Peut-on remplacer chaque scientifique humain dans un laboratoire de pointe par des agents IA capables de concevoir et de mener des expériences révolutionnaires en chimie, physique ou médecine ?

L'IA accélère la génération d'hypothèses et l'analyse des données, mais les travaux de laboratoire nécessitent toujours une présence incarnée et un jugement sensoriel. La confiance dans la découverte pilotée par l'IA sans intuition humaine reste faible.

Background

AI accelerates hypothesis generation and data analysis, but lab work still requires embodied presence and sensory judgment. Trust in AI-driven discovery without human intuition remains low.

Current systems can autonomously propose chemical syntheses or run simulations in narrow subfields, but no AI agent exists that could independently conceive, finance, secure regulatory approval for, and safely execute a true breakthrough experiment in chemistry, physics, or medicine as a human-led team in a top-tier lab does. Techniques like generative molecular design, automated lab platforms (e.g., closed-loop experimentation), and AI-guided hypothesis generation can accelerate parts of the research pipeline, yet they still depend on human oversight for goal-setting, ethical review, and cross-domain integration. Physics experiments often require complex theoretical insight and large-scale instrumentation that AI alone cannot currently orchestrate end-to-end, while medical trials involve patient safety, regulatory compliance, and unpredictable biological variability that exceed current autonomous capabilities. Systematic integration across these dimensions remains an open challenge rather than a present reality.

— Enriched May 10, 2026 · Source: Royal Society

While AI has made significant progress in assisting scientists in various tasks, such as data analysis and hypothesis generation, it is still far from being able to fully replace human scientists in top-tier labs. Current AI systems lack the creativity, intuition, and critical thinking skills that human scientists possess, which are essential for designing and conducting breakthrough experiments. Additionally, AI systems require significant human oversight and validation to ensure the accuracy and reliability of their results. The current state of the art in AI research is focused on developing more advanced tools for scientific discovery, but these tools are designed to augment human capabilities, not replace them.

— Status checked on May 10, 2026.