By 6:30 a.m. most mornings, Hany Farid is already hard at work. It’s early in Vermont, but Europe and the Middle East have been awake for hours. Conflicts in Ukraine, Gaza and Iran are generating a steady stream of footage — some authentic, some fabricated — and much of it is increasingly dif ficult to verify. From his sparely furnished home office in the Upper Valley, the Dartmouth College computer science professor and digital forensics expert fields requests from journalists, technology companies and others who want him to determine what’s real. Some need answers within hours.  By midmorning, he often makes the short commute to the Dartmouth campus with his wife, neuroscientist and fellow professor Emily Cooper, if the work hasn’t taken him farther afield. Wherever he goes, the inquiries continue to pile up, through meetings and into the evening. Fourteen-hour workdays have become the norm. For more than two decades, Farid has developed tools to identify manipulated images and address some of the internet’s most troubling problems. After spending seven years at the University of California, Berkeley, he recently returned to Dartmouth, where he helped pioneer digital forensics in the early 2000s.  Now artificial intelligence has transformed the work. Digital fabrications proliferate at unprecedented speed and scale, and with increasing sophistication that makes them harder to detect. His expertise has never been more in demand, even as the limits of what he can discern have become increasingly clear. When Farid first arrived at Dartmouth in 1999, digital photography was becoming more accessible, and software was making it easier to alter images without leaving obvious telltale signs. While experimenting with Adobe Photoshop to put a friend’s face on a tennis pro’s body, he realized that resizing the image left traces. Few researchers were studying how to detect those changes. Hany Farid Credit: Rob Strong Farid began looking for clues invisible to the naked eye: subtle patterns in pixels, inconsistencies in lighting and other marks that could distinguish real from fake. His work helped lay the foundation for a field of research that would become increasingly important as images spread across the internet and shaped public discourse.  Devin Balkcom, now chair of Dartmouth’s computer science department, remembers Farid as a generous mentor and a researcher drawn to important questions before others thought to ask them. For Farid, that meant more than publishing research. He wanted to turn his findings into tools that could address real-world harm. He made one of his most consequential contributions in 2009, when he worked with Microsoft to develop PhotoDNA, a tool that creates digital fingerprints of images so that copies of known child sexual abuse material can be identified even after they’ve been altered. The technology generates digital fingerprints that allow participating organizations to identify copies of previously identified child sexual abuse material, including some altered versions, and take steps to remove them.  Project Arachnid, an initiative of the Canadian Centre for Child Protection, uses PhotoDNA to proactively detect such material online and seek its removal. Farid has volunteered his expertise for years.  A few years ago, PhotoDNA helped Project Arachnid identify a young woman who was being sexually abused. By matching images of her to copies circulating online, the technology helped investigators piece together clues that contributed to locating and rescuing her. After learning about PhotoDNA’s role, the woman sent Farid a video thanking him. At the end, she lifted her sleeve to reveal a tattoo of Project Arachnid’s spider logo.  “It was really tear-jerking, emotional, gut-wrenching,” Farid recalled. He got the same tattoo on his forearm, adding wings representing the goddess Isis, who is associated with protection, in a nod to his Egyptian heritage. Farid sent a video of his tattoo back to the young woman.  “You won’t find someone with a bigger heart than him,” said Lloyd Richardson, the center’s director of technology. Farid also investigates questionable footage from wars, elections and breaking news, often collaborating with his wife, Cooper, whose expertise in how the human brain works complements his knowledge of technology. Many of the inquiries come through GetReal, an eight-person company he cofounded to detect digital deception. Its revenue comes largely from products that protect large businesses against AI-powered fraud, including impersonation during video calls. GetReal’s investigations for journalists and researchers typically operate at a loss, but Farid said they were a driving reason he started the company. Demand for that work has surged, from roughly one inquiry a month to several a day. The volume oscillates with world events, but the trend is unmistakably upward. When questionable footage is already circulating online, every minute or hour spent investigating it gives a potential falsehood more time to spread.  Earlier this week, Farid examined footage purportedly showing a building being blown up in Ukraine. It appeared authentic, but he couldn’t establish when it had been recorded. Farid said Russian disinformation campaigns increasingly use recycled footage, false attribution and subtle alterations to genuine images rather than creating outright fabrications. Even when footage passes every forensic test, Farid is wary it may have been fabricated using methods those tests cannot detect. Finding no evidence of manipulation is not the same as proving something is real. “I would much rather say, ‘I don’t know,’ than be wrong,” Farid said. As AI content becomes increasingly sophisticated, he finds himself saying it more often than ever. His assessments can influence national security decisions, news coverage and court cases. Last week, he met with federal judges and lawyers in Boston to discuss evidence rules that must guard against convincing fakes without excluding legitimate evidence. Farid’s assessments can influence national security decisions, news coverage and court cases. Farid doubts Vermont’s law regulating AI-generated political content (see story, page 15) will do much to curb deception, particularly when misleading material can spread rapidly across state lines. But he doesn’t fault lawmakers for trying. Other states have struggled with the problem, he said, and stronger restrictions risk running afoul of the First Amendment. While Farid supports disclosing when political videos are fabricated, he cautions against laws that could restrict people’s right to criticize elected officials, even through manipulated imagery. He said the law is unlikely to do any harm, “but let’s not pretend that it’s actually going to help us that much.” Farid has seen how misinformation erodes trust in journalists, scientists and other institutions that might correct the record. That distrust leaves people more susceptible to falling for additional falsehoods, fueling what he calls a “death spiral” — even in the United States, where reliable information is readily available. Sarah Barrington, a PhD candidate at Berkeley whom Farid continues to advise, has watched the technological changes fueling those concerns unfold firsthand. Since she began studying under Farid in 2021, AI has transformed digital forensics. She now uses the technology almost daily in her research. While AI has made many tasks easier, detecting manipulated images and videos has become more difficult. “The fakes are harder to spot, and the algorithms we build have to become more and more sophisticated,” she said. Barrington has seen the strain take a toll on Farid over the past five years, but she said it hasn’t diminished his “silliness” or generosity as a mentor. Earlier this month, he invited her to the University of Oxford for discussions organized by human rights lawyer Amal Clooney. He also found time to browse Paddington Bear souvenirs, inspired by his love of the movie Paddington 2. When difficult cases or research setbacks frustrate them, Barrington said, they joke around to keep each other going, occasionally marking a breakthrough with a celebratory cigar.  The fakes are harder to spot, and the algorithms we build have to become more and more sophisticated.sarah barrington Farid and Cooper had long discussed returning to Vermont. A sabbatical in Grand Isle settled the question: California was beautiful but lacked the peace they missed. Now, on 100 acres in the Upper Valley, they have both. Farid can step away from his computer between meetings to chop wood or walk through the forest. He rides his bicycle and Harley-Davidson motorcycle along the area’s quiet roads in warmer weather and just recently started planning a network of mountain bike trails through their property. At Dartmouth, where he is spending this academic year establishing his lab, he is getting to know a new generation of students.  The quieter surroundings and distance from Silicon Valley haven’t softened his criticism of the technology industry, particularly social media. Farid argues that companies and policy makers need to address AI’s risks before harmful content spreads rather than relying on tools or forensic experts to assess the damage afterward.  Despite those criticisms, Farid works with social media companies to respond to incidents involving questionable content. He draws a distinction between their executives and the employees who turn to him for help verifying material. The latter, he said, are often smart, dedicated people trying to do the right thing. He believes it’s their bosses who have failed to invest adequately in protecting users, in part because removing harmful content can work against the companies’ financial interests.  “Computer scientists going out there have to stop asking themselves, What can I do?, and they have to start asking themselves, What should I do?” Farid said.  But that change has yet to come. Farid still spends many of his days confronting the consequences of technology that has outpaced efforts to control it. Recently, he and Cooper spent part of a Sunday at home analyzing footage purportedly showing military operations abroad. They checked the shadows, aircraft markings and geography. Nothing they found suggested the footage was fake. But neither could say with certainty that it was real.  When they finished, it was nearly dinnertime. By morning, there would be another inbox full of questions, and Farid would be back at his desk, doing what he can to bring a fractured world closer to a shared sense of reality. ➆ Seven Days reporter Hannah Bassett will moderate a QA with Hany Farid following his presentation, “Totally Unreal: How Deepfakes Are Changing Everything,” at the Vermont Tech Jam on Saturday, October 24, 1 p.m., at Hula in Burlington. Preregister for the free event. The original print version of this article was headlined “Seeing Is Deceiving | Dartmouth digital forensics pioneer Hany Farid is racing to expose ever more realistic AI-generated fakes” The post Dartmouth’s Hany Farid Confronts the Rise of AI-Generated Fakes appeared first on Seven Days. ...read more read less