In just a few months, you could hold the technology to literally put words in anyone’s mouth.
Deepfake technology uses artificial intelligence to combine and superimpose existing images and videos onto source media. Essentially, you can copy one person’s face and paste it on another body.
Hao Li, an associate professor of Computer Science at USC, is one of the major experts in this field. He has created some of the most convincing deepfakes currently available online.
As the director of the Vision and Graphics Lab at the USC Institute for Creative Technologies and CEO/Co-Founder of Pinscreen, Li’s work has placed an emphasis on digitizing humans and capturing their performances for immersive communication and telepresence in virtual worlds.
Li told attendees at an MIT conference in September that he believes average people will be able to create deepfakes within two to three years.
Li’s prediction changed two days later.
He now believes that people will be able to create “perfectly real” deepfakes in 6 to 12 months. This recalibration is due to a heightened focus on deepfake technology and the launch of the Zao app, which lets users insert their faces into famous films and television shows.
Deepfake AI technology is commonly used in film production for face swapping applications. Li was part of the research team that worked on "Furious 7″ to recreate scenes with the late Paul Walker by replacing the face of a body double with a digital version of Walker’s face.
Similar technology is also used to create convincing digital avatars for gaming, immersive communications and virtual fashion.
These projects cost millions of dollars and require months of work from engineers and artists.
“Traditionally, you need a sophisticated visual effects pipeline or game studio to create animated faces, and from a research standpoint we were developing technologies to both scale the production of animation and offer consumers similar capabilities, such as Snapchat lenses or animojis,” Li said.
However, deepfake technology is improving to become cheaper and more accessible. The tools to manipulate videos are becoming available to anyone with a computer.
“This raises the concern when harmful deepfakes are created where the intention is not for entertainment purpose but for disinformation, especially when media can spread very rapidly on social media and when the content can be created by anyone,” Li said.
For now, Deepfake is mostly harmless. Other experts, however, are concerned about its potential “nefarious uses," such as scamming companies, stealing identities and interfering with elections.
That’s why Li is working with Hany Farid, one of the top media forensics experts at UC Berkeley, to develop technology that can detect realistic deepfakes. Not much has been done yet, but with the 2020 US election coming soon, they face a time crunch.
After fake news permeated social media during the 2016 presidential election, Li expressed concern about deepfake videos and they way to regulate the spread of disinformation - a concern that still lingers with him moving forward.
“I think that it will have a similar effect as photoshop, where people will no longer blindly believe in videos that they see,” he said. “We will probably have better ways of fact checking, and also different ways of obtaining trusted sources of information, rather than anything that we read on social media.”