Google DeepMind has launched a new watermarking tool called SynthID for AI-generated images. This makes Google the first Big Tech firm to publicly launch such a tool after a group of them pledged to develop them at the White House. SynthID will initially be available only to users of Google’s AI image generator Imagen. The tool aims to help people identify AI-generated content and protect copyright. Watermarking is considered an effective technique to curb the misuse of AI-generated content.
| Signal | Change | 10y horizon | Driving force |
|---|---|---|---|
| Google DeepMind launches watermarking tool | Watermarking AI-generated images | More widespread use of AI watermarking tools | Combat misinformation and protect copyright |
| Big Tech firms pledge to develop watermarking | Industry commitment to combat AI-generated harm | Improved techniques and sharing of information | Combat misinformation and protect copyright |
| Watermarking used to curb harms of AI-generated content | Prevention of deepfakes and copyright infringements | More robust and tamper-resistant watermarks | Combat misinformation and protect copyright |
| SynthID tool detects watermarks in AI-generated images | Improved detection of AI-generated content | Better identification of AI-generated content | Combat misinformation and protect copyright |
| Watermarking methods exist, but not widely available | Limited availability of watermarking tools | Increased availability of watermarking tools | Combat misinformation and protect copyright |
| Watermarks may not be robust over time | Challenges in maintaining robust watermarks | Continued efforts to improve watermarking | Attempts to discredit or promote AI-generated content |
| Google DeepMind’s launch is a first step | Progress in the field of watermarking | Further development and improvement of tools | Advancement in combatting AI-generated harm |
| Watermarking tool is experimental | Limited use and testing of the tool | Iterative improvement and wider availability | Evaluation and learning from user feedback |
| Proprietary watermarking limits usefulness | Limited access to embedding and detection | Wider adoption of watermarking across the board | Mitigation of harms caused by AI-generated content |