Decision-Making AI “Scientists” Perform Sophisticated, Interdisciplinary Research

Imagine you’re a molecular biologist wanting to launch a project seeking treatments for a newly emerging disease. You know you need the expertise of a virologist and an immunologist, plus a bioinformatics specialist to help analyze and generate insights from your data. But you lack the resources or connections to build a big multidisciplinary team. Researchers at Chan Zuckerberg Biohub San Francisco (CZ Biohub SF) and Stanford University now offer a novel solution to this dilemma. They’ve developed an AI-driven “Virtual Lab,” through which a team of AI agents, each equipped with varied scientific expertise, can tackle sophisticated and open-ended scientific problems by formulating, refining, and carrying out a complex research strategy. The agents can even conduct virtual experiments, producing results that can be validated in real-life laboratories. In a paper published in Nature, “The Virtual Lab of AI agents designs new SARS-CoV-2 nanobodies”, the team, headed by John Pak, PhD, of CZ Biohub SF, and Stanford’s James Zou, PhD, report on their development of the Virtual Lab platform, which they describe as “… an AI-human research collaboration to perform sophisticated, interdisciplinary science research.” “Interdisciplinary science research is complex, requiring increasingly large teams of researchers with expertise in diverse fields of science,” the authors wrote. But assembling and coordinating potentially large teams of multidisciplinary researchers for different specialties can be challenging. “Furthermore, it can be harder for under-resourced groups without connections to many experts across fields to engage in complex, interdisciplinary science, especially when dedicated interdisciplinary research funding is lacking,” they pointed out. In the Virtual Lab a human user creates a “Principal Investigator” AI agent (the PI) that assembles and directs a team of additional AI agents emulating the specialized research roles seen in science labs. The agents are run by a large language model (LLM), giving them scientific reasoning and decision-making capabilities. The human researcher proposes a scientific question and then monitors meetings in which the PI agent exchanges ideas with the team of specialist agents to advance the research. In addition to the PI agent and specialist agents, the Virtual Lab platform includes a Scientific Critic agent, a generalist whose role is to ask probing questions and inject a dose of skepticism into the process. “We found the Critic to be quite essential, and also reduced hallucinations,” Zou said. The team further explained, “In the Virtual Lab, a human researcher guides a set of interdisciplinary AI agents, such as a biologist or computer scientist, through a set of research meetings that tackle the different phases of a research project…The AI agents are run by an LLM that powers their scientific reasoning abilities with instructions that guide each agent’s scientific expertise and interaction with the other agents and the human researcher.” The Virtual Lab performs research both through team meetings and through individual meetings. In both cases, the human researcher provides an initial agenda to guide the discussion, and then the agents discuss how to address the agenda. In team meetings, a broad research question is discussed by all of the agents, which work together to come up with an answer. In individual meetings, a more specific task is given to a single agent, for example, writing code for a machine learning model. Given this task the agent then either works alone or together with another agent providing “critical feedback.” For their reported study, the team used the Virtual Lab platform to investigate a timely research question, designing antibodies or nanobodies to bind to the spike protein of new variants of the SARS-CoV-2 virus. After just a few days working together, the Virtual Lab team had designed and implemented an innovative computational pipeline and had presented Pak and Zou with blueprints for dozens of binders. Experiments in Pak’s lab found that two of the newly designed nanobodies bound to the spike protein of recent SARS-CoV-2 variants, a significant enough finding that Pak expects to publish studies on them. “We experimentally validated 92 mutant nanobodies designed by the Virtual Lab and found that over 90% of the nanobodies were expressed and soluble, with two promising candidates showing unique binding profiles to the recent JN.1 and KP.3 spike RBD variants,” the investigators reported in their paper. This result, they suggest, demonstrates that the Virtual Lab’s AI-human collaboration can perform complex, interdisciplinary science research that translates to validated results in the real world. The overall Virtual Lab study was led by Kyle Swanson, a PhD student in Zou’s group. The team found that while human researchers participated in AI scientists’ meetings and offered guidance at key moments, their words made up only about one percent of all conversations. The vast majority of discussions, decisions, and analyses were performed by the AI agents themselves. “What was once this crazy science fiction idea is now a reality,” said Pak, group leader of the Biohub SF Protein Sciences Platform. “The AI agents came up with a pipeline that was quite creative. But at the same time, it wasn’t outrageous or nonsensical. It was very reasonable—and they were very fast…You’d think there’d be no way AI agents talking together could propose something akin to what a human scientist would come up with, but we found here that they really can. It’s pretty shocking.” Zou is a pioneering AI researcher who has been recognized widely for breakthroughs in using AI for biomedical research, including winning the International Society of Computational Biology’s 2025 Overton Prize and being named in the New York Times’ 2024 Good Tech Awards for SyntheMol, an AI system that can design and validate novel antibiotics. “This is the first demonstration of autonomous AI agents really solving a challenging research problem, from start to finish,” said Zou, an associate professor of biomedical data science who leads Stanford University’s AI for Health program and is also a CZ Biohub SF Investigator. “The AI agents made good decisions about complex problems and were able to quickly design dozens of protein candidates that we could then test in lab experiments.” It’s become increasingly common for human scientists to employ LLMs to help with science research, such as analyzing data, writing code, and even designing proteins. Zou and Pak’s Virtual Lab platform, however, is to their knowledge the first to apply multistep reasoning and interdisciplinary expertise to successfully address an open-ended research question. “The strength of the Virtual Lab comes from its multi-agent architecture, which empowers an AI-human scientific collaboration via a series of meetings between a human researcher and a team of interdisciplinary LLM agents,” the scientists stated. “The different backgrounds of the various scientist agents leads to discussions that approach complicated scientific questions from multiple angles, thereby contributing to comprehensive answers.” The human researcher’s input is also vital for providing high-level guidance where the agents lack relevant context, for example, choosing readily available computational tools, or introducing constraints in experimental validation, they stated. Zou and Pak first met at one of the biweekly Biohub SF Investigator Program meetings. “I had just seen James give a talk at the previous Investigator meeting, where he said he wished he could do more experimental work,” Pak said. “So I decided to introduce myself.” That conversation, in the spring of 2024, sparked a collaboration that drew on Zou’s AI expertise and Pak’s expertise in protein science. When asked if he’s worried about AI scientists replacing him, Pak says no. Instead, he thinks these new virtual collaborators will just enhance his work. “This project opened the door for our Protein Science team to test a lot more well-conceived ideas very quickly,” he said. “The Virtual Lab gave us more work, in a sense, because it gave us more ideas to test. If AI can produce more testable hypotheses, that’s more work for everyone.” The results, said Pak and Zou, not only demonstrate the potential benefits of human–AI collaborations but also highlight the importance of diverse perspectives in science. Even in these virtual settings, instructing agents to assume different roles and bring varying perspectives to the table resulted in better outcomes than one AI agent working alone, they said. And because the discussions result in a transcript that human team members can access and review, researchers can feel confident about why certain decisions were made and probe further if they have questions or concerns. “The agents and the humans are all speaking the same language, so there’s nothing ‘black box’ about it, and the collaboration can progress very smoothly,” Pak said. “It was a really positive experience overall, and I feel pretty confident about applying the Virtual Lab in future research.” Zou says the existing platform is designed for biomedical research questions, but modifications would allow it to be used in a much wider array of scientific disciplines. “We’re demonstrating a new paradigm where AI is not just a tool we use for a specific step in our research, but it can actually be a primary driver of the whole process to generate discoveries. It’s a big shift, and we’re excited to see how it helps us advance in all areas of research.” The authors added, “While the experimental results here are limited to the domain of nanobody design, with future work, we envision the Virtual Lab as a powerful framework for human researchers to engage in interdisciplinary science research with the help of LLMs.” They continued, “… the Virtual Lab architecture of LLM agents and meetings is agnostic to specific research questions or scientific domains. The Virtual Lab architecture can be implemented with any set of scientist agents and any human researcher, and the conversations in the meetings will naturally adapt based on the human researcher’s agenda and the backgrounds of the agents.”

Frequently Asked Questions (FAQ)

About the Virtual Lab and AI Researchers

Q: What is the Virtual Lab platform? A: The Virtual Lab is an AI-driven platform developed by researchers at Chan Zuckerberg Biohub SF and Stanford University. It allows a team of AI agents, each with specialized scientific expertise, to conduct sophisticated and interdisciplinary research in collaboration with human researchers. Q: How does the Virtual Lab function? A: A human user acts as a Principal Investigator (PI) and directs a team of AI agents, each possessing specific scientific knowledge. These agents, powered by a large language model (LLM), engage in virtual meetings to formulate, refine, and execute research strategies, including conducting virtual experiments. A Scientific Critic agent is also included to ensure rigor and reduce potential AI hallucinations. Q: What kind of research can the Virtual Lab perform? A: The Virtual Lab can tackle complex, open-ended scientific problems that require interdisciplinary expertise. The initial study demonstrated its capability in designing new nanobodies to bind to the SARS-CoV-2 spike protein. Q: Can AI agents replace human scientists? A: According to Dr. John Pak, one of the researchers, the Virtual Lab is seen as an enhancement to human work, allowing for the rapid testing of more well-conceived ideas. It generates more testable hypotheses, potentially increasing the workload for human scientists. Q: What is the role of the human researcher in the Virtual Lab? A: The human researcher acts as a guiding force, setting the initial scientific question and agenda, and monitoring the AI agents' discussions and decisions. They provide high-level guidance, especially where AI agents might lack context or need constraints on experimental validation. Q: How are the AI agents trained or given expertise? A: The AI agents are run by a large language model (LLM) that is instructed with specific scientific expertise and interaction guidelines to emulate specialized research roles. Q: How effective is the Scientific Critic agent? A: The researchers found the Scientific Critic agent to be "quite essential" and noted that it helped to reduce "hallucinations" in the AI agents' outputs. Q: Can the Virtual Lab be applied to fields outside of biology? A: While the current platform is designed for biomedical research, Dr. Zou suggests that modifications could allow it to be used in a much wider array of scientific disciplines, as the underlying architecture is agnostic to specific research questions or domains. Q: What are the benefits of using interdisciplinary AI agents compared to a single AI? A: The multi-agent architecture allows for diverse perspectives to approach complex scientific questions from multiple angles, leading to more comprehensive answers and better outcomes, even in virtual settings.

Crypto Market AI's Take

The development of AI-driven "Virtual Labs" for sophisticated, interdisciplinary research marks a significant leap forward in scientific discovery. This mirrors the advancements we are seeing in the financial sector, particularly within the cryptocurrency market. Just as these AI agents can simulate and validate experiments, our platform leverages advanced AI and machine learning to provide sophisticated AI-powered crypto trading bots and analytical tools. These tools are designed to navigate the complexities of the crypto market, offering data-driven insights and automated trading strategies to enhance decision-making for traders and investors. The principle of specialized AI agents collaborating to solve complex problems is directly applicable to the dynamic and data-intensive world of cryptocurrency trading, where speed, accuracy, and diverse analytical perspectives are paramount for success.

Decision Making AI Scientists Perform Sophisticated, Interdisciplinary Research

Decision-Making AI “Scientists” Perform Sophisticated, Interdisciplinary Research

Frequently Asked Questions (FAQ)

About the Virtual Lab and AI Researchers

Crypto Market AI's Take

More to Read: