Multi-agent AI system built with Gemini 2.0 set to become a virtual scientific collaborator, a pseudo-exo skeleton for researchers

Google’s AI co-scientist, built on Gemini 2.0, collaborates with researchers for discoveries
It uses specialized agents to generate, evaluate, and refine scientific hypotheses
Scientists can interact naturally, providing ideas or feedback to guide AI research

Artificial intelligence has already had a major impact on scientific research by accelerating discoveries, improving accuracy, and handling vast datasets that would be near-impossible for humans to analyze efficiently. AI-powered algorithms can assist in the discovery of new drugs, optimize materials for energy storage, and aid in modeling climate change.

A number of projects have been set up to make AI more useful and more reliable in a scientific setting. We’ve previously written about the concept of the “exocortex,” which aims to provide a bridge between the human mind and a network of AI agents, and more recently, an Australian research team developed a generative AI tool called LLM4SD (Large Language Model for Scientific Discovery), designed to speed up scientific breakthroughs.

Now, Google is also launching a similar initiative, which aims to turn AI into a co-scientist that can accelerate scientific discoveries. The tech giant explains, “The AI co-scientist is a multi-agent AI system that is intended to function as a collaborative tool for scientists.”

Deploying specialized scientific agents

The AI co-scientist is built on Google’s Gemini 2.0 and is the result of collaboration between Google Research, Google DeepMind, and Google Cloud AI teams. It is designed to “mirror the reasoning process underpinning the scientific method.” Google says that its system is intended to “uncover new, original knowledge and to formulate demonstrably novel research hypotheses and proposals, building upon prior evidence and tailored to specific research objectives.”

The system will use a number of specialized agents – Generation, Reflection, Ranking, Evolution, Proximity, and Meta-review – that can iteratively generate, evaluate, and refine hypotheses. Google says that scientists will be able to interact with the system in whatever way best suits their needs. This will include providing their own seed ideas or feedback on generated outputs in natural language.

“The AI co-scientist also uses tools, like web search and specialized AI models, to enhance the grounding and quality of generated hypotheses,” Google says.

Not wishing to rush its deployment, the company plans to offer access to the system for research organizations through a trusted tester program.

Source link