The authors argue that AI is shifting from a complement—where LLMs handle analysis and tools like AlphaFold provide predictions—into an automated orchestrator that manages the discovery loop. In this a la model, scientists shift from executing individual research steps to designing the systems that prompt models and audit outputs before wet lab testing. This transition makes hypothesis generation cheap, which moves bottlenecks to the verification stage and will require investment in infrastructure and institutional reform.
Tuesday, October 20, 2026
Story
DeepMind Forecasts AI Science Shift Based on 15M Gemini Interactions
AI12 days ago
James Manyika and Alex Olegimas have detailed a new framework for how artificial intelligence can fundamentally expand the boundaries of scientific discovery. In an essay for the DeepMind Institute titled "Bending the Curve of Discovery," the authors analyze the use of the GNoME model to predict millions of crystal structures and deep learning to stabilize plasma in fusion reactors. The research is based on usage data from 15 million Gemini interactions, 2,600 specialized models, and surveys of more than 600 scientists.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.