An artificial intelligence system presented in the journal Science can autonomously perform several biomedical research tasks. Named Biomni, it is not limited to answering questions: it can chain together operations necessary for the analysis of biological data.
The researchers describe Biomni as a generalist agent. This type of system receives an objective, chooses intermediate steps, uses specialized tools, and then adapts its work according to the results obtained. It thus distinguishes itself from a model that only produces a textual response.
A set of test tubes filled with liquids of different colors.
Illustration image Unsplash
In biomedical research, tasks are often distributed across multiple software programs. For example, one must prepare data, query biological resources, launch analyses, and interpret results. The interest of an agent is to link these operations into a single procedure.
Biomni was designed to perform diverse tasks in several fields of biology. The publication presents this capability as a step toward assistants that can help researchers explore hypotheses more quickly and process ever more complex datasets.
This autonomy does not mean that the system replaces a scientist. Results depend on the data provided, the tools available, and the quality of decisions made at each step. An error in an analysis can also propagate to subsequent steps.
Human verification therefore remains essential, especially when results can guide an experiment or a medical decision. An agent can accelerate exploration, but it does not automatically transform a hypothesis into a proven discovery.
Biomni's interest lies mainly in the organization of work. Systems capable of planning and executing analyses could reduce certain repetitive tasks, while leaving the formulation of questions and critical evaluation of results to the researchers.