At Google I/O, DeepMind CEO Demis Hassabis signaled a strategic shift in AI-driven science: from specialized tools like AlphaFold and WeatherNext toward agentic, LLM-based systems capable of conducting research autonomously. While specialized tools remain widely used — AlphaFold has served over 3 million researchers — signs of realignment are emerging, including Nobel laureate John Jumper moving to AI coding work. Google's new Gemini for Science package bundles LLM-based scientific agents like AI Co-Scientist and AlphaEvolve under one brand. Meanwhile, OpenAI demonstrated that a general-purpose reasoning model disproved a mathematics conjecture, suggesting general agents may soon contribute meaningfully to scientific research without domain-specific training. Google is carefully framing these agents as human collaborators rather than replacements, but the trajectory points toward increasingly autonomous AI scientists.

7m read timeFrom technologyreview.com
Post cover image

Sort: