Multi-agent artificial intelligence system developed with Gemini 2.0 emerges as a virtual scientific collaborator, an exoskeleton for researchers.
The system will make use of several specialized scientific agents.
Google has developed a new AI co-scientist, based on Gemini 2.0, which collaborates with researchers in the process of scientific discoveries. This system is designed to interact with scientists in a natural manner, allowing them to share ideas or feedback that guide AI-assisted research.
Artificial intelligence has revolutionized the field of scientific research by accelerating discoveries, improving accuracy, and managing data volumes that would be almost impossible for humans to analyze efficiently. AI-driven tools can assist in developing new medications, optimizing materials for energy storage, and facilitating climate change modeling.
Recently, several projects have been implemented aimed at making AI more useful and reliable in a scientific context. For example, an Australian research team created a generative AI tool called LLM4SD, which seeks to accelerate scientific advancements. Now, Google is also joining this trend with the goal of turning AI into a co-scientist that speeds up the discovery process.
Google's AI co-scientist is a multi-agent system that serves as a collaborative tool for scientists. It was developed through a collaboration between the teams at Google Research, Google DeepMind, and Google Cloud AI. Its purpose is to reflect the reasoning process underlying the scientific method, enabling the discovery of new original knowledge and the formulation of novel hypotheses and research based on previous evidence and tailored to specific objectives.
This system employs several specialized agents, including Generation, Reflection, Classification, Evolution, Proximity, and Meta-review, which can generate, evaluate, and refine hypotheses iteratively. Google has stated that scientists will be able to interact with the system in ways that best suit their needs, including the ability to provide initial ideas or comments on the generated outputs using natural language. Additionally, the AI co-scientist will utilize tools such as web searches and specialized AI models to enhance the foundation and quality of the generated hypotheses.
The company intends to deploy this system carefully, offering access to research organizations through a trusted testing program.
