HOT
- Xsens’ new MTi 600-series wins AUVSI XCELLENCE …
- The Future Group, Xsens and IKINEMA partner up
- Live data analysis of the long jump at the FBK G…
- Xsens and Ergoneers join forces to launch ergono…
- New MTi 600-series industrial-grade inertial sen…
- Xsens North America Celebrates New Office Grand …
- Xsens’ latest update takes outdoor motion captu…
Deepfake Pioneer: “Perfectly Real” Fake Vids Are Six Months Away
Deepfake technology is advancing so rapidly, even Hao Li — one of the leading experts in the field — is having trouble keeping up.
Li has watertight credentials. When he isn’t teaching classes as a professor of computer science at the University of Southern California, he’s the creator of some of the most convincing deepfakes out there.
And he’s in a predictive mood. On Wednesday, he told attendees at an MIT conference that he believes easy-to-use tech to create perfect deepfakes — ones that are virtually impossible to detect with the naked eye — is just two to three years away.
Scary, right?
Well, two days later, Li appeared on CNBC’s “Power Lunch” program and told the show’s hosts he expects everyday people to be able to create “perfectly real” deepfakes in just six to 12 months.
When CNBC later reached out to Li to ask him what changed in the two days since the MIT conference, he replied that he was forced to “recalibrate” his expected timeline due to an increased focus on deepfake tech and the growing popularity of Zao, a deepfake app that lets users insert their faces into famous films and television shows.
While Zao’s application of deepfake tech is relatively harmless, Li and other experts are highly concerned about its potential nefarious uses. If widely available, there’s no telling how many bad actors might use deepfake tech to scam companies, steal identities, or even interfere with elections, for example.
To prevent that from happening, Li is currently working with University of California at Berkeley computer science professor Hany Farid to develop technology to detect deepfakes that look entirely real.
“We are working together on an approach that assumes that deepfakes will be perfect,” Li told attendees at the MIT conference, adding that “there will be no way to tell if it’s real or not, so we have to take a different approach.”
Just what that approach will be isn’t yet clear — but with the 2020 U.S. presidential election now expected to occur after perfect deepfake technology becomes available to everyday internet users, there’s no time to waste in figuring it out.