Deepfakes: Hany Farid Warns About Reality Distorted by AI

Hany Farid: A Pioneer Facing the Evolution of Deepfakes

Hany Farid, an iconic figure in the field of digital forensics, has expressed his concerns about the growing impact of artificial intelligence on our ability to distinguish between the true and the false. Since he began exploring this field in 1999, when digital forensics was still in its infancy, Farid has witnessed a radical transformation. Today, it is nearly impossible for the average person to discern whether content is authentic or AI-generated simply by looking at it. This evolution raises profound concerns, as it alters our perception of reality itself.

The Illusion of Reality in the Digital Age

For the majority of Internet users, it has become extremely difficult to determine the authenticity of an image, video, or audio recording that appears on their screens. Perceptual studies have shown that our visual and auditory senses are not sharp enough to reliably accomplish this task. However, this does not mean that the distinction is impossible. With advanced computational and mathematical tools, it is possible to unmask falsified content. But this capability remains largely inaccessible to the average social media user, who lacks both the tools and the time necessary to conduct such analyses.

The Beginnings of Digital Forensics

In 1999, Hany Farid began his academic career at Dartmouth College, at a time when the world was still predominantly analog. Digital cameras were just beginning to emerge, and the Internet was far from the omnipresence it is today. At that time, the idea of malleable digital evidence in courts seemed distant and untroubling. Yet, Farid and a few visionary students began laying the groundwork for what would become digital forensics, a field that seemed esoteric then but would quickly gain relevance with the rise of digital technology.

The Evolution of Image Manipulation

Early in his career, Farid focused primarily on photographs, as video manipulation was still complex due to the need to process 24 to 30 frames per second, accompanied by an audio track. Tools like Photoshop facilitated the manipulation of still images, but this task still required technical skills. Manipulation errors, such as incorrect shadows or geometric inconsistencies, could be detected. Today, however, the situation has radically changed. Thanks to AI, it only takes a keyboard and an Internet connection to generate impressive visual and audio content, without requiring any special skills.

The Visual Indistinction of AI-Generated Content

With technological advancements, it is crucial not to focus solely on the current state of technology, but on its future direction. Images were the first to become visually indistinguishable from reality, followed by voice, with its nuances of intonation and rhythm. Video is now crossing this threshold, making it difficult to distinguish between the real and the generated for the short videos commonly found online. While AI-generated videos were once limited to a few seconds, they can now reach 30 to 40 seconds, making the content increasingly realistic and accessible.

The Race Against Time to Identify Fakes

Although generative AI can produce convincing content, it does not understand the three-dimensional world or the laws of physics. These limitations provide clues for detecting falsified content. However, authenticating content is a long and complex process, often more difficult than detecting a fake. On average, an analysis can take about an hour, a timeframe that seems endless in the scale of the Internet, where content can quickly reach millions of views. Verifications often arrive too late, after the content has already been widely disseminated.

The Threat to Our Shared Perception of Reality

The stakes related to content authenticity are increasingly high, affecting crucial decisions, whether judicial or geopolitical. What worries Hany Farid the most is the gradual loss of a shared sense of reality within society. Debates are no longer just about political or economic issues, but about fundamental facts. This situation threatens democratic stability, as a society cannot function without a common foundation of truth. Disagreement is healthy, but it must be based on a shared reality to be constructive.

GetReal: A Response to the Era of Deepfakes

To address these challenges, Hany Farid co-founded GetReal, a startup specializing in digital forensics. This company aims to provide tools and solutions for identifying and authenticating digital content, thereby helping to restore some trust in the visual and auditory information circulating online. GetReal positions itself as a key player in the fight against digital misinformation, offering services that allow for faster and more reliable verification of content authenticity.