In the past year and a half, the generation of AI-generated images has advanced by leaps and bounds, only raising concerns about the potential misuse of these fake images…a debate that becomes more relevant as more Silicon Valley CEOs See US Presidential Election Approaching of 2024.
In response to this, the Google DeepMind team (Alphabet’s department responsible for AI-related developments) announced today that they have been working on an innovative tool called SynthID, the purpose of which would be embed ‘watermarks’ invisible to the human eye in AI-generated images… but in such a way that said brands continue to be identifiable by digital tools created for this purpose.
“While not a silver bullet to address the problem of disinformation, SynthID is an early and promising technical solution to this pressing AI security issue.”
One of the key features of SynthID is its resistance to tampering: it already simply modifying or cropping an image will not suffice, as the SynthID digital watermark persists even after the image has been significantly altered.
Although the tool is still in an experimental phase and is only available to a few Google Cloud customers using its Vertex AI platform, it is expected to be further refined as it is more widely deployed.
In fact, Demis Hassabis, CEO of Google DeepMind, seems to expect that, over time, SynthID will become a ‘de facto’ standard on the Internet; in fact, he is convinced that the fundamentals of this technology could even be used in other media such as video and text.
Other companies, such as Microsoft and OpenAI, are also researching and developing methods to detect and mark AI-generated images.
Despite advances in this area, the race between AI developers and those seeking to circumvent detection tools is a constant challenge. Experts anticipate that fake imagers will also find more ingenious ways to evade detection. This is why Google has chosen not to divulge too many technical details about SynthID in an effort to prevent malicious actors from discovering ways around this technology.
Image | Pixabay
In Genbeta | The video of Princess Leonor dancing on TikTok was a deepfake that has deceived users in Spain