Light-Based AI System Achieves Near-98% Accuracy in Deepfake Detection
Researchers at UCLA have built an AI system that leverages light-based (photonic) computing to detect deepfakes with nearly 98% accuracy. The system can analyze more than a dozen videos at once, offering a significant speed advantage over traditional deepfake detection methods that typically process content sequentially.
The photonic approach offers several practical advantages beyond accuracy. By performing computations using light rather than electrical signals, the system achieves low energy demands compared to conventional GPU-based detection tools. This efficiency could make large-scale video screening more accessible for news organizations, social media platforms, and verification services.
Researchers also note the system's resistance to adversarial attacks, where malicious actors attempt to fool detection models by modifying deepfakes in subtle ways. This robustness addresses a common vulnerability in existing deepfake detection technologies, which can be circumvented by sophisticated manipulation techniques.
As AI-generated video content becomes increasingly prevalent, tools capable of processing large volumes of media quickly and accurately are growing in importance. The UCLA system's combination of speed, energy efficiency, and attack resistance positions it as a promising candidate for deployment in content verification pipelines where computational resources may be limited.