Virginia Tech is pioneering a new approach to scientific research known as agentic science, which employs artificial intelligence (AI) agents to facilitate and enhance the research process. This concept, while seemingly complex, aims to simplify scientific inquiry through the use of trained AI systems. Hongliang Xin, a professor of chemical engineering at Virginia Tech, along with collaborators John Kitchin from Carnegie Mellon University and Heather Kulik from the Massachusetts Institute of Technology, recently published a commentary in Nature Machine Intelligence, discussing the foundational aspects, potential applications, and ethical considerations surrounding agentic science.
In an interview, Xin elaborated on the essence of agentic science, describing it as a novel paradigm that positions AI at the core of scientific reasoning, planning, and execution in both digital and physical environments. He likened the relationship between human scientists and AI agents to that of musicians in an orchestra, where each AI agent contributes to a collective output that is richer than what either could achieve individually. In this analogy, human scientists act as conductors, ensuring that all elements work cohesively.
Kulik noted the transformative potential of agentic science, particularly in enabling scientists to formulate and test hypotheses that may have previously been overlooked. She highlighted how AI can automate tedious laboratory tasks, significantly accelerating innovation across various scientific domains.
Kitchin expressed enthusiasm for recent advancements, pointing out that AI agents can now perform tasks that were unimaginable just a year ago, such as gathering information from diverse sources to inform experimental planning. Kulik added that the efficiency of AI could lead to breakthroughs in areas where traditional methods have fallen short, particularly in environmental and health sciences.
The timing for such developments is crucial. Kulik explained that recent improvements in language models and the ability to train large AI systems on extensive datasets have paved the way for agentic science. Additionally, Xin mentioned the overwhelming increase in scientific publications, which makes it challenging for researchers to keep up. AI”s ability to process large volumes of information offers a solution to this problem, serving as a “brain” that makes informed decisions within autonomous laboratories.
As this technology evolves, Kitchin emphasized the importance of maintaining human oversight in laboratory settings to ensure safety and reliability. He argued that human responsibility remains paramount, even as AI systems take on more complex roles.
Collaboration among research institutions is vital to the success of agentic science, according to Kitchin. With the myriad of challenges and potential applications of AI agents, no single organization can address every aspect. He stressed the importance of sharing knowledge and experiences to maximize the benefits of these technological advancements.
