Google's SynthID watermark tested: robust but not a cure for AI misinformation

Google's SynthID watermarking system, which aims to identify AI-generated material, has undergone testing and proved resilient against modifications. However, specialists warn that it does not tackle the larger issue of misinformation generated by AI. According to the Starling Lab, a joint initiative of Stanford University and the University of Southern California, it took 149 years after the invention of the camera for humanity to produce 1.5 billion images by 1975, while generative AI accomplished this feat in a mere 18 months. During Google I/O this spring, the company revealed that its tools had created over 100 billion AI images and videos in just a few years. Google is enhancing SynthID through collaborations with OpenAI, Runway, Nvidia, and others. The technology embeds watermarks within pixels or audio, maintaining integrity through compression and alterations, yet metadata standards like C2PA can still be easily removed. Pushmeet Kohli, a scientist at Google DeepMind, highlighted SynthID's resilience, but the article points out that invisible watermarks cannot completely resolve issues of authenticity in AI content.