What does the Mission look like to an artificial intelligence?
Chimeric bird-snake creatures.
Tamales sold from roadside carts.
When artificial intelligence DALL-E was asked to portray the Mission, these are some of the motifs that were conjured from its digital depths.
Tech company OpenAI, which created DALL-E, is based in the Mission on Folsom and 18th Streets. Last weekend, it opened up access to its AI to the general public, allowing anyone and everyone to try out the model. Businesses that use the service extensively are charged, but nosey journalists or members of the public can poke around for free.
It works like this. Users enter a text prompt – for example, “The Day of the Dead procession in San Francisco’s Mission District, in the style of Vincent Van Gogh” – and, after a few seconds of computation, DALL-E spits out an image to match your description. In this case:
DALL-E creates its pictures using a process a bit like a super-powerful visual autocorrect. Working backward from a random assortment of pixels, it tries to build up a picture that is likely to match your text prompt. It can understand how text and images are related because it has ingested a vast number of image and caption pairs from across the internet. Apparently, the math behind its operation is laid out here, although to my non-engineer brain it may as well be witchcraft.
The technology is not without its ethical quandaries. Because it is based on publicly available data, the images it produces can replicate biases seen in the wider world (for example, its pictures may represent men more often than women). AI art has also been criticized for displacing human artists, as when AI-generated art won first prize at the Colorado State Fair.
Nonetheless, its interpretations of the Mission – its home turf – are striking. Without further ado, here is a glimpse of our neighborhood through the eyes of an artificial intelligence.