Large language models (LLMs), particularly transformer-based models, have experienced rapid advancements in recent years. These models have been successfully applied to various domains, including natural language, biological and chemcial research, and code generation.
Combining laboratory automation technologies with powerful LLMs opens the door to developing a sought-after system that autonomously designs and executes scientific experiments.
In a new study, Carnegie Mellon University scientists present a multi-LLMs-based intelligent agent, Coscientist, that can autonomously design, plan, and perform complex scientific experiments.
Coscientist uses large language models (LLMs), including OpenAI’s GPT-4 and Anthropic’s Claude, to execute the full range of the experimental process with a simple, plain language prompt.
For instance, a scientist can task a Coscientist to locate a compound with specific properties. The system extensively searches the internet, documentation data, and other sources, synthesizes the information and proposes an experimental plan utilizing robotic application programming interfaces (APIs). This plan is then sent to automated instruments, which carry out the experiment. With the system’s assistance, a human can design and execute experiments more rapidly, accurately, and efficiently than working alone.
National Science Foundation (NSF) Chemistry Division Director David Berkowitz said, “Beyond the chemical synthesis tasks demonstrated by their system, Gomes and his team have successfully synthesized a hyper-efficient lab partner. They put all the pieces together, and the result is far more than the sum of its parts — it can be used for genuinely useful scientific purposes.”
In the study, scientists showcased that Coscientist is capable of planning the chemical synthesis of known compounds, navigating hardware documentation, executing high-level commands in a cloud lab, controlling liquid handling instruments, performing scientific tasks involving multiple hardware modules and diverse data sources, and solving optimization problems through the analysis of previously collected data.
Brian Frezza, co-founder and co-CEO of ECL, said, “Professor Gomes and his team’s groundbreaking work here has not only demonstrated the value of self-driving experimentation but also pioneered a novel means of sharing the fruits of that work with the broader scientific community using cloud lab technology.”
Coscientist serves to open the “black box” of experimentation. By meticulously following and documenting each research step, the system ensures the work is fully traceable and reproducible.
Kathy Covert, director of the Centers for Chemical Innovation program at the U.S. National Science Foundation, which supported this work, said, “This work shows how two emerging tools in chemistry — AI and automation — can be integrated into an even more powerful tool. Systems like Coscientist will enable new approaches to improve how we synthesize new chemicals rapidly, and the datasets generated with those systems will be reliable, replicable, reproducible, and re-usable by other chemists, magnifying their impact.”
Assistant Professor of Chemistry and Chemical Engineering Gabe Gomes said, “I believe the positive things AI-enabled science can do far outweigh the negatives. But we must acknowledge what could go wrong and provide solutions and fail-safes.”
“By ensuring ethical and responsible use of these powerful tools, we can continue to explore the vast potential of large language models in advancing scientific research while mitigating the risks associated with their misuse,” the authors wrote in the paper.”
Journal Reference:
- Boiko, D.A., MacKnight, R., Kline, B. et al. Autonomous chemical research with large language models. Nature 624, 570–578 (2023). DOI: 10.1038/s41586-023-06792-0