Google Develops Translucent Watermarks for AI Videos

Of the various AI features and products that Google showcased at Google I/O 2024, there was one security feature for AI that was also showcased.

The feature in question is SynthID, which can detect and mark videos created by AI. It was Demis Hassabis, CEO of Google DeepMind who introduced this new feature in his keynote.

SynthID was actually first introduced in August 2023, and its initial function was to tag by providing a watermark for images created by AI. This watermark can be said to be transparent because it is not visible to humans, but can be detected by the system.

But now its capabilities have been expanded to mark AI-generated content in the form of video and text.

The watermark method used is different from similar protocols such as C2PA, which provides watermarks in the form of encrypted metadata in AI-generated content, as quoted by from The Verge, Friday (17/5/2024).

In addition to inserting watermarks into videos, SynthID can also insert special watermarks into music created by DeepMind's Lyria model. It is called special because this watermark is made in the form of sound but cannot be heard by humans.

SynthID is just one of a number of protections being developed to protect AI from abuse. The importance of watermarks for content from AI is increasingly important because misuse is increasingly occurring.

So far, AI content has been used for several types of crimes, such as spreading false information related to politics, slandering someone by distributing fake voice recordings, and even being used to make bad videos.

There was a lot of new technology that Google showed off at I/O 2024, but everything that was showcased was related to AI. In fact, Android 15 didn't even get a place at the I/O 2024 keynote on the first day. Even though Android updates usually get a main place in the I/O keynote from year to year.

Share :