Google has released an AI co-scientist agent

Unlike many similar projects, the agent does not write articles or conduct end-to-end research. Its purpose is to be an idea generator and toss real scientists fresh, useful hypotheses.

The AI ​​co-scientist implements the entire cycle of the scientific method: idea generation, selection of the best, refinement, review, and so on in a circle. This is similar to how the reasoning structure is arranged in many models, but here a separate super-elaborate agent with its own tools is responsible for each stage:


The most interesting thing is that they have attached an asynchronous task management system here, so that the test-time computer is distributed dynamically, and due to this the system is easily scaled. 

By the way, the agent was tested on medical research, and it turned out that it is really capable of generating ideas that are then confirmed experimentally. 

I

Impressive! The distributed approach for hypothesis generation will accelerate research significantly.