Google DeepMind expands Co-Scientist AI to physical laboratory automation

Google DeepMind has expanded its Co-Scientist system, a multi-agent AI architecture based on Gemini, to operate within physical laboratory environments. Moving beyond virtual simulations, the system can now autonomously formulate hypotheses, propose experiments, control laboratory equipment, and analyze resulting data to refine its research. This closed-loop process allows the AI to interact with physical constraints and adapt protocols to real-world conditions.

In material science applications, the system successfully used Gemini 3 Deep Think to design synthesis procedures for 2D materials, such as molybdenum disulfide (MoS2), molybdenum diselenide (MoSe2), and tungsten disulfide (WS2), achieving results on the first attempt. This capability is significant due to the precise control required for synthesizing such materials.

Separately, Google is updating Gemini Live on Android with a real-time translation feature. This tool allows the voice assistant to recognize a spoken language and provide live translations, appearing as progressive subtitles on the screen. This update aims to facilitate communication between users speaking different languages by acting as a real-time interpreter.