The following article originally appeared in CNN and was written by Rachel Metz of CNN Business
The other day I found myself looking at a startling image of my Northern California home, the daylight dampened by an eerie orange glow as wildfire smoke blocked the sun.
It looked exactly as it had on September 9, 2020, when the entire San Francisco Bay Area was shrouded in orange smoke — a scary scene I hoped I’d never see again. But this time the disturbing sight was generated by a new website with the help of artificial intelligence. The goal: to draw attention to the perils of our changing environment, and inspire people to take action against it, by showing how climate disasters would look in our own backyards.
On ThisClimateDoesNotExist.com you can look up any address — your house, landmarks like the Arc de Triomphe in Paris, Times Square in New York or San Francisco’s “Painted Ladies” homes — and get a surprisingly realistic sense of what it could look like if that place was struck by flood, wildfires or smog. The website, which does this by using AI trained on images of such scenes to re-imagine pictures from Google Street View, was created by researchers at Mila, an AI research institute in Montreal.
The website, which was unveiled Thursday, has been in the works for two years. It began with the realization that while we have tools to face climate change, an enormous obstacle to dealing with it is public awareness and political will, said Yoshua Bengio, a professor at the University of Montreal and founder of Mila, who also led the research team for the project. Bengio, a Turing Award winner, said the researchers wanted to create images that felt personal, which led to the idea of using AI to show what your house might look like during an environmental catastrophe.
“Citizens in the past have been hearing about climate change coming from scientists, reports, and graphs,” Bengio said. “And there is a cognitive aspect, which is, something doesn’t scare us so much if it’s not right in front of our eyes.”
Climate scientists reported in August the world is already around 1.2 degrees Celsius warmer than pre-industrial levels. Temperatures should stay below 1.5 degrees, they say — a critical threshold to avoid the most severe impacts of the climate crisis. With every fraction of a degree of warming, the consequences of climate change worsen. Even limiting warming to 1.5 degrees, scientists say the kinds of extreme weather the world experienced this summer, including flash floods and more devastating hurricanes, will become more severe and more frequent.
ThisClimateDoesNotExist can show you such images that make your home or, say, the Leaning Tower of Pisa appear to be truly flooded or covered in smoke or smog. But creating these visuals isn’t as simple as placing an image of water in front of a building or adding a diaphanous filter.
There aren’t many pairs of images out there showing homes before and after a flood — the kind of data that would be helpful for training an AI system on the relationship between the image it’s being fed and what it should turn it into. To compensate for this, researchers started by building a computerized virtual world. This world, the equivalent of several blocks of a city, let them control flooding and other elements so they could create synthetic images of places “before” and “after” a climate catastrophe, said Sasha Luccioni, a postdoctoral researcher at the University of Montreal and Mila and a lead researcher on the project.
This synthetic data, along with real images showing flooded houses and pictures of smokey orange skies and smog, was used to train an AI system to take any given image from Google Street View and make it look like a climate catastrophe was at hand. To do this, the system needed to first learn where, for instance, water should go in a given image, and then essentially paint water in a realistic-looking way, including reflections of objects poking out of the water and considerations for the lighting of the image.
After a user types in an address and ThisClimateDoesNotExist generates images, the website encourages the user to share them with others and presents resources to learn more about climate change and fight it.
“I think what we want is to channel that initial, like, ‘Oh man, my house is underwater,’ into climate action,” Luccioni said.