Google introduces the SynthID detector tool for labeling AI-generated content, but it’s far from perfect

Google LLC is making its SynthID identification tools for artificial intelligence-generated and AI-edited videos and images more widely available with today’s launch of a new service that anyone can use.

The web-based SynthID Detector can identify images, video and audio clips using AI models from Google, OpenAI Group PBC, Nvidia Corp. and Cocoa Corp. were created. However, to use it, users must first sign in with a Google, OpenAI or Apple account. In a few weeks, the tool’s capabilities will expand to include images created using Apple Inc.’s AI tools, such as Image Playground.

Google first developed SynthID in 2023. This is an imperceptible watermark added to AI-generated content created using tools from the companies mentioned above. The only problem is that it has always required special tools to detect it.

Although the company has made various versions of its detection tool available to journalists and other media professionals, availability has been severely limited. Google also offers some other ways to detect AI-generated content. For example, there is the “Circle to Search” smartphone feature, and users can also upload content to Gemini to ask whether it was created with AI or not. But with the introduction of a dedicated detector tool, these features become more accessible.

The disadvantage of SynthID Detector is that it has drawbacks. For example, since it is designed to identify the hidden SynthID watermarks in AI content, it is not very good at detecting AI-generated content that does not have a SynthID watermark. And there are a lot of them.

Anthropic PBC’s Claude doesn’t use SynthID and xAI Corp’s Grok chatbot. neither. None of the many open Chinese AI models use SynthID either. So there will definitely be a lot of AI-generated content missing.

Another problem with SynthID Detector is that it cannot distinguish between images, videos, and audio that were created entirely from scratch or only edited using AI. With edited content, it is not possible to tell which part is fake and which part is real.

SynthID Detector is still a step in the right direction, but it’s far from enough. The reality is that there is still no foolproof method for verifying AI-generated content, in part because there are so many different standards for creating watermark models.

Meta Platforms Inc. has experimented with its own watermarking tools and developed a system for identifying images created with Muse Image, but it is useless when confronted with images created with third-party models. Unless all major AI model manufacturers, including China’s, decide to get on the same page and create some kind of universal standard for watermarking AI content, there will likely never be a reliable way to detect such material.

Given its influence and the fact that it already has several tech heavyweights on board, it’s likely that Google is in the best position to realize some kind of universal standard. The company at least seems to recognize the problem, saying on the SynthID Detector website that it is “continuously working to expand our partnerships” to be able to detect more types of AI-generated media.

“Building industry-wide standards takes time, and different content types (image, audio, video) have different technical requirements,” Google said. “The goal is a multi-layered ecosystem where watermarking, metadata standards and verification tools work together across the industry.”

Image: Google