Prelims: Current events of national and international importance | Science & Technology
Why in news?
Researchers at the University of California, Los Angeles, have found a way to tell if a video is a deepfake based on the light in it.
- New Technique – A hybrid optical-electronic computer that uses the physics of light to detect deepfake videos with remarkable speed and accuracy.
A hybrid optical-electronic computer combines light-based (photonic) processing with traditional electricity-based (electronic) processing to achieve ultra-fast speeds, high bandwidth, and low energy consumption.
- Need for a detection system – It was designed to address the massive energy, speed, and security limitations of traditional digital deepfake detectors.
- Because humans only catch deepfakes about 50% of the time, this light-powered technology provides a critical first line of defence for social media platforms struggling against a flood of AI-generated misinformation.
Deepfakes – It is synthetic media (images, video, or audio) created or altered using AI and deep learning techniques to make it look or sound like a real person is saying or doing something they never did.
- Working – Unlike traditional digital processors that evaluate data frame-by-frame using silicon chips, this hybrid system offloads the heaviest computational work onto photons (light).
- Frame Selection – A lightweight digital encoder selects 12 specific frames from a video and maps unique facial features.
- Light Modulation – These features are converted into a phase pattern and projected onto a programmable spatial light modulator.
- Diffraction Analysis – The light travels through a passive optical decoder. The machine splits the light wavefront toward sensor spots and analyses the differences in brightness caused by light diffraction.
- Based on the difference in brightness between these spots, the system could say whether a video was fake.
- Instant Classification – If the brightness pattern matches known anomalies typical of generative AI, it instantly flags the video as synthetic.
- Performance & Capabilities –
- 97.79% average accuracy while processing 15 videos simultaneously.
- 99.86% sensitivity, meaning it correctly identifies almost all fake videos.
- Only about 0.14% false negatives.
- Effective against advanced AI videos – When tested on videos generated using Google's VEO-3 model, which produces highly realistic synthetic footage, the system achieved 94.8% accuracy with minimal additional training.
Reference
The Hindu | New technique uses light to spot deepfakes in seconds
