Google’s AI “Ko-Scientist” company is based on the Gemini Language Model
Raa / Nurphoto / Shutterstock
Google has presented an experimental artificial intelligence system “to help scientists synthesize many quantities of literature, create new hypotheses, and propose specific research plans, according to his press release. “The idea of the AI co-scientist” is to give scientists’ superpowers, “says Alan Karthikosalingam on Google.
Tool, who does not have an official name, builds it on Google Gemini large language models. When a researcher asks a question or determines a goal, to find a new drug, say – the tool comes out with initial ideas in 15 minutes. Several Gemini’s agents, then “discussed” these assumptions, classify them and improved during the next few hours and days, Vivek Natarajan says.
In this process, the agents can seek scientific literature, use databases and use tools such as the Google Alphafold system to predict the structure of proteins. “In addition to continuous improvement ideas, they discuss ideas, they criticized ideas,” says Natarajan.
Google has already accessed a system of research groups that have released short papers describing their use. The teams who have tried are enthusiastic about its potential, and these examples will be helpful to the AI co-scientist to synthesize discoveries. However, it is questionable whether examples accept a claim that new hypotheses can cause.
For example, Google says that a team uses the system to find “new” “new” ways to treat liver fibrosis potential. However, the drug proposed by AI has been previously studied for this purpose. “The identified drugs are well established to be antifibrotic,” he says Steven o’reilly United Kingdom Biotechnology Company in Alcyomics. “There’s nothing new here.”
This potential use of treatments is not new, teammate Gary Pelz At the school of California Stanford University, two of the three drugs selected by AI cient scientists were promises with the tests of human liver organs, and there were no more evidence to protect the two options that were not personalized. Peltz says Google gave a small amount of finance to cover the test costs.
In another paper, José Penadés Imperial College in London and his colleagues describe how scientists suggest that the hypothesis that matches an unpublished discovery. He and his team learn mobile genetic elements – DNA bits that can be moved between bacteria according to various resources. Some mobile genetic items kidnapped the bacteriophagous viruses. These viruses form a tail that are associated with DNA and specific bacteria and DNA is injected. So when an item can be accessed by a Phagan virus shell, he gets a free route to another bacterium.
A type of mobile genetic item makes its shells. This type is particularly widespread, the penadés and his team, in fact, the virus of any Phagen can only contaminate a narrow range of bacteria. The answer, they recently found, which is that these shells can be linked to tails of different faje so that the mobile element can be accessed in a wide range of bacteria.
While the discovery was still unpublished, the team asked the co-scientist to explain the puzzle – and a suggestion asked to steal tails from different faje.
“We were surprised,” says Penadés. “I have sent Google to access my computer, you can access my computer. Is that that? Otherwise I can’t believe what I’m reading here.”
However, the Group published a paper in 2023 – it was nurtured to the system – this family of mobile genetic elements how it was “Steals bacteriocerous tails to spread in nature.” At the time, the researchers believed that the items were limited to the phases of the phases that pollute the same cell. Only later discovered items can pick up tails that float around external cells.
Therefore, the AI KO-Scientist is the way you explained how the correct answer was created has lost his apparent limit.
What is clear is that he had to find everything he needed to find the answer, rather than come with a completely new idea. “Everything was already published, but in different bits,” says Penadés. “The system has been able to put everything together.”
The groups were already in the market tried by some AI systems, and one of them is not created with the answer, he says. In fact, some did not manage when the role that describes the answer even when it feeds. “The system suggests things you never thought about,” says Penadés, Google has not received any financing. “I think it will be a variable game.”
Changing the game really will be clearer over time. Google follow-up when making claims about AI tools to mix scientists. Its system lives on the Alfafo-Wear Hype, The Nobel Prize won the team behind it Last year.
In 2023, however, the company announced 40 “about new material” It has been Synthesized with his support of Gnome AI. However, according to 2024 analysis Robert Palgrave University College in London, One of the synthesized materials was not really new.
Despite findings, Palgrave believes that he can help AI scientists. “In general, I think AI has a large number to help science, if established in collaboration with the corresponding areas,” he noted.
Themes:
