A DESPAIRING prediction for the digital future came from an unlikely source recently. Speaking of “deepfakes,” or media manipulated through artificial intelligence, the actress Scarlett Johansson told The Post that “the Internet is a vast wormhole of darkness that eats itself.”
A stark view, no doubt, but when it comes to deepfakes, it may not be entirely unmerited. The ability to use machine learning to simulate an individual saying or doing almost anything poses personal and political risks that societies around the world are ill-equipped to guard against.
Ms. Johansson’s comments appeared in a report in The Post about how individuals’ faces, and celebrities’ faces in particular, are grafted onto pornographic videos and passed around the Web — sometimes to blackmail, sometimes just to humiliate. But deepfakes could also have applications in information warfare. A foreign adversary hoping to influence an election could plant a doctored clip of a politician committing a gaffe. Convincingly edited video could confuse military officers in the field. The ensuing uncertainty could also be exploited to undermine journalistic credibility; tomorrow’s deepfake may be today’s “fake news.”
Perhaps the scariest part of these Frankenstein-ish creations is how easy they are to make, especially when the software for a specific application — such as pornography — is publicly available. A layman can simply plug sufficient photos or footage into prewritten code and produce a lifelike lie about his or her subject. Deepfakery is democratizing, and malicious actors, however unsophisticated, are increasingly able to harness it.
Deepfakes are also inherently hard to detect. The technology used to create them is trained in part with the same algorithms that distinguish fake content from real — so any strides in ferreting out false content will soon be weaponized to make that content more convincing. This means online platforms have their police work cut out for them, though investment in staying one step ahead, along with algorithmic tweaks to demote untrustworthy sources and de-emphasize virality, will always be needed. Some suggest holding sites liable for the damages caused by deepfakes if companies do too little to remove dangerous content.
Like technical solutions, policy answers to the deepfake problem are elusive, but steps can be taken. Many harmful deepfakes are already illegal under copyright, defamation and other laws, but Congress should tweak existing fraud-related regulations to cover the technology explicitly — amping up penalties and bringing federal resources, as well as public attention, to bear on a devilish problem. Humans have so far hardly had to think about what happens when someone else uses our faces. To avoid that wormhole of darkness, we will have start thinking hard.