Researchers in Sweden have developed an artificial intelligence system capable of performing several stages of the scientific discovery process with limited human intervention. The system can generate biological hypotheses, design experiments, work with laboratory robots and analyse the results to refine its next research steps.
The research was developed by scientists at Chalmers University of Technology, together with researchers from the University of Gothenburg and collaborators. The project combines large language models, automated reasoning, scientific databases and laboratory automation into what researchers describe as a closed-loop AI laboratory.
AI Moves From Analysis to Experimentation
Artificial intelligence is already widely used to analyse scientific data, search research literature and predict biological or chemical outcomes. The Swedish system takes that approach further by connecting AI reasoning directly to physical laboratory equipment.
Instead of simply giving researchers a list of possible experiments, the AI can identify potential questions, develop testable hypotheses, create experimental instructions and evaluate what happens after an experiment is performed.
The system then uses the experimental results to update its understanding and determine which questions or experiments should be investigated next.
Testing the System With Brewer’s Yeast
The researchers tested the AI scientist using Saccharomyces cerevisiae, commonly known as brewer’s yeast or baker’s yeast.
The system was provided with scientific information relating to the yeast’s genome, metabolism and previous research. According to the researchers, the database contained around 60,000 biological relationships that could be analysed by the AI.
Using this information, the system generated a large number of potential biological predictions and selected hypotheses that could be tested experimentally.
The selected ideas were converted into detailed instructions that could be executed using automated laboratory equipment.
A Robotic Scientist Called Eve
The physical experiments were carried out using a robotic laboratory platform known as Eve.
Eve was originally developed for automated biological research and drug-discovery experiments. In the new research, the platform was combined with modern AI systems and automated reasoning capabilities.
The combination allows the AI to connect digital reasoning with physical laboratory operations. Once an experiment is completed, the resulting information can be analysed by the AI and used to guide subsequent experiments.
How the Closed-Loop System Works
The research process can be divided into several connected stages.
- The AI analyses existing scientific knowledge.
- It identifies potentially important biological questions.
- It creates hypotheses that can be experimentally tested.
- The system selects and designs appropriate experiments.
- Robotic laboratory equipment performs the experiments.
- The AI analyses the resulting data.
- Successful or unsuccessful predictions are used to refine future hypotheses.
This creates a continuous feedback loop between artificial intelligence and physical experimentation.
Why This Could Matter for Science
Scientific research can require large amounts of repetitive experimentation. Researchers may spend significant amounts of time preparing samples, adjusting experimental conditions, recording results and analysing data.
An autonomous laboratory could potentially perform many of these repetitive activities continuously while maintaining digital records of the experimental process.
This could be particularly useful in areas such as biotechnology, drug discovery, genetics and systems biology, where researchers often need to test large numbers of possible biological relationships.
Automated systems may also make it easier to reproduce experiments because laboratory procedures and parameters can be recorded electronically and executed using standardized equipment.
The AI Does Not Replace Human Scientists
Despite the high level of automation demonstrated in the research, the scientists involved do not describe the system as a replacement for human researchers.
Human experts remain important for deciding research priorities, understanding the wider scientific context and evaluating ethical and safety considerations.
The researchers say future autonomous laboratories are more likely to operate as collaborative tools that work alongside scientists, helping them explore more hypotheses and perform experimental cycles faster.
Potential Applications in Medicine and Biotechnology
The technology could eventually be applied to more complex biological research.
In drug discovery, for example, autonomous systems could potentially help researchers screen biological pathways, test candidate compounds and learn from experimental results.
Similar approaches could also be explored in biotechnology, metabolic engineering, materials science and other fields where repeated experimentation is necessary.
However, the current research was conducted in a specific biological environment using yeast. The results therefore do not mean that AI systems are already capable of independently conducting all forms of medical or biological research.
A New Model for Scientific Discovery
The Swedish research highlights a broader shift in the role of AI in science. Earlier AI systems were largely designed to assist researchers with calculations, predictions and data analysis.
Newer agentic systems are increasingly being developed to perform multiple connected tasks, including planning, reasoning, tool use and decision-making.
When these capabilities are connected to laboratory robotics, AI can move beyond the computer screen and interact with physical experiments.
The researchers believe this type of autonomous laboratory could eventually help scientists investigate complex biological questions more efficiently.
What Comes Next
The next challenge will be expanding the system beyond relatively controlled experiments and determining how reliably it can operate across more complex biological problems.
Researchers will also need to ensure that autonomous laboratory systems operate under appropriate safety procedures and that discoveries generated by AI can be independently validated.
For now, the Swedish project demonstrates an important step toward laboratories in which artificial intelligence can continuously generate questions, test ideas and learn from experimental evidence.
If these systems continue to improve, the laboratory of the future could involve humans and AI working together in a continuous cycle of hypothesis, experimentation and discovery.
Journalist: Vijay Singh