From algorithms to experiments: AI’s evolving role in biological research

Artificial intelligence (AI) is increasingly playing a significant role in biological research, with agentic AI systems capable of performing tasks autonomously and adapting by revising their workflows. Google's AI Co-Scientist, for instance, is a multi-agent AI system that assists researchers in generating ideas, identifying similar concepts, conducting peer review, and refining research hypotheses based on scientific literature and data.

Collaboration between computational and lab experiments is expected to improve with the advent of automated labs, cloud computing, and autonomous labs, as seen in Anthropic's recent setting up of a wet lab in San Francisco Bay area.

One notable example of AI's role in biological research is the autonomous lab system conducted by OpenAI and Ginkgo Bioworks, where GPT-5 was connected to a cloud lab to optimise a biological method - cell-free protein synthesis. This AI-powered lab could execute pre-defined experimental tasks and make decisions, but its capabilities might be limited to simpler processes.

However, another study revealed that an autonomous lab system could determine optimal growth conditions for a particular strain of E.coli, but could not determine conditions suitable for glutamic acid production, a more complex task.

Despite the growing role of AI in biological research, human judgement remains essential. AI-generated claims may contain errors, as models can hallucinate or misinterpret data. Researchers must verify AI findings and possess critical thinking skills, practical expertise in troubleshooting experiments, and an understanding of underlying biology, data, laboratory infrastructure, reagents, and materials.

Furthermore, transparency about AI's contribution to the research process is crucial, with disclosure of when and how AI has been used. AI's effectiveness also relies on the quality of data it has been trained on, necessitating the careful curation and standardisation of biological databases, such as the Protein Data Bank used by AlphaFold.