Researchers at Carnegie Mellon University report in the December 21 issue of Nature that a non-organic intelligent system designed, planned and executed a chemical experiment for the first time. Carnegie Mellon University's artificial intelligence system Coscientist successfully automatically completed a complex Nobel-winning chemical reaction, marking a breakthrough in artificial intelligence-driven scientific research and experiments.

"We anticipate that intelligent agent systems for autonomous scientific experiments will lead to tremendous discoveries, unforeseen treatments, and new materials." The Carnegie Mellon University research team wrote in the paper: "While we cannot predict what these discoveries will be, we hope to see a collaborative relationship between humans and machines lead to a new way of doing research."

Carnegie Mellon University researchers report in the December 21 issue of Nature that a non-organic intelligent system designed, planned and executed a chemical experiment for the first time. Source: Carnegie Mellon University

Bringing artificial intelligence and chemistry together

The system, called Coscientist, was designed by Gabe Gomes, assistant professor of chemistry and chemical engineering, and chemical engineering doctoral students Daniil Boiko and Robert MacKnight. It uses large language models (LLMs), including OpenAI’s GPT-4 and Anthropic’s Claude, to perform the entire experimental process with simple and clear language prompts.

For example, a scientist can ask a Coscientist to find a compound with a given property. The system searches the Internet, literature data, and other available resources, synthesizes the information, and selects an experimental protocol using the Robot Application Programming Interface (API). The experimental plan is then sent to and completed by the automated instrument. In summary, humans working alongside a system can design and run experiments much more quickly, accurately, and efficiently than humans working alone.

Carnegie Mellon University's Cloud Lab is a remotely operated, automated laboratory that gives researchers access to more than 200 pieces of scientific equipment. Source: Carnegie Mellon University

"In addition to the chemical synthesis tasks demonstrated by their system, Gomez and his team have also succeeded in synthesizing a super-efficient lab companion," said David Berkowitz, director of the National Science Foundation's (NSF) Chemistry Division. "They put all the parts together and the end result is much more than the sum of its parts - and it can be used for truly useful scientific purposes."

Specifically, in the Nature paper, the team shows that Coscientist can plan chemical syntheses of known compounds; search and browse hardware documentation; use documentation to execute high-level commands in automated laboratories known as cloud labs; control liquid handling instruments; complete scientific tasks that require the use of multiple hardware modules and different data sources; and solve optimization problems by analyzing previously collected data.

Expand access to advanced scientific research

"Using LLM will help us overcome one of the biggest barriers to using automated laboratories: programming capabilities," Gomes said. "If scientists can interact with automated platforms using natural language, we can open this field to more people."

This includes academic researchers who do not have access to advanced scientific instruments typically only available at top universities and institutions. Remotely controlled automated laboratories, often referred to as cloud laboratories or self-driving laboratories, bring access to these scientists and democratize science.

The Carnegie Mellon University Cloud Lab is a remotely operated, automated laboratory that provides researchers with access to more than 200 pieces of scientific equipment. Source: Carnegie Mellon University

Collaborative efforts and future prospects

Researchers at Carnegie Mellon University, working with Ben Kline of Emerald Cloud Lab (ECL), demonstrated that Coscientist can be used to perform experiments in automated robotic laboratories.

ECL co-founder and co-CEO Brian Frezza said: "The pioneering work of Professor Gomez and his team here not only demonstrates the value of autonomous driving experiments, but also pioneers a new way to use cloud laboratory technology to share the results of the work with the wider scientific community."

Carnegie Mellon University has partnered with ECL to open the first cloud laboratory at a university in early 2024. The Carnegie Mellon University Cloud Lab will provide more than 200 devices to the university's researchers and their collaborators. Gomez plans to continue developing the technology described in the Nature paper for use with the Carnegie Mellon University Cloud Lab and other self-driving labs in the future.

Gabe Gomes is an assistant professor of chemistry and chemical engineering at Carnegie Mellon University. He and his team created Coscientist, an intelligent system that can design, plan and execute scientific experiments. Source: Jonah Bayer, Carnegie Mellon University

Improve research traceability and reproducibility

Coscientist actually opened the "black box" of the experiment. The system tracks and records every step of the research, making the research work fully traceable and reproducible.

"This work demonstrates how two emerging tools in chemistry - artificial intelligence and automation - can be integrated into a more powerful tool," said Kathy Covert, program director at the National Science Foundation's Center for Chemistry Innovation. "Systems like Coscientist will enable new methods to rapidly improve how we synthesize new chemicals, and the data sets generated by these systems will be reliable, replicable, reproducible, and reusable by other chemists, thereby amplifying their impact."

Address safety and ethical issues

For Gomes, safety issues surrounding LLM, especially those related to scientific experiments, are crucial. In supporting information to the paper, Gomez's team investigated the possibility that artificial intelligence could be coerced into making dangerous chemicals or controlled substances.

"I believe that the positive impacts that AI science can bring far outweigh the negative impacts. But we have a responsibility to acknowledge the problems that may arise and provide solutions and failsafes," Gomez said.

"By ensuring that these powerful tools are used ethically and responsibly, we can continue to explore the enormous potential of large language models to advance scientific research while reducing the risks associated with the misuse of these tools," the authors wrote in the paper.

Reference: "Autonomous Scientific Research Capabilities of Large Language Models", December 20, 2023, "Nature" magazine.

DOI:10.1038/s41586-023-06792-0

Compiled source: ScitechDaily