Google used its I/O keynote to showcase a changing vision for scientific Artificial Intelligence. Demis Hassabis paired sweeping language about being near the “foothills of the singularity” with a demonstration of WeatherNext, a weather prediction system that gave advance warning ahead of Hurricane Melissa’s landfall in Jamaica. That contrast captured a wider divide in scientific Artificial Intelligence: one path emphasizes tightly trained systems built for specific scientific tasks, while the other centers on agentic, large language model-based systems that could eventually conduct research with limited or no human involvement.
Google has not stepped away from specialized scientific systems. AlphaGenome and AlphaEarth Foundations, which are trained for genetics and Earth science applications respectively, were released last summer, and the newest version of WeatherNext came out in November. Last year, for instance, Google reported that protein structure predictions from AlphaFold have been used by over three million researchers worldwide. The company also continues to highlight real-world utility, especially where specialized systems can produce practical benefits for scientists and the public.
Even so, signs of a strategic realignment are becoming harder to ignore. Last month, the Los Angeles Times reported that Google fellow John Jumper, who won the Nobel for AlphaFold, is now working on Artificial Intelligence coding, not on science-specific Artificial Intelligence tools. Google’s major scientific announcement at I/O was Gemini for Science, a new package that groups several of its large language model-based research systems under one brand. That includes Co-Scientist, which generates hypotheses, and AlphaEvolve, which optimizes algorithms, with Google opening applications for researchers to gain access.
The broader industry is also reinforcing this direction. OpenAI said this week that one of its models had disproved an important mathematics conjecture, a result some mathematicians viewed as a major step for generative Artificial Intelligence in research. Google is framing its own systems as collaborators rather than replacements. Hassabis said, “For the next decade or so, we should think about Artificial Intelligence as this amazing tool to help scientists,” while leaving open the possibility that systems may later become more like collaborators. That framing still points toward a future in which scientific Artificial Intelligence becomes more autonomous, and in which Google appears increasingly interested in building general agents that can contribute across domains rather than only highly specialized tools.