Will self-driving 'robot labs' replace biologists? Paper sparks debate

"Last year, synthetic biologist Meagan Olsen performed the biggest experimental campaign of her career. The PhD student at Northwestern University in Evanston, Illinois, was trying to make proteins in a test tube more efficiently. Across more than 40 experiments over 4 months, she tested 1,231 combinations of sugars, amino acids and other ingredients, including cellular machinery, before landing on a cocktail that was at least six-times cheaper than existing cell-free protein synthesis recipes."

"Now, an 'autonomous laboratory' system made up of a large language model (LLM) 'scientist', lab robotics that automate simple tasks such as liquid transfer and human overseers created by scientists at artificial-intelligence firm OpenAI in San Francisco, California, and Ginkgo Bioworks, a biotechnology company in Cambridge, Massachusetts, has eclipsed Olsen's record. It achieved a further 40% reduction in cost, after testing more than 30,000 experimental conditions over 6 months."

"The findings - described in a paper posted on the bioRxiv preprint server on 5 February - have sparked discussion over the extent to which chatbot-controlled robots could replace humans. "That is going to be the future of biology," says Philip Romero, a protein engineer at the University of Wisconsin-Madison. However, the technology has some way to go before it can gain wide usage."

Meagan Olsen ran 1,231 cell-free protein synthesis experiments across more than 40 trials over four months and identified a cocktail at least six-times cheaper than previous recipes. An autonomous laboratory combining a large language model 'scientist', liquid-handling robotics and human overseers tested over 30,000 conditions across six months and reduced costs by a further 40%. The system automated simple tasks such as liquid transfer while relying on human oversight. Current lab robotics struggle with dexterous or bespoke experiments involving tissue samples or animals, and some complex research goals remain beyond existing AI tools, so human expertise remains essential.

Read at Nature