“The woman in the photo seems familiar,” write Cade Metz and Keith Collins in The New York Times. “She looks like Jennifer Aniston, the ‘Friends’ actress, or Selena Gomez, the child star turned pop singer. But not exactly.”
You can’t quite place her because she is not real, they explain. Nvidia software generated the photo based on analysis of thousands of real celebrity photos. The goal of the work, based in a lab in Finland, is to move beyond still images to speed the creation of computer interfaces, games, and movies, accomplishing in minutes what would take human developers days. As of now, though, the software still takes days to create a still image. Click here to see the “faux celebrity photo” and a time-lapsed sequence of the 18-day process that created it.
The Nvidia team employs progressive generative adversarial networks (GANs), in which one neural network generates an image and another tries to determine whether the image is real or fake. Metz and Collins quote Jaakko Lehtinen, one of the researchers behind the project, as saying, “The computer learns to generate these images by playing a cat-and-mouse game against itself.”
Metz and Collins say a second team at Nvidia has developed a system that can convert a summer street scene to a winter one, and researchers at the University of California, Berkeley, have designed another that converts horses into zebras and Monets into Van Goghs.
Such concerns raise ethics issues. Metz and Collins quote Tim Hwang, director of the Ethics and Governance of Artificial Intelligence Fund, as saying, “The concern is that these techniques will rise to the point where it becomes very difficult to discern truth from falsity. You might believe that accelerates problems we already have.”
