw anything at them and they’ll throw information back at you. So there’s a critique required in order to interpret and understand what of this information is useful and valid. What issues do we want to address with these technologies?”
James Bridle's Cloud Index can be viewed on screens at the Serpentine Gallery, and online, here. The artist will speak about the project during Miracle Marathon arts festival on October 8, 2016 at the Serpentine Sackler Gallery.