Blab AI
Google Research VP Yossi Matias Says AI Will Accelerate Scientific Discovery but Human Judgment Remains Essential
On 11 June 2026, Yossi Matias, Google Research’s Vice‑President, told reporters that artificial intelligence will speed up scientific discovery yet cannot replace the human element of scientific judgment.
Matias, who has steered Google’s research agenda for more than two decades, underscored the company’s health‑AI initiatives. He cited a 2016 JAMA paper in which machine learning identified diabetic macular edema from retinal images. The study, conducted with partners in Thailand and India, enabled diagnoses in under two minutes—a turnaround that could prevent blindness.
He also referenced a recent collaboration with the National Health Service in England, where an AI system served as a second reader for breast‑cancer mammograms and flagged 25 % of cases that human experts had missed.
"The world of generative AI opens up a whole new realm of opportunity," Matias said. "The most exciting chapter today is how AI is being used to accelerate research itself."
At the Google Developers Conference in May, the company unveiled three tools designed to streamline the research workflow.
Literature Insights helps researchers navigate the ever‑growing volume of scientific papers. According to Matias, the tool can summarize articles, generate infographics, translate content into presentations, or even produce a podcast. It also assists with the early stages of a literature review—a process that traditionally consumes a large portion of a researcher’s time.
"Once we ask a scientific question, we still need to conduct a full literature review," he explained. "But the real work begins when we synthesize the information, generate hypotheses, filter and rank them, and present them back to the researcher."
Co‑Scientist is a multi‑agent AI partner built on Gemini. It helps researchers generate, critique, refine, and prioritize new hypotheses across life sciences, natural sciences, and engineering. Matias described it as a virtual collaborator that can act as a polymath, connecting ideas across disciplines.
"The AI Co‑Scientist is a kind of polymath in your pocket," he said. "Each of us can have a partner that connects across fields and helps make those links."
The third tool, the Empirical Research Assistant, focuses on model building. Researchers often spend days or weeks tuning parameters to find the best model for a given problem. The assistant uses generative AI to search among thousands of model configurations and return a solution that can be tested.
Matias emphasized that these tools do not eliminate the need for rigorous scientific methods. "All these tools must remain grounded in the scientific method," he said. "We need strict discipline when using AI for scientific research to ensure everything follows a rigorous verification process."
He also addressed concerns about the evolving skill set required for researchers. "Education remains extremely important," he said. "The tools accelerate research, but the ability to ask the right question and drive it forward remains essential."
Matias referenced Google’s AlphaFold project, which earned a share of the 2024 Nobel Prize in Chemistry. AlphaFold solved the protein‑folding problem, enabling the prediction of structures for hundreds of millions of proteins. "AlphaFold took a problem that once required PhD‑level work and now provides solutions for millions of proteins," he said.
He concluded that AI will enhance human innovation. "The role of humans in this equation is greater than ever," he said. "Human creativity and human connection remain essential."
In summary, Google Research is positioning AI as a tool that can accelerate discovery, streamline literature reviews, generate hypotheses, and build models. The company stresses that human judgment, rigorous validation, and the scientific method remain central to responsible research. The tools are currently available to researchers through Google’s research platform, and further development is expected as AI models continue to mature.
Matias, who has steered Google’s research agenda for more than two decades, underscored the company’s health‑AI initiatives. He cited a 2016 JAMA paper in which machine learning identified diabetic macular edema from retinal images. The study, conducted with partners in Thailand and India, enabled diagnoses in under two minutes—a turnaround that could prevent blindness.
He also referenced a recent collaboration with the National Health Service in England, where an AI system served as a second reader for breast‑cancer mammograms and flagged 25 % of cases that human experts had missed.
"The world of generative AI opens up a whole new realm of opportunity," Matias said. "The most exciting chapter today is how AI is being used to accelerate research itself."
At the Google Developers Conference in May, the company unveiled three tools designed to streamline the research workflow.
Literature Insights helps researchers navigate the ever‑growing volume of scientific papers. According to Matias, the tool can summarize articles, generate infographics, translate content into presentations, or even produce a podcast. It also assists with the early stages of a literature review—a process that traditionally consumes a large portion of a researcher’s time.
"Once we ask a scientific question, we still need to conduct a full literature review," he explained. "But the real work begins when we synthesize the information, generate hypotheses, filter and rank them, and present them back to the researcher."
Co‑Scientist is a multi‑agent AI partner built on Gemini. It helps researchers generate, critique, refine, and prioritize new hypotheses across life sciences, natural sciences, and engineering. Matias described it as a virtual collaborator that can act as a polymath, connecting ideas across disciplines.
"The AI Co‑Scientist is a kind of polymath in your pocket," he said. "Each of us can have a partner that connects across fields and helps make those links."
The third tool, the Empirical Research Assistant, focuses on model building. Researchers often spend days or weeks tuning parameters to find the best model for a given problem. The assistant uses generative AI to search among thousands of model configurations and return a solution that can be tested.
Matias emphasized that these tools do not eliminate the need for rigorous scientific methods. "All these tools must remain grounded in the scientific method," he said. "We need strict discipline when using AI for scientific research to ensure everything follows a rigorous verification process."
He also addressed concerns about the evolving skill set required for researchers. "Education remains extremely important," he said. "The tools accelerate research, but the ability to ask the right question and drive it forward remains essential."
Matias referenced Google’s AlphaFold project, which earned a share of the 2024 Nobel Prize in Chemistry. AlphaFold solved the protein‑folding problem, enabling the prediction of structures for hundreds of millions of proteins. "AlphaFold took a problem that once required PhD‑level work and now provides solutions for millions of proteins," he said.
He concluded that AI will enhance human innovation. "The role of humans in this equation is greater than ever," he said. "Human creativity and human connection remain essential."
In summary, Google Research is positioning AI as a tool that can accelerate discovery, streamline literature reviews, generate hypotheses, and build models. The company stresses that human judgment, rigorous validation, and the scientific method remain central to responsible research. The tools are currently available to researchers through Google’s research platform, and further development is expected as AI models continue to mature.
