Scientists in Sweden Build an AI System That Can Design and Test Its Own Experiments

Researchers in Sweden have developed an artificial intelligence system that can generate scientific questions, design experiments and analyse the results with minimal human intervention.

The system was created by researchers at Chalmers University of Technology, the University of Gothenburg and the University of Cambridge. Their work focuses on biological research using brewer’s yeast.

The researchers connected large language models and automated reasoning with laboratory equipment. The AI can examine scientific information, identify promising questions, suggest experiments and then use the results to refine its next set of hypotheses.

In the study, the system analysed a large collection of information about the biology of brewer’s yeast and generated thousands of testable predictions about how different nutrients could affect the cells.

Selected predictions were then converted into instructions for automated laboratory equipment. After the experiments were completed, the AI analysed the results and used the new evidence to update its understanding.

The researchers describe the work as an important step toward so-called self-driving laboratories, where machines can perform repeated cycles of scientific investigation much faster than researchers working entirely by hand.

However, the system is not a replacement for human scientists. Researchers still set the boundaries of the work, determine important priorities and provide oversight. The team says humans will remain essential for understanding the wider significance of discoveries and ensuring that research is conducted responsibly.

The development could eventually help scientists investigate complex biological questions more quickly and make better use of laboratory resources. It also shows how AI is moving beyond analysing existing information and beginning to participate directly in the experimental process.

Source: Chalmers University of Technology, September 30, 2026; Journal of the Royal Society Interface