Knock.. Knock... Dear Biology, Open the Door. AI Is Here!

For years, scientists assumed their jobs were safe from AI because discovery requires creativity, intuition, and critical thinking. Three recent Nature papers challenge that assumption, showing that AI is beginning to act less like a tool and more like a research collaborator.

AI is moving at a pace that no one could have imagined a decade ago. What feels particularly unusual is how quickly it has entered our lives compared with previous technological revolutions. Electricity took nearly 50 years to become part of everyday life. The internet took around 20 years. Smartphones took almost a decade. Generative AI, however, took only two years.

In the blink of an eye, AI became a day to day tool. Whether you appreciate it, understand it, or even dislike it, AI has become an integral part of modern life. Perhaps the most striking aspect is the speed at which it arrived. Many developments that experts expected to occur over the next three or four decades have appeared within a few years. Improvements that were projected for the next decade are now happening within a single year. That is some speed.

But are we ready?

Every transformative technology creates disruption. Electricity replaced human and mechanical power. Computers replaced manual calculations. The internet made information freely available and connected people across the world. AI is now beginning to amplify human cognition itself.

The disruption has already begun. Job losses have been reported across several industries, particularly in software related fields. Like every major technological transition before it, there will be a period of uncertainty and adjustment. Eventually, things will settle, much like dust settling after a sandstorm.

Art credit: Diyansh Y. Achar

Amid all this disruption, one group appeared relatively unfazed.

Scientists!

Research is often non repetitive, poorly defined by rigid rules, and heavily dependent on creativity. Among all professions, scientific research seemed one of the least likely to be affected by AI. Scientific discovery requires generating new ideas, developing novel hypotheses, designing experiments, and interpreting complex results. These were tasks many believed belonged exclusively to trained human minds.

Until recently, researchers used AI primarily as an assistant. It helped classify images, analyze sequencing data, predict protein structures, and accelerate drug discovery. Useful tools, certainly, but still tools.

That perception changed on May 19, 2026, when three papers were published in Nature. Together, these studies suggest that AI may be capable of contributing to something much larger than routine analysis.

The first paper introduces Robin, a multi agent AI system developed by FutureHouse. Unlike conventional AI tools that perform a single task, Robin attempts to automate several core intellectual steps involved in scientific discovery.

Robin can search and synthesize scientific literature, generate biological hypotheses, propose experiments, analyze experimental results, and refine its hypotheses as new data become available. In other words, it does not simply retrieve information. It participates in an iterative cycle of hypothesis generation, experimentation, and interpretation much like a human researcher.

To test the system, researchers challenged Robin with a real biomedical problem: dry age related macular degeneration, a leading cause of blindness. Robin proposed an unexpected therapeutic strategy and identified the drug Ripasudil as a promising candidate. Follow up laboratory experiments validated the prediction. The system then analyzed RNA sequencing data from these experiments and suggested a potential molecular mechanism involving ABCA1 regulation.

Many improved versions and competing systems will undoubtedly emerge in the future, but for now Robin represents a remarkable milestone.

The second paper comes from Google DeepMind and introduces Co-Scientist. Unlike Robin, Co-Scientist is not designed to execute an entire discovery pipeline. Instead, it functions as an intellectual collaborator.

The system employs multiple AI agents that generate, critique, debate, rank, and refine scientific ideas. Rather than producing a single answer, it creates a process that resembles an exceptionally productive laboratory meeting. Researchers provide a scientific objective, and the system explores potential explanations, mechanisms, and experimental directions.

The authors tested Co-Scientist across several biomedical challenges, including drug repurposing, target discovery, and antimicrobial resistance. In multiple cases, the generated hypotheses were experimentally validated. Co-Scientist suggest that AI is moving beyond information retrieval and entering the territory of scientific reasoning and imagination.

The third paper addresses a less glamorous but equally important bottleneck in modern science: software development.

Contemporary biology depends heavily on custom computational tools. Whether working with RNA sequencing, single cell data, Hi-C datasets, or complex multiomics projects, researchers often spend months developing software before they can begin answering their biological questions.

The system described in this study, called Empirical Research Assistant or ERA, was designed to create expert level scientific software. Rather than generating isolated code snippets, ERA combines large language models with tree search algorithms to systematically improve software quality and generate complete computational solutions.

For many biologists, this may be the most immediately impactful development of the three. ERA has the potential to dramatically shorten the time between generating data and extracting biological insight.

Individually, each of these papers is impressive. Together, they reveal something far more important. For the first time, AI is beginning to participate in nearly every major intellectual component of scientific discovery.

Does this mean human scientists are becoming obsolete?

In my view, we are nowhere near that point.

Every result still requires experimental validation, biological intuition, critical thinking, and careful interpretation. However, these systems will undoubtedly increase the speed at which researchers move from questions to answers.

The more pressing question is whether researchers, particularly biologists, are prepared for this transition. How do we train the next generation of scientists for a world where AI becomes a collaborator rather than merely a tool?

The three Nature papers discussed here may eventually be remembered as early milestones in that transition. 

Robin resembles an entire research team compressed into a single system. One person collects samples, another isolates DNA, someone prepares libraries, another runs sequencing, and yet another analyzes the resulting data. Robin attempts to coordinate many of these intellectual steps simultaneously.

Co-Scientist resembles that exceptionally knowledgeable colleague who never gets tired of discussions and somehow seems to have read millions of papers.

And ERA?

ERA is simply the bioinformatician sitting quietly in the corner, staring at a screen, trying to make sense of mountains of data and present the results in a way that biologists can actually understand.

-------

References:

1. Ghareeb, A.E., Chang, B., Mitchener, L. et al. A multi-agent system for automating scientific discovery. Nature (2026). https://doi.org/10.1038/s41586-026-10652-y

2. Gottweis, J., Weng, WH., Daryin, A. et al. Accelerating scientific discovery with Co-Scientist. Nature (2026). https://doi.org/10.1038/s41586-026-10644-y

3. Aygün, E., Belyaeva, A., Comanici, G. et al. An AI system to help scientists write expert-level empirical software. Nature (2026). https://doi.org/10.1038/s41586-026-10658-6