The way we think about scientific research is changing. AI has moved from being a passive tool to an active collaborator in the scientific process. Research shows that LLM is rapidly emerging as an essential research tool for PhD students.
“In the next few years, we will get to a place where we have AI scientists, so we can apply scientific advances faster,” venture capitalist Vinod Khosla said in a recent interview. , talked about the potential of AI in scientific discovery.
This vision is already taking shape. In a new study, researchers at Stanford University describe an AI-human collaboration framework designed for interdisciplinary scientific research, in this case the design and validation of new nanobodies for SARS-CoV-2, called “virtual”. Introduced the concept of “laboratory”.
The system employs a team of multidisciplinary large-scale language model (LLM) agents to assist human researchers in addressing scientific problems.
It is built in Python and uses GPT-4o as the default LLM, but unlike other research frameworks such as ChemCrow and Coscientist, which rely on fixed or predefined LLM agents, it is not possible to connect to other models. Supports easy replacement.
In such a structure, agents do more than just perform computations. They participate in discussions, propose solutions, and critically evaluate results, making them more than just tools in the scientific process.
Inside the virtual lab
The multi-agent AI architecture described in this study is designed to mimic a multidisciplinary team of scientists. The workflow begins by defining the roles of the AI agent, such as principal investigator (PI), scientist agent, and science reviewer. Research proceeds through structured “team meetings” for broad discussions and “individual meetings” for specific tasks.
AI agents will design SARS-CoV-2 variant nanobodies by collaboratively discussing, critiquing, and refining ideas using tools such as AlphaFold, Rosetta, and machine learning models. Human researchers set questions, synthesize results, and streamline complex interdisciplinary research with the help of AI.
The “Science Critic” agent ensures rigorous evaluation and increases the robustness of the output, while the “Principal Investigator” agent integrates inputs, simulates leadership, and assists in coordinating knowledge across domains. Overcome challenges.
“The LLM agent in the virtual lab serves as a very useful resource for developing research ideas, experimental workflows, and code for research projects. However, reconfirming the validity of the decisions made by the LLM agent It’s up to human researchers,” Kyle Swanson, a researcher at Stanford University and co-author of the study, told AIM. At least for now, such experiments can be carried out.
Although LLM agents can help generate research ideas, workflows, and code, researchers must verify their accuracy because they can be subject to illusions and errors.
Swanson explained that using multiple agents and scientific reviewers can help spot mistakes, but the method is not foolproof because all agents can make the same mistakes. He also noted that providing additional context, such as relevant research papers, can reduce hallucinations, but not eliminate them completely.
The future of interdisciplinary scientific discovery
A recent essay from Google DeepMind also highlights the growing role of AI in science. He said AI will accelerate knowledge sharing with LLM, help generate and label data (such as predicting protein function), and facilitate research.
Tools like AlphaProof and AlphaGeometry 2 are solving difficult problems in drug design and algorithm optimization. In May 2024, DeepMind released AlphaFold 3, an innovative open-source protein folding model that predicts with 50% higher accuracy. Two Google DeepMind scientists, Demis Hassabis and John M Jumper, have been awarded the 2024 Nobel Prize in Chemistry for their work on AlphaFold2.
The Stanford study references Jumper’s AlphaFold 2 paper as an example of interdisciplinary research. This book highlights how modern scientific research involves large and diverse teams, illustrating the collaborative nature of complex scientific research.
Beyond the virtual lab
Interestingly, the Virtual Lab architecture can also be extended to other areas of drug discovery and materials science, leveraging the broad expertise of agents. It is designed to be easily adapted to a wide range of scientific applications, and its core elements are domain independent.
“As the underlying LLM improves, we expect the virtual lab to continue to improve,” Kyle said. “We also want to incorporate more tools and provide more functionality,” he added, suggesting improvements such as enhancing internet searches and providing access to GitHub repositories. said that there is potential to acquire new knowledge and utilize tools beyond training.
Such experiments democratize access to entry-level knowledge. The virtual lab designed and experimentally validated 92 nanobody candidates over four rounds. This efficiency shows how AI collaboration can be used to achieve impactful results in resource-constrained settings.
