it takes a few steps to make a face-swap video. first, you run thousands of face shots of the two people through an ai algorithm called an encoder. the encoder finds and learns similarities between the two faces, and reduces them to their shared common features, compressing the images in the process. a second ai algorithm called a decoder is then taught to recover the faces from the compressed images. because the faces are different, you train one decoder to recover the first person's face, and another decoder to recover the second person's face. to perform the face swap, you simply feed encoded images into the "wrong" decoder. for example, a compressed image of person a's face is fed into the decoder trained on person b. the decoder then reconstructs the face of person b with the expressions and orientation of face a. for a convincing video, this has to be done on every frame.
it gets harder as the technology improves. in 2018, us researchers discovered that deepfake faces don't blink normally. no surprise there: the majority of images show people with their eyes open, so the algorithms never really learn about blinking. at first, it seemed like a silver bullet for the detection problem. but no sooner had the research been published, than deepfakes appeared with blinking. such is the nature of the game: as soon as a weakness is revealed, it is fixed.
facebook last week banned deepfake videos that are likely to mislead viewers into thinking someone "said words that they did not actually say", in the run-up to the 2020 us election. however, the policy covers only misinformation produced using ai, meaning "shallowfakes" (see below) are still allowed on the platform.
there is still ample room for mischief-making, though. last year, tesla stock crashed when elon musk smoked a joint on a live web show. in december, donald trump flew home early from a nato meeting when genuine footage emerged of other world leaders apparently mocking him. will plausible deepfakes shift stock prices, influence voters and provoke religious tension? it seems a safe bet.
the more insidious impact of deepfakes, along with other synthetic media and fake news, is to create a zero-trust society, where people cannot, or no longer bother to, distinguish truth from falsehood. and when trust is eroded, it is easier to raise doubts about specific events.
donald trump, who admitted to boasting about grabbing women's genitals in a recorded conversation, later suggested the tape was not real. in prince andrew's bbc interview with emily maitlis, the prince cast doubt on the authenticity of a photo taken with virginia giuffre, a shot her attorney insists is genuine and unaltered.
in another incident, jim acosta, a cnn correspondent, was temporarily banned from white house press briefings during a heated exchange with the president. a shallowfake video released afterwards appeared to show him making contact with an intern who tried to take the microphone off him. it later emerged that the video had been sped up at the crucial moment, making the move look aggressive. acosta's press pass was later reinstated.
the uk's conservative party used similar shallowfake tactics. in the run-up to the recent election, the conservatives doctored a tv interview with the labour mp keir starmer to make it seem that he was unable to answer a question about the party's brexit stance. with deepfakes, the mischief-making is only likely to increase. as henry ajder, head of threat intelligence at deeptrace, puts it: "the world is becoming increasingly more synthetic. this technology is not going away."
designer bags cyber mondaygucci replica bags
www.popninja.net |
|
|