During Google’s recent I/O event, DeepMind CEO Demis Hassabis discussed AI’s role in science, suggesting the world is in the “foothills of the singularity,” a term referring to the point where AI surpasses human intelligence. This statement was made after presenting WeatherNext, a tool that successfully forecast Hurricane Melissa’s impact on Jamaica, highlighting the tension between current AI capabilities and future potential.
The event showcased two main AI approaches: specialized tools like WeatherNext, designed for specific tasks, and agentic AI systems capable of autonomous scientific research. This divergence marks a shift in AI development focus from specialized applications, such as the Nobel Prize-winning AlphaFold for protein structure prediction, to more generalized AI agents.
OpenAI recently demonstrated agentic AI’s potential when one of its models disproved a significant mathematical conjecture without being specifically designed for that purpose. Google’s renewed efforts toward agent-based AI are reflected in initiatives such as Gemini for Science, which combines various AI systems offering hypothesis generation and algorithm optimization.
Despite unveiling new scientific AI systems, Google maintains its development of specialized AI tools, evident in projects like AlphaGenome and AlphaEarth Foundations. However, the company appears to be reallocating resources toward agentic AI, with key figures like John Jumper pivoting from specialized tools to AI coding projects.
While Google’s new AI systems are framed as tools to assist human scientists, the longer-term vision includes AI becoming more like scientific collaborators. Such a shift could signify a future where AI not only aids but potentially surpasses human scientific capabilities, aligning with Hassabis’s vision of AI addressing stagnation in established sciences.
Source: Technology Review
Read Original:
https://www.technologyreview.com/2026/05/22/1137813/google-i-o-showed-how-the-path-for-ai-science-is-shifting/