Artificial intelligence is now so powerful that it can trick you into believing an image of Pope Francis in a white puffy Balenciaga coat is real, but digital tools that reliably identify counterfeit images are the only way to make the content Struggling to keep up with the pace of generation.
Just ask the researchers at Deakin University’s School of Information Technology outside Melbourne. According to Stanford University’s Artificial Intelligence Index 2023, their algorithm performed best last year at identifying images of tampered celebrities from a series of so-called deepfakes.
“Pretty good performance,” said Professor Chang-Tsun Li of Deakin’s Center for Cyber Resilience and Trust, who developed the algorithm, and it was proven correct 78% of the time. “But the technology is really still under development.” Li said the method needs to be further enhanced before it is ready for commercial use.
Deepfakes have been around for years and are causing concern. Former House Speaker Nancy Pelosi appeared to denigrate her words in a defaced video that went viral on social media in 2019. Meta Platforms Inc. About a month later, Facebook After refusing to remove Mr. Pelosi’s video, a modified video was released to show CEO Mark Zuckerberg saying something he didn’t say.
The pufferfish pope’s image was a relatively harmless manipulation, but as technology advances, deepfakes can do more harm than good, from election manipulation to sex acts. Last year, a fake video of Ukrainian President Volodymyr Zelensky asking his soldiers to surrender to Russia could have serious repercussions.
Tech giants and a series of start-ups are pouring tens of billions of dollars into generative AI to claim a leading role in the technology that could change the face of everything from search engines to video games. pouring. However, the global market for technology to eradicate manipulated content is relatively small. According to research firm HSRC, the global market for deepfake detection was valued at $3.86 billion in 2020 and is expected to grow at a compound annual growth rate of 42% through 2026.
Claire Leibowicz, director of the AI and media integrity program at the nonprofit The Partnership on AI, agrees that experts are paying too much attention to AI generation and not enough to detect it. increase.
While the buzz around the technology, dominated by applications like OpenAI’s ChatGPT, has reached a feverish pitch, executives from Tesla CEO Elon Musk to Alphabet CEO Sundar Pichai are warning of too soon. warns of risks.
It will be some time before detection tools are available to combat waves of realistic-looking altered images from generative AI programs such as Midjourney and OpenAI’s DALL-E that generated Pope images. takes. Part of the problem is the prohibitive cost of developing accurate detection, with little legal or financial incentive to do so.
Forrester Research analyst Jeff Pollard said: “They are concerned about generative AI. But when it comes to things like deepfake detection, that’s not what they spend their budgets on.
Still, a handful of startups, such as Netherlands-based Sensity AI and Estonia-based Sentinel, and many big tech companies are developing deepfake detection techniques. Intel Corp. launched his FakeCatcher product last November as part of its commitment to responsible AI. According to the company, the technology can look for real cues in real videos and detect fakes with 96% accuracy by evaluating human features such as blood flow within pixels of the video.
“Currently, the motivation for deepfake detection is not money,” said Ilke Demir, senior staff research scientist at Intel.
So far, deepfake detection startups have mostly served governments and businesses looking to reduce fraud, not consumers. Reality Defender, a Y-Combinator-backed startup, charges based on the number of scans it performs. These costs range from tens of thousands to millions of dollars to cover expensive graphics processing chips and cloud computing power.
Consumers are in the dark because platforms like Facebook and Twitter aren’t required by law to detect and warn about deepfake content on their platforms, said Reality Defender CEO Ben Colman. says. “The only organizations that do anything are the banks that are directly involved in financial fraud.”
Current methods for detecting fake images and videos involve training computers to learn from examples and comparing visual features of content by embedding watermarks or camera fingerprints in the original work. But the rapid spread of deepfakes requires more powerful algorithms and computing resources, said his Xuequan Lu, another Deakin University professor who worked on algorithms.
And without a widely-used commercial tool to distinguish between fake online content and real content, malicious actors can easily break in.
said Ted Schlein, chairman and general partner of Ballistic Ventures. initial. As hacks became more sophisticated and damaging, antivirus software was developed, eventually becoming cheap enough for consumers to download onto their PCs. “We are in the very early stages of deepfakes,” but so far it is mostly done for entertainment purposes. “We’re just starting to see some malicious cases,” Schlein said.
But even if it’s cheap enough, consumers may not be willing to pay for such technology, said the head of artificial intelligence at security and fraud prevention company F5 Inc. One Shuman Ghosemajumder said:
“Consumers don’t want to do the extra work themselves,” he said. “They want to be automatically protected as much as possible.”
