A Brooklyn-based artist is using DALL-E to rethink roads so that they are more friendly for bicyclists and pedestrians.
What would a promenade bordered by trees and plush grass look like in place of a six-lane highway? Now, DALL-E can simply demonstrate that.
DALL-E creates photorealistic graphics based on a text input and was developed in the San Francisco lab of the artificial intelligence startup OpenAI. Additionally, you can send an image and request that the AI replace a certain portion of it with something else.
The 28-year-old former open-street activist filled in how he wanted to replace the pavement and parked cars with a “strikingly beautiful cobblestone European pedestrian promenade, with an ornate stone water fountain and children playing” after selecting the elements to be removed.
DALL-E had given him its opinion on the redesign in a matter of seconds. It looked quite beautiful.
Thanks to a massive amount of training data in the form of images and their descriptions, the technology, known as a neural network, is able to convert language into features and illustrate how these qualities may connect to one another. In January 2021, the first iteration of DALL-E, a combination of the surrealist artist Salvador Dal and the Pixar robot WALL-E, was released.
The AI-generated photographs have been making the rounds of social media, and despite occasionally being delightfully bizarre, they are also remarkably lifelike. By using text searches like “in the style of Mark Rothko,” users can specify the styles they want. Additionally, they might request details they’d like to see in the result.
I thought it was really cool, what it was spitting out, Katz said, and I was kind of blown away at the transformative potential it has for helping people envision better cities.
Katz set up a Twitter account on July 20 and began sharing his works there. It exploded. He currently has a backlog of hundreds of requests for him to remodel streets using AI coming from all over the world, from Thailand to Canada.
Visualizing things is the most powerful way to effect change, Katz said. Before, to make renders of a street, it’d cost hundreds if not thousands of dollars and take days or weeks to produce.
In New Port Richey, Florida, where Vox recently labeled US-19 “the deadliest road in America” for pedestrians, Katz asked DALL-E to remodel it. His instruction was to draw a large avenue in the style of the Champs Elysees with bike paths on both sides, a streetcar in contemporary European design, and apartment buildings with a variety of uses. The outcome? Very convincing.
My favorite one I’ve done is Stuart Road, in my hometown of Herndon Virginia, Katz said. It’s one of the most unassuming on the page: It’s literally just adding sidewalks. Katz’s parents have lived there for close to three decades and complain about the lack of sidewalks.
There are issues with these AI-generated photos. The results may be biased towards women and persons of color due to a lack of diversity in the training data. Some experts are worried about the photorealism of the output because it may be used to propagate incorrect information in an overly convincing manner, blurring the barrier between fiction and truth.
Katz hopes that his work will motivate both activists and decision-makers and lead to local change. He’ll continue processing his backlog of requests in the meanwhile.
The most amazing part is feedback from people that I never expected to chime in, he said. One very notable example is the mayor of Sheboygan, Wisconsin. He tweeted at me and asked me to reimagine some streets in his hometown. Council members and urban planners from all over the country seem to be very inspired by the images, Katz said. I think this is how it will lead to change: people with real decision making power being influenced and inspired by these images.