The rise of Deepfakes has prompted intense coverage in the press, concern from government officials, and fear among the public. Deepfakes, along with GANs (generative adversarial networks), are a class of "generative" neural networks capable of creating highly realistic synthetic images. The widespread availability of open-source Deepfake tools means anyone with access to a computer can potentially create photorealistic fake videos and images. These fakes can, for instance, portray high-profile individuals in arbitrary--possibly compromising--situations. Because of their wide availability, relative ease of use, and harm potential, the technology has been the subject of considerable scrutiny and debate.
In this talk, we'll discuss the current state of AI-generated imagery, including Deepfakes and GANs: how they work, their capabilities, and what the future may hold. We'll try to separate the hype from reality, and examine the social consequences of these technologies with a special focus on the effect that the idea of Deepfakes has had on the public. We'll consider the visual misinformation landscape more broadly, including so-called "shallowfakes" and "cheapfakes" like Photoshop. Finally, we'll review the challenges and promise of the global research community that has emerged around detecting visual misinformation.
Nick Dufour is a Senior Software Engineer at Google Research, and a member of the team overseeing research into Deepfakes and visual misinformation. Nick joined Google in 2016 following his graduate work in Machine Learning at Stanford University.