Google has introduced a new technology called SynthID, designed to identify content generated or edited using its artificial intelligence (AI) tools. Developed by Google DeepMind, SynthID embeds a digital watermark into AI-generated content, such as images, videos, audio files, and text. This watermark is invisible to humans but can be detected by SynthID systems. The technology aims to address the growing challenge of distinguishing between human-created and AI-generated content.
SynthID works by adding a digital watermark directly to the content produced by Google's AI systems. This watermark does not affect the quality of the content and can be detected even after common edits, such as cropping images, adding filters, or changing frame rates. Google uses SynthID with its Lyria music generation model and NotebookLM podcast creation feature. The company claims the audio watermark is inaudible and can be detected even after some edits, like adding noise or compressing files.
The verification process is also available within Google's Gemini chatbot. Users can upload an image, video, or audio file to the chat and ask if it was generated or edited using Google's AI tools. Gemini searches for the digital watermark and informs the user if it detects one. However, Google notes that SynthID is not a universal tool capable of detecting all types of AI-generated content, as it primarily focuses on recognizing the watermark associated with content produced by Google's systems.
Google has expanded SynthID to include text generated by Gemini and its web experience. The watermarking mechanism in text differs from other media, as it involves making subtle adjustments to the probabilities used by language models to select words or symbols. This creates a pattern that SynthID can detect later. Google states that this pattern is not apparent to readers and is designed not to affect the quality of the generated text.
In addition to integrating SynthID with Gemini, Google has launched the SynthID Detector, a separate portal that allows users to upload images, videos, or audio files to verify if they contain the SynthID watermark. Google DeepMind is collaborating with journalists and media professionals to test and evaluate the portal's performance. This cooperation aims to gather feedback before expanding the tool's usage.
While SynthID provides a means to track content associated with Google's AI systems that use the watermark, it does not claim to detect all forms of AI-generated content. The verification process relies on the presence of the SynthID watermark from the outset, so its absence does not necessarily mean the content was created by humans. Therefore, SynthID offers a way to monitor content linked to specific AI systems but is not a comprehensive tool for determining the origin of all digital content.
Google's introduction of SynthID is a significant step in addressing the challenges posed by AI-generated content. As AI technology continues to evolve, the ability to distinguish between human-created and AI-generated content becomes increasingly important. SynthID's development reflects Google's efforts to provide transparency and accountability in the use of AI tools, and it will be interesting to see how this technology is adopted and expanded in the future.
Key points
- Google's SynthID technology embeds a digital watermark into AI-generated content to identify its origin.
- SynthID is integrated with Google's Gemini chatbot and is available as a separate detector tool.
- The technology currently focuses on content produced by Google's AI systems using the SynthID watermark.
