Toward autonomous science with agentic artificial intelligence.

Abstract

Artificial intelligence (AI) in biomedicine has evolved from pattern-recognition for medical image analysis to generative models like AlphaFold that predict protein structures. Tools such as Elicit, GPT-Rosalind, and Claude Science now bridge prediction and reasoning through literature synthesis, domain-tuned models, and agentic workbenches. Three recent systems (Co-Scientist, Robin, and Biomni) demonstrated agentic AI capable of generating hypotheses, designing experiments, computational analysis, and refining reasoning through experimental feedback. This perspective traces the emergence of agentic scientific reasoning, discusses scaling laws and autonomous laboratories, and argues that AI's transformative potential is inseparable from challenges in hallucination, bias, irreproducibility, and dual use. Scientific progress will depend on responsible human-AI collaboration.