Google presents a “co-scientific” AI: accelerates research and creates new hypotheses

Usually, researchers combine their ingenuity and creativity with knowledge, scientific publications and experience, to generate new research addresses and achieve scientific advances. However, in many fields, the rate of scientific publications grows rapidly, and navigate between these publications to achieve new research is raised as a “challenge.”

Faced with this scenario, Google has announced a new AI to which it refers as a ‘co-scientist’, which is based on a collaborative tool promoted by the AI ​​of the Gemini 2.0 model, capable of helping scientific experts to Carry out tasks such as collecting information and “perfecting your work” to accelerate discoveries.

However, as the technology explained in a statement in her research blog, this AI has not been devised to automate the scientific process of an investigation, but has been raised as an option to assist professionals when generating new Hypothesis or raise new research proposals.

To use AI ‘co-scientist’, researchers can specify a research objective through natural language such as better understanding the spread of a microbe that causes diseases. Based on this, the AI ​​will propose verifiable hypotheses, together with a summary of the relevant published documentation and a possible experimental approach.

Likewise, scientists can interact in other ways with AI, for example, directly providing their own ideas for exploration or offering comments on the results generated in natural language.

Thus, Google has detailed that, in terms of operation, this AI system is built on multiple agents based on Gemini 2.0 and, with this, it manages to reflect the reasoning process that supports the scientific method.

Specifically, the process is based on the fact that the IA analyzes the objective assigned in the configuration of the research plan, administered by a supervisory agent, which, in turn, assigns the agents specialized to the work tail and establishes resources to carry out The tasks.

In this way, the “co-scientific” AI is able to synthesize complex issues and carry out a long-term planning and reasoning. In addition, it is able to “take advantage of the scalability of the calculations in time of proof” to reason, evolve and improve iteratively in its results.

“The nature of the system facilitates recursive self -criticism, including the use of tools for feedback in order to refine hypotheses and proposals,” said the technological.

Therefore, as the company has underlined, in addition to reviewing published scientific documents, summarizing and using ‘Deep Research’ tools, this system is intended to “discover new and original knowledge”, as well as formulate hypotheses and research proposals ” demotrably novel ”, adapting to specific research objectives.

So much so that, according to their internal tests, the capacities of the “co-scientific” AI to generate novel and verifiable hypothesis in various scientific and biomedical domains-some already experimentally validated-as well as its ability to recursive selfmexing with greater capacity to Computing, “they demonstrate their potential to accelerate the efforts of scientists to address great challenges in science and medicine.”

With all this, Google has pointed out that scientists who are part of their trusted testers will have early access to this new “co-scientific”, in order to start trying their abilities.