AI Experts Warn of Worsening Deepfake Threat and the Danger of Denying Real Content
Artificial intelligence (AI) experts have issued a warning about the escalating threat of deepfakes and the potential risks associated with denying real content. During a presentation on AI and TV at the Edinburgh TV Festival, experts highlighted the positive aspects of AI and its potential for creating new jobs. However, they emphasized the rapid advancement of deepfake technology and its potential impact on society.
The event, held at the Edinburgh International Conference Centre, featured presenters Alex Connock, a senior fellow in management practice at The University of Oxford, mathematician and author Hannah Fry, and Muslim Alim, commissioning editor for the BBC. The trio showcased various deepfake examples, including a video of Martin Lewis, where he seemingly endorsed a product created by Elon Musk.
While some deepfakes were less convincing, such as a humorous debate between Donald Trump and Joe Biden, others were more realistic, like a video of Facebook founder Mark Zuckerberg speaking fluently in Hindi. Despite the varying levels of authenticity, the experts expressed concern that deepfakes will become increasingly advanced in the future.
Mr. Connock admitted that he may not have initially recognized the Martin Lewis video as a deepfake and urged individuals to be vigilant. He warned, We’ve all read a lot about deepfakes in recent years, and the good news is that they haven’t been that convincing so far. But the bad news is that they’re about to become incredibly realistic. As we approach the US 2024 election, we can expect the situation to worsen.
He also cited academic research indicating that the more fake something appears, the more credible it becomes. Differentiating between real and fake content becomes increasingly challenging when content is entirely fabricated. Mr. Connock emphasized the importance of remaining alert to this issue, especially for those in the news industry. Deepfakes are already a significant concern in conflicts like the Ukraine war, employed by both sides, and experts anticipate their proliferation.
When the presenters were asked about the future development of deepfakes and measures to identify them, Ms. Fry mentioned the possibility of integrating techniques into generated content that could enable algorithms to recognize deepfakes. She referred to mathematical tricks like barcodes, which, although not immediately visible, can still be effective.
However, Ms. Fry highlighted a more significant concern associated with deepfakes: the denial of real content. She suggested that this aspect is even more worrisome than fake content itself. While acknowledging the potential for political misinformation and deepfakes, she drew parallels to past concerns about the invention of Photoshop. Establishing the source of a video is crucial in determining its credibility, and context plays a vital role in this assessment.
In conclusion, AI experts have alerted society to the growing threat posed by deepfakes and the dangers of denying real content. As deepfake technology advances, it becomes increasingly difficult to distinguish between authentic and fabricated material. While techniques to detect deepfakes are being developed, the experts highlighted the importance of remaining vigilant and critically assessing the source and context of content. With proper awareness and measures in place, society can navigate the pitfalls of deepfakes and preserve the integrity of information.
