Biochemist Anna Pertl typed her prompt into Google’s Co-Scientist AI system on a rainy Tuesday in July and left the laboratory for the night, letting autonomous agents search hundreds of papers to tackle MYC, a protein that runs amok in most cancers and has long defied attack attempts. Advanced AI research assistants are transforming how scientists approach complex biological problems, but researchers warn that human judgment remains irreplaceable for asking the right questions and verifying machine-generated hypotheses.
AI Agents Target Complex Cancer Biology
Working at the Whitehead Institute for Biomedical Research in Cambridge, Massachusetts, Pertl and bioengineer Kalon Overholt used Co-Scientist to find practical ways of harnessing molecular droplets known as condensates to shut down MYC. The request built on 2018 research by Whitehead biologist Richard Young, who found that cells switch on crucial genes by gathering regulatory proteins into condensates at genome-control regions called super-enhancers. While Young co-founded Boston-based Dewpoint Therapeutics to explore dissolving tumor cell condensates directly, Pertl wanted to see if AI could generate a fresh line of attack.
Getting a workable hypothesis required careful human intervention. Co-Scientist initially assumed the protein clusters were the drug rather than the target and that they always activated gene expression. After Pertl and Overholt corrected these misconceptions, the AI combed through more than 700 scientific papers overnight and generated 108 possible strategies. It rejected all but one approach, proposing to glue protein clusters together using click chemistry to form a mass that halts MYC DNA reading—a concept Overholt calls extremely compelling.
Trust, But Verify in Automated Research
Frontier AI labs including Anthropic, OpenAI, FutureHouse, and Phylo are rolling out systems that tackle tasks once reserved for human scientists. Vivek Natarajan, an AI researcher at Google who helped develop Co-Scientist, describes the tool as a collaborative partner. However, Fyodor Urnov, who studies genome editing at the University of California, Berkeley and uses FutureHouse’s Kosmos platform, relies on a Reagan-Gorbachev era proverb: trust, but verify. Researchers caution that taking AI outputs on faith risks scientific accuracy, especially as platforms handle complex multistep workflows.
Pro Tip: When using AI research assistants for open-ended science, hold back unpublished experimental data to test the system objectively, as researchers at Imperial College London did during a 2025 study on bacterial DNA swapping.
The Changing Value of Human Expertise
As machines take over routine information synthesis and hypothesis generation, the scarce resource in laboratories is shifting toward human scientific judgment. Ajay Agrawal, an economist at the University of Toronto Rotman School of Management, notes that the most valuable part of the process is now asking the question. Similarly, Stanford molecular geneticist Le Cong emphasizes that tools from start-ups like Phylo aim to move humans up the value chain rather than replace them entirely, freeing investigators to focus on consequential decisions.
This reliance on automation creates a distinct paradox, according to biomedical computing researcher Hector Zenil of King’s College London. Investigators may need deeper knowledge than ever to evaluate AI-generated ideas, even as students get fewer opportunities to build that expertise through manual experimentation. Philine Guckelberger, a molecular geneticist at FutureHouse and Stanford, experienced this firsthand when Kosmos initially stumbled over data labeling in a 30-million-row data set before uncovering an unexpected pattern in cancer cells once corrected.
Frequently Asked Questions
What is Google’s Co-Scientist?
Co-Scientist is an artificial intelligence system designed to produce innovative scientific hypotheses by launching autonomous AI agents that search literature, evaluate explanations, and test ideas against published evidence, according to developers at Google in Mountain View, California.
Can AI systems conduct scientific research entirely on their own?
No. According to researchers using platforms like Co-Scientist and Kosmos, human scientists must frame the initial research questions, correct system misconceptions, and rigorously verify machine-generated hypotheses against experimental data.
Which companies and institutions are developing AI research assistants?
Major contributors include Google, Anthropic, OpenAI, and specialized start-ups such as FutureHouse and Phylo based in San Francisco, alongside groups like Stanford University and Huawei Technologies.
Explore Further Scientific Innovations
Are you tracking the intersection of artificial intelligence and laboratory research? Share your thoughts in the comments below, explore our archive of scientific technology coverage, or subscribe to our newsletter for weekly updates on automated discovery.
Related reading