Google DeepMind scientists propose a watermark for AI-designed proteins
Science’s post about Google DeepMind scientists describes a proposed way to mark AI-designed protein sequences by subtly shifting amino-acid probabilities, leaving a detectable pattern while preserving structure and function.
TLDR
Science’s post says Google DeepMind scientists have proposed a way to watermark AI-designed proteins by slightly changing the odds of which amino acids appear in a sequence. The proposed signature would be statistically detectable while preserving the protein’s structure and function, a partial response to concerns that AI is making novel functional proteins easier to create and potentially misuse.
Combined views
5.1K
1 Source, first seen ago
