What’s going on: In October, Christie’s will become the first auction house to sell a piece of art created by artificial intelligence, Artnet news reports: a portrait developed by a French collective named Obvious.
- Obvious sold a painting in the same series to a private collector for about $12,000 in February.
How it works: The pieces were made using generative adversarial networks, or GANs, a state-of-the-art AI technique.
- The painting algorithm took a crash course in portraiture, reviewing 15,000 portraits painted between the 14th and 20th centuries.
- A second algorithm called the discriminator judged the result to see if it could discern AI- from human-created art.
- If it could, it sent the generator back to the easel, and the cycle repeated until the generator produced something indistinguishable from human art.
The final products may not be exactly human-like, but they draw a viewer in. One member of the fictional Belamy family, Edmond, is an inscrutable, slightly pixellated ghost. Another — the count himself — looks like a wispy, glowing J.S. Bach.
The backstory: Computers have been making art for decades, from a system called AARON that has been drawing since the 1970s to "The Painting Fool," an algorithm that creates art based on the emotional content of news stories.
Humans are bad at telling human-painted art from computer-generated art, a 2017 Rutgers study found. Artworks created by GANs tuned to simulate creativity fooled humans into thinking they were human-made 75% of the time.
The big question: What does it mean for computers to be creative?
- Computer-generated art generally needs a seed: Some inspiration from which to start. For The Painting Fool, it’s text; for the Rutgers system, a random input.
- But Ahmed Elgammal, director of Rutgers’ Art and Artificial Intelligence Laboratory, writes that humans need inspiration, too, and that artists draw on their experiences and other art to seed their own creations.
Go deeper: A thoughtful 2016 piece in MIT Technology Review.
