Samsung researchers have developed a new way to streamline the process of creating deepfakes, allowing what they call “Realistic Neural Talking Head models” to be created using only a single image. Say, the one on your Facebook profile page. Think about that.
According to the project’s researchers (their paper can be read here), this new method of deepfaking involves “lengthy meta-learning,” during which the AI analyzes videos to understand human facial movements. Afterwards, the AI can take a single image and create a more-or-less accurate moving representation of the individual in the image.
Things get really weird when you see 2D paintings brought to life, like the Mona Lisa.
Many have questioned just what this technology will be used for.
Good? Bad? Will it be used to deceive? Even the description on the YouTube video, posted by research lead Egor Zakharov, goes into a lengthy comment on the nature of these kinds of technologies. He compares them to Hollywood’s special effects, and how over time they’ve become available to everyone:
“Our work (and quite a few parallel works) will lead to the democratization of the certain special effects technologies. And the democratization of the technologies has always had negative effects…[but] the net effect of democratization on the World has been positive, and mechanisms for stemming the negative effects have been developed.”
There are, however, other applications that aren’t quite so malicious. Imagine you’re at a museum in the future, and a replica of the Mona Lisa pops to life and starts describing her own history. That’d be neat, I guess.
Or, if we want to dive into the weird AI futures of something like Revelation Space, I wonder if such deepfakes could some day be coupled with artificial intelligence (constructed from brain scans, of course) to “resurrect” the deceased. Have a picture of Einstein sitting on your desk, ask him questions about physics when you get bored.