Google DeepMind has recently developed an open-source AI watermarking tool called SynthID with the aim of fighting proliferating AI-generated text.
SynthID was released on October 23, 2024. It currently offers text watermarking to businesses as well as other developers to help users better determine whether content was generated using AI. The new tool is available on Google's Responsible Generative AI Toolkit as well as through its listing on Hugging Face.
A recent AWS AI lab study showed that over 57% of translated sentences are AI-generated. SynthID puts the issue of misinformation and misleading content to rest. It allows machine learning to predict the placement of words within sentences and replacing some words with watermarked synonyms. Thus, SynthID ensures reliable verification of the content authenticity.
Although its current core focus is on texts, Google DeepMind plans to launch AI for images, videos, and even audio in the near future. This new and innovative approach may ensure that online content remains unique, free of copied AI-generated content.