MIT Technology Review is reporting that a significant tension exists between two approaches to artificial intelligence in science, highlighted by recent statements from Google DeepMind CEO Demis Hassabis. During Google I/O, Hassabis declared humanity is in the “foothills of the singularity,” a theoretical point where AI surpasses human intelligence. The Review noted this lofty rhetoric was juxtaposed with a presentation on WeatherNext, Google’s weather prediction software, which provided an advance alert for Hurricane Melissa in Jamaica, potentially saving lives.
The Review explained that one approach focuses on specialized AI tools, like WeatherNext or AlphaFold, designed for specific scientific problems. The other involves agentic, large language model-based systems that could eventually conduct cutting-edge research autonomously. It said there is growing enthusiasm for these agentic systems, including concepts like recursive self-improvement, where AI drives its own advancement. The Review cited Google Cloud’s chief scientist, Pushmeet Kohli, who wrote that AI is moving beyond merely facilitating science to actively doing science.
Google appears to be realigning its focus and resources. The Review reported that John Jumper, a Nobel laureate for AlphaFold, is now working on AI coding rather than science-specific tools. This shift may signal a prioritization of agentic science, as coding abilities are crucial for such systems. OpenAI also recently announced that one of its general-purpose reasoning models disproved an important mathematics conjecture, marking a significant contribution.
Google is dedicating attention to an agent-driven scientific future with its new Gemini for Science package, which includes the hypothesis-generating AI Co-Scientist and algorithm-optimizing AlphaEvolve. While specialized tools remain popular, the company seems to be shifting its public image and some personnel toward agentic systems. Hassabis frames AI as an “amazing tool to help scientists” for the next decade, potentially evolving into collaborators or even surpassing human capabilities, driven by his belief that AI could overcome stagnation in scientific progress.
Full Article: Google I/O showed how the path for AI-driven science is shifting