AI for Research &​ Experimental Design​

Experimental design sits at the intersection of scientific creativity and resource constraints. This review examines how AI supports hypothesis generation, selection and planning of experiments, exploration of complex design spaces and adaptive scientific discovery. The analysis covers probabilistic and adaptive design methods such as Bayesian optimisation, active learning and surrogate modelling, as well as AI-assisted research planning with large language models and AI agents. It also looks at autonomous laboratory systems and their role in accelerating scientific discovery across domains.

• Bayesian optimisation
• Active learning / adaptive design
• ML-driven Design of Experiments (ML-DOE)
• Large language models for hypothesis generation and planning
• Autonomous and self-driving laboratories
• Data encoding and preprocessing for ML-ready experiments

• Data dependence: models are only as good as theirtraining data
• Validation and reproducibility challenges
• Miscalibrated uncertainty and confidently wrong recommendations in early-stage research areas
• Safety and cost risks of direct robotic execution
• Skills gaps between AI and scientific communities

Experimental design is one of the areas where AI can have the greatest impact on research productivity and scientific discovery. By helping researchers identify the most informative experiments, optimise scarce resources and reduce costly trial-and-error processes, these technologies can fundamentally reshape how scientific knowledge is generated. The topic is directly connected to SCIANCE activities on AI-enabled discovery, self-driving laboratories, adaptive experimentation and the identification of future research priorities for AI-enabled science.​

Europe is developing a growing ecosystem supporting AI-enabled experimentation through EOSC, EuroHPC, AI Factories, emerging Data Labs, domain-specific research infrastructures and Horizon Europe investments. These initiatives are complemented by RAISE, AI in Science Working Groups and sectoral communities seeking to connect AI expertise, scientific infrastructures and research priorities across Europe.​

• AI Scientists and autonomous research systems​
• Self-driving laboratories​
• Human-AI collaborative discovery​
• Agent-based scientific workflows​
• Generative experiment design​
• AI-assisted hypothesis generation​

Adoption is growing rapidly in materials science, chemistry, biotechnology and simulation-intensive fields. While optimisation methods are increasingly mature, fully autonomous scientific discovery remains an emerging area requiring further validation, auditing practices, reliable uncertainty estimates, safeguards for robotic execution and interdisciplinary skill building. Shared data and computing infrastructure is key to scaling these methods and to strengthening technological sovereignty.