Artificial intelligence watermarking needs to be effective

AI company DeepMind, in its experiment, analysed a dataset of AI text and found that it was much easier to remove a watermark in the AI-generated text as compared to written text. It used a watermarking process that alters the way in which it statistically selects its ‘tokens’, which are the words from a given book or article. In its experiment, DeepMind added over 100 billion words.