AI’s Promise in Biomedical Research Comes With Risks for Future Biologists

The stakes are inside the lab

AI’s most immediate stakes for science may be less dramatic than rogue appliances or threats to humanity. For biomedical researchers, the pressing question is how it could change the work of doing science: how researchers form questions, assess evidence, interpret results and learn the habits that make scientific work reliable. Those changes may arrive through ordinary research tasks, rather than through a single dramatic breakthrough.

That distinction matters. Public discussion often treats AI as either a tool that will accelerate discovery or a force that could put people at risk. Biomedical research raises a more practical question: what happens when software becomes part of the day-to-day process by which scientists learn and make judgments? The answer depends on how researchers use it, what tasks they hand over, and whether they can still assess the output for themselves.

Convenience can change how scientists learn

AI can be useful, but usefulness doesn't guarantee that a tool improves a researcher’s judgment. Systems that help summarize material, draft text, generate code or organize information can reduce the time spent on routine work. They can also make it easier to produce an answer before a researcher has fully understood the question. In science, that difference matters because the process of getting to an answer is part of the training.

Early-career biologists learn by taking on work that can look repetitive from the outside: reading closely, checking methods, tracing how a result was produced, and noticing when a conclusion outruns the evidence. If an AI system takes on parts of that work, researchers may have fewer chances to practice those skills. They might still complete tasks faster, yet have a weaker grasp of the choices behind the output.

That risk isn't inevitable. A researcher who uses AI to draft a first pass, then checks it against the underlying papers or data, may use the system as a prompt to think more carefully. Someone who accepts a polished answer without checking its basis may miss errors that are hard to spot precisely because the answer sounds confident. The practical issue isn't whether AI appears in the workflow. It's whether the scientist can explain and defend what the system contributed.

Scientific work includes judgment

Research isn't a simple sequence of tasks that can be divided neatly between people and software. A biomedical project can involve decisions about which question to pursue, how to interpret an ambiguous result, whether an experimental approach is appropriate, and what limitations to disclose. Some parts may be easier to automate than others, but the boundaries aren't fixed by the software alone. They depend on how a lab defines responsibility and what it expects researchers to understand.

Even a system that produces a useful answer doesn't establish that the answer is sound. Researchers need ways to check where information came from, whether assumptions are appropriate, and whether a result can be reproduced or challenged. If those checks become an afterthought, speed can create a misleading impression of progress. More output isn't automatically more knowledge.

(And this is the part most people overlook.) The concern isn't only that a system might make a mistake. It's also that repeated reliance could change which skills researchers practice and which ones they stop developing. A mistake can be corrected when someone recognizes it. A gap in judgment is harder to see.

Jobs and career paths may change

The source raises another concern: AI could displace jobs in science. That possibility deserves attention without assuming that every task now done by a person will disappear. In many fields, new tools shift the mix of work before they eliminate a whole role. In research, the effects could depend on which tasks become easier to automate and how institutions respond.

For people starting a scientific career, the consequences could be especially complicated. Some entry-level work is tedious, but it can also teach how evidence is gathered and where errors tend to enter a process. If that work shrinks, labs may need other ways to give newer researchers hands-on experience. Otherwise, they could ask people to make more sophisticated judgments without giving them enough practice to develop those judgments.

For established researchers, the pressure may look different. AI could change expectations about how quickly work gets done, or about how many tasks one person can manage. That might reduce some burdens, but it could also shift attention toward reviewing more material and checking machine-produced work. A tool that saves time on one step can create new work elsewhere.

What responsible use would require

There isn't a single rule that will fit every research task. A lab using AI to help organize notes faces different questions from one using it to analyze data or shape a scientific argument. Researchers need to know what a system was asked to do, what material it handled, and how its contribution was checked. Without that clarity, it becomes difficult to judge the reliability of the work or to learn from mistakes.

Training matters, too. If you're working in this space, the useful question isn't just which tool can finish a task fastest. Ask what you need to understand well enough to verify the result, and whether using the system helps you build that understanding. That may mean treating an AI-generated response as a draft or suggestion, not as a substitute for reading the methods or examining the evidence.

Here’s the thing: scientific competence includes knowing when an answer is uncertain, incomplete or beside the point. A fluent system can make uncertainty less visible, which places more weight on the person using it. Researchers and institutions will need to decide how to preserve opportunities for learning while adopting tools that can make some work easier.

The significance is in the habits that follow

This is more significant than it looks. The effect of AI on biomedical science won't be measured only by whether it helps produce a result more quickly. It will also be seen in whether researchers retain the ability to question that result, explain how they reached it and recognize when the available evidence doesn't support a confident conclusion.

And yet, rejecting useful tools outright wouldn't answer the underlying question. AI may help with parts of research, while its use creates new demands for checking, training and accountability. The challenge is to make those demands explicit, particularly for people whose first experience of scientific work may include AI from the start.

There’s no simple verdict in the source material on how large these changes will be. For now, the central concern is a familiar one in science: tools can extend what people do, but they can also alter what people learn to do. The balance will depend on choices made in labs and training, not just on what the technology can produce.

Continue to STAT+ to read the full story…