Google has made its watermarking technology for AI-generated text, called SynthID Text, available through its updated Responsible Generative AI Toolkit and Hugging Face, a repository of open-source AI tools. This tool, which is now generally available, allows developers to determine whether text has come from their own large language models (LLMs) with the goal of promoting responsible AI development.
SynthID Text works by observing a series of words in the text. LLMs use tokens to process information and generate output, and these tokens can be a single character, word, or phrase. The tool assigns each token a score based on its probability of appearing in the output for a given prompt. It also embeds imperceptible watermarks directly into the text during token distribution. When a text output is verified, SynthID compares the expected pattern of scores for watermarked and unmarked text to determine whether the text was generated by an AI tool or came from another source.
However, there are limitations to this technology. It requires at least three sentences for detection, and its accuracy and robustness increase with longer text. It is also less effective on factual text and AI-generated text that has been thoroughly rewritten or translated.
Google acknowledges that SynthID Text is not designed to directly stop motivated adversaries from causing harm. However, it can make it harder to use AI-generated content for malicious purposes and can be combined with other approaches to provide better coverage across content types and platforms.
SynthID Text is part of a larger family of tools created by Google to detect AI-generated outputs. Last year, the company released a similar tool to watermark AI images.
This AI-text detection tool comes at a time when AI-powered misinformation is on the rise, as well as false-positive detections. About two-thirds of teachers reportedly use AI detection tools for student assignments and essays, and students using English as their second language have been victims of false detection.
