The Problem With Information Overload in Biomedical Science
We interviewed dozens of researchers before building NousLab. Almost all of them described the same problem, and the same quiet resignation about it.
When we started building NousLab, we spent the first few months talking to researchers before writing a single line of code. We interviewed oncologists, pharmacologists, people running university research units, people working inside pharma. We asked them all a version of the same question: where does your time actually go?
The answers were remarkably consistent. Not in the specific details, but in the shape of the problem. Almost everyone described some version of the same thing: they were drowning in literature they could not read, making decisions based on a picture of the evidence they knew was incomplete, and feeling vaguely guilty about it.
That guilt is worth sitting with for a moment, because it tells you something important about how science actually works versus how it is supposed to work.
The numbers are not abstract
PubMed indexes roughly 1.5 million new biomedical articles every year. That works out to around 4,000 papers per day, every day, including weekends. In any reasonably active subspecialty, a researcher who wanted to read everything published in their field would need to get through 50 to 100 new papers a week just to stay current, before doing any actual research.
Nobody does that. What researchers actually do is develop filters. They follow specific journals. They read abstracts and skip full texts. They rely on colleagues to surface relevant work. They set up keyword alerts and skim the results. They read the papers that get cited in the papers they are already reading.
These filters work, up to a point. The problem is that they are invisible and inconsistent. Two researchers working on adjacent questions in the same institution may be drawing on almost entirely non-overlapping bodies of literature, with neither of them aware of what the other is missing. A finding that would change someone's experimental design sits unread in a database because the title did not match the search terms they happened to use.
This is not a new problem. But it is getting worse faster than people realise.
The volume of biomedical literature has been doubling roughly every nine years since the 1950s. PubMed currently holds over 36 million citations. What has changed recently is not just the volume but the velocity, the rate at which the literature is growing has itself accelerated, partly because of preprint culture, partly because of publication pressure in academia, and partly because AI-assisted writing tools have made it easier to produce papers faster.
The result is a strange situation: we have more scientific knowledge available than at any point in human history, and researchers feel less confident than ever that they have a complete picture of it.
One of the oncologists we spoke to early on put it this way: "I know that somewhere in the literature there is probably a paper that is directly relevant to what I am doing. I just have no way of finding it in a reasonable amount of time. So I proceed without it." She said this matter-of-factly, not as a complaint. It had simply become the normal condition of doing research.
The solutions people reach for, and why they fall short
The standard responses to information overload are journal clubs, systematic reviews, meta-analyses, and clinical guidelines. All of these are valuable. All of them are also slow.
A Cochrane systematic review takes on average 67 weeks from protocol registration to publication, according to data from the Cochrane Collaboration itself. Clinical guidelines are updated infrequently, sometimes years after the evidence has moved. By the time the formal synthesis machinery catches up with the literature, the literature has moved again.
There is also a coverage problem. Systematic reviews and guidelines concentrate on the questions that someone decided were worth the investment of time and resources to answer formally. The long tail of more specific, more contextual, more institution-specific questions mostly goes unanswered, or gets answered informally through the kind of selective literature reading described above.
What we kept hearing that shaped how we built NousLab
One pattern that emerged repeatedly in those early conversations was the distinction between known unknowns and unknown unknowns. Researchers were reasonably good at finding literature on questions they already knew to ask. What they struggled with, and what caused the most discomfort, was the sense that there were relevant papers they had no way of knowing they were missing.
A pharmacologist might search for drug interactions using a specific set of terms, never knowing that a relevant mechanism had been described in a nephrology paper using completely different language. An oncologist designing a trial might be unaware of a small pilot study from a different country that had already encountered the recruitment problem they were about to face.
This is the problem we built NousLab to address, and it shaped almost every decision we made about how the platform works. The goal was never to help researchers find papers they already knew existed. It was to surface the ones they did not know to look for.
The deeper issue: time as the real constraint
Everything comes back to time. The researchers we spoke to were not incurious or lazy. They were people doing three jobs at once, running experiments, supervising students, writing grants, teaching, sitting on committees, and trying to stay current in a field that was moving faster than anyone could reasonably track.
The information overload problem in biomedical research is not going to be solved by asking researchers to read more or search more cleverly. It requires tools that do a different kind of work, that can move across the literature at a scale no individual can match, and return not raw results but structured, contextualised insight.
We are still early in understanding what those tools can reliably do. There are real limits, and we have run into most of them while building this product. But the direction is right. The researchers we work with now spend less time searching and more time thinking. That shift, small as it might sound, changes the quality of the work they produce.
The literature will keep growing. The question is whether the tools researchers use to navigate it grow with it, or whether we keep asking human beings to do a job that has quietly outgrown human capacity.
