Google DeepMind’s Joëlle Barral argues that AI is about to change how science is done, not just how fast papers are read. She sees large language models as new research tools that can search across fields, reason over data, and suggest next steps for experiments. Her view comes as AI’s role in science has just been recognized at the highest level, with the 2024 Nobel Prize in chemistry honoring work that used AI to solve core problems in biology.
Concrete wins already show what this acceleration can look like. AlphaFold 3 can predict the structures of many types of molecules and how they interact, which can guide drug design and basic biology. In materials science, scaled graph networks have helped propose promising new crystals for energy and electronics. These systems do not replace lab work, they focus it, by narrowing the search and pointing to the most useful tests.
AI will also speed discovery by linking software to robots. Early “self-driving labs”, which are automated labs controlled by machine learning, can plan and run cycles of experiments with little human intervention. This can turn slow trial and error into fast, data-driven loops while keeping scientists in charge of the goals and the checks.
Limits remain. Today’s models can still invent facts, they need high-quality data, and their results must be verified. Openness and access also matter. For example, some researchers have criticized restricted access to state-of-the-art biology models, which can slow outside testing and trust. Guardrails, reproducible methods, and better sharing will be key if AI is to help science at scale.
Taken together, these trends support Barral’s claim. AI is not a magic wand, but it is already changing the pace and the method of research, from idea generation to experimental design to analysis. With careful validation and responsible deployment, the impact is likely to be much larger than most people expect.
Nature – Accurate structure prediction of biomolecular interactions with AlphaFold 3 – 2024
Describes AlphaFold 3, which predicts the structures of complexes such as proteins, nucleic acids, and small molecules, enabling faster hypothesis testing in biology and drug discovery.
NobelPrize.org – Press release, The Nobel Prize in Chemistry 2024 – 2024
Awards chemistry’s Nobel to David Baker for computational protein design and to Demis Hassabis and John Jumper for protein structure prediction, confirming AI’s central role in modern science.
Nature – Google AI and robots join forces to build new materials – 2023
News article on automated, AI-driven materials labs that plan and run synthesis, an example of AI connected to robotics to speed experiments.
Chemical Reviews – Self-Driving Laboratories for Chemistry and Materials Science – 2024
Review of automated labs that use machine learning to plan and execute experiments, outlining benefits, challenges, and paths to wider use.
0 Comments