Google launched a new artificial intelligence (AI) system on Wednesday that could help scientists conduct scientific discovery processes. Called AI Co-Scientist, the tool is powered by the Gemini 2.0 AI model and includes a multi-agent system that specializes in different tasks in the discovery process. The mountain view-based tech giant said AI systems are designed to be a collaborative tool, not to replace human scientists. Currently, the public domain is not available, and Google plans to further test the system before it releases it.
Google reveals co-scientious scientists in artificial intelligence
In a blog post, the tech giant details the new AI system. Multi-institutional tools can reflect the reasoning process used by the scientific community for new discoveries. Some of the tasks it can perform include standard literature reviews, abstracts, research hypotheses and suggestions, and achieving specific research objectives. It is worth noting that AI co-scientists have also adopted in-depth research tools recently released to Gemini’s advanced users.
AI systems have several professional AI agents including power generation, reflection, ranking, evolution, proximity and meta-comment. Google says these agents are inspired by popular scientific methods. These agents can complete a single task and interact with each other to iterate, evaluate and refine assumptions. These agents are also monitored by the supervisory agent.
“We see promising early results in important areas of research, such as liver fibrosis treatment, antimicrobial resistance and drug reuse,” Google CEO Sundar Pichai said in a post on X (formerly Twitter).
To begin the process, scientists can interact with the system by specifying research goals through natural language. They can also propose seed ideas and suggestions that require AI to propose hypotheses around the surrounding hypotheses. Despite the work of AI co-scientists, scientists can also provide feedback on the generated output to further adjust the process. AI systems use tools such as web search and professional AI models to enhance the grounding and quality of generated outputs. Details of how the tool works are detailed in the research paper.
One of the unique aspects of this AI system is the test time calculation scaling feature, which makes it guess for a second time and verify its response. This allows AI to co-scientists to iterate over causes, evolve and improve output.
However, it should be mentioned that Google’s AI joint scientists cannot make real innovations, which is usually a prerequisite for making scientific discoveries. Finally, all its information and assumptions come from its existing database or web search. Although its reasoning ability does allow it to expand ideas and test its effectiveness, it is unlikely to be more effective than an assistant.
The full functionality of the AI system can only be evaluated after the tool is released. Currently, Google is evaluating its advantages and limitations in science and biomedical aspects. The company has launched a trusted tester program through which it can access AI co-scientists into research organizations. Those interested can join the program by filling out this Google Form.
