Robin, an agent system that automates scientific research using AI, has been born, and has actually achieved a new scientific discovery and published a paper in two and a half months
FutureHouse, a nonprofit organization that builds AI agents to automate scientific research, has announced that it has built an AI agent called 'Robin' that can actually obtain new scientific knowledge.
Demonstrating end-to-end scientific discovery with Robin: a multi-agent system | FutureHouse
[2505.13400] Robin: A multi-agent system for automating scientific discovery
https://arxiv.org/abs/2505.13400
Today, we're announcing the first major discovery made by our AI Scientist with the lab in the loop: a promising new treatment for dry AMD, a major cause of blindness.
— Sam Rodriques (@SGRodriques) May 20, 2025
Our agents generated the hypotheses, designed the experiments, analyzed the data, iterated, even made figures… pic.twitter.com/ZiX9uj9s7e
Robin is an integration of two AI agents, Crow and Falcon, specialized in literature search, and Finch, a specialized AI agent in data analysis, into a single system that automates scientific research through an iterative cycle of hypothesis generation, experiment design, and data analysis.
When Samuel Rodriguez and his colleagues at FutureHouse tested the technology, they were able to identify a potential treatment for dry age-related macular degeneration, one of the major diseases that causes vision loss.
Dry age-related macular degeneration is a disease caused by the atrophy of the retinal pigment epithelium, a cell that nourishes photoreceptors. Robin, who was tasked with researching this disease, first conducted an extensive literature search using Crow and hypothesized that increasing
Robin then used Falcon to evaluate a group of molecules that might enhance phagocytosis, and Rodriguez and his team tested 10 of them in the lab and had Finch analyze the experimental data.
As a result of the analysis, Finch found that a drug called 'ROCK inhibitor Y-27632' enhanced phagocytosis of retinal pigment epithelium in cell culture.
Robin then suggested conducting an experiment called RNA-seq to see if Y-27632 was causing changes in gene expression that enhanced phagocytosis. Rodriguez and his team conducted the experiment and had Finch analyze it, revealing that Y-27632 increased the expression of a protein called ABCA1, which helps to export lipids out of cells.
As a final step, Robin investigated existing drugs that have the same effect as Y-27632 and proposed Ripasudil, which is already used to treat glaucoma, as a candidate. After Rodriguez and his colleagues performed experiments with Ripasudil, they identified it as an effective new treatment for dry age-related macular degeneration.
The entire process of Robin, from conception to publication, was completed in just two and a half months by a small research team. 'All hypotheses, experiment selection, data analysis, and figures in the main text of the paper related to this study were created autonomously by Robin. Human researchers performed the physical experiments, but the intellectual framework was entirely led by the AI. Validating these findings will require clinical trials, which will take much longer, but this is an unprecedented short period for a preliminary study,' Rodriguez and his colleagues explained.
In addition, Rodriguez and his colleagues said, 'We were able to automate hypothesis generation, experimental design, and data analysis in one system. While Robin's first results were therapeutic drugs, its versatility means it can be used in a variety of fields, from materials science to climate technology.' They also promised to release Robin as open source.
Related Posts:
