From Narrow AI to Autonomous Systems: How AI has and might transform science

Narrow AI has already improved science performance through applications like fault prediction, physics‑aware reinforcement learning, and uncertainty‑aware data analysis, the next logical step is to embed these capabilities within fully autonomous scientific infrastructures. Autonomous systems efforts promise to accelerate discovery by deploying agents that manage the entire research workflow—from literature review and hypothesis generation to instrument control and adaptive experimentation. By orchestrating multi‑agent workflows, these systems enable real‑time instrument adaptation, automated protocol generation, and data‑driven decision making, thereby transforming traditionally manual processes into intelligent, scalable pipelines that enhance both efficiency and innovation.