Using AI to detect seemingly perfect deep-fake videos

One year ago, Maneesh Agrawala of Stanford helped develop a lip-sync technology that allowed video editors to almost undetectably modify speakers' words. The tool could seamlessly insert words that a person never said, even mid-sentence, or eliminate words she had said. To the naked eye, and even to many computer-based systems, nothing would look amiss. The tool made it much easier to fix glitches without re-shooting entire scenes, as well as to tailor TV shows or movies for different audiences in different places. But the technology also created worrisome new opportunities for hard-to-spot deep-fake videos that are created for the express purpose of distorting the truth.