Two AI-based science assistants succeed with drug-retargeting tasks

Two new AI systems, Google’s Co-Scientist and FutureHouse’s Robin, are designed to help scientists navigate the overwhelming volume of scientific literature. These "agentic" systems operate in the background, utilizing various tools to synthesize information and propose hypotheses. Their primary goal is to assist with drug repurposing and accelerate scientific discovery by identifying non-obvious connections across disparate fields. This tackles a significant challenge in modern research: the sheer quantity of published papers makes it difficult for any individual to stay abreast of developments, even within their own field. These AI tools excel at processing vast amounts of information that humans would struggle to analyze. This frees up researchers to focus on other aspects of their work. These systems represent a collaborative approach, rather than an attempt to replace human scientists or the scientific process. Google’s Co-Scientist, for example, prioritizes a "scientist in the loop" model, ensuring human judgment guides the AI. It uses the Gemini large language model to interpret research goals and conduct literature searches, leading to hypotheses that are then evaluated for plausibility, novelty, testability, and safety. This constant human oversight helps prevent the generation of implausible or "hallucinated" hypotheses. FutureHouse’s Robin takes a similar approach but emphasizes speed and efficiency. Its tools, like Crow and Falcon, rapidly summarize and provide overviews of research papers, allowing Robin to analyze hundreds of papers in minutes—a task that would take human researchers hundreds of hours. This efficiency enables Robin to quickly generate hypotheses about disease mechanisms, as demonstrated in its work on macular degeneration. Both systems proved effective in identifying known drugs for conditions like acute myeloid leukemia and macular degeneration. While the drug suggestions were often effective for subsets of cells, this is a common outcome in cancer therapies, reflecting the diverse pathways of disease. The focus on drug repurposing highlights the practical applications of these AI assistants in accelerating medical research.