Google Research's senior vice president, James Manyika, used the company's annual Research@ conference to argue that artificial intelligence has already delivered more scientific progress than most people recognize, pointing to long-running programs in genomics, brain mapping, and quantum computing that predate the current wave of public attention on AI.
Speaking after a keynote that showcased work ranging from natural disaster prediction to medical AI and technical fact-checking, Manyika described Google Research as an organization modeled partly on Bell Labs, with a mandate covering both foundational computer science and applied science. The lab, now 26 years old, was founded with the idea of maintaining a dedicated research organization inside the company, and its teams have produced work that feeds directly into Google's core products while also pursuing scientific questions with no immediate commercial application.
Manyika highlighted several programs that have operated largely out of public view. Google Research has run a genomics program for 12 years, a connectomics effort to map the wiring of the brain for a decade, and a quantum computing moonshot for 13 years. The team that invented the transformer architecture, which underpins modern large language models, was also part of Google Research. «The parts that people aren't as aware of are the scientific work,» Manyika said, describing these efforts as central to the lab's identity rather than peripheral projects.
The conference also emphasized translational work in crisis response and community resilience. Manyika framed flood forecasting and similar tools as part of a broader moonshot asking what it would take for no human or computer on Earth to be surprised by a natural event. That list includes earthquakes, floods, wildfires, monsoons, and cyclones, with progress varying across categories. In health, a parallel ambition asks what it would look like if all diseases could be diagnosed, detected, understood, and eventually treated, driving research into language, accessibility, and medical AI.
Language accessibility emerged as a concrete measure of progress. Manyika said one current moonshot is supporting 1,000 languages, and that Google recently reached a milestone of 300 languages. He tied the effort directly to Google's mission of organizing the world's information and making it universally accessible, arguing that usefulness requires overcoming barriers not only of language but also of disability and speech challenges.
Manyika was candid about the risks that accompany these advances. He said Google Research spends roughly as much time on potential harms as on the exciting possibilities, dividing the challenges into performance risks such as hallucinations and risks related to misuse of the technology. He pointed to investments in privacy and SynthID, a watermarking tool that Google has expanded beyond its own systems, as examples of technical responses. But he stressed that technology alone is insufficient. «You need society to come to collective views about how we deal with those things,» he said, describing the approach as advancing ambitious and useful work while also addressing downsides and complexities.
External collaboration is a significant part of the lab's model. Manyika cited a decade-long partnership with the Howard Hughes Medical Institute and the Lichtman Lab at Harvard on connectomics, which recently produced published work on the fruit fly brain. The quantum computing team has similarly maintained long-term collaborations with academic researchers at Caltech and other institutions. These relationships, he suggested, allow Google Research to pursue scientific questions that exceed what any single organization could accomplish alone.
The broader message from the conference was that AI's role in science extends well beyond consumer-facing products. While public debate often focuses on chatbots and job disruption, Manyika's remarks placed genomics, neuroscience, disaster forecasting, and language preservation at the center of Google Research's agenda. He acknowledged that AI is not always as accurate as desired and that it may disrupt jobs and the economy in ways that cause problems, but he argued that the company is structured to pursue ambitious goals while working on the complexities they create.