The Impact of AI on Scientific Research — Evolutionary or Revolutionary?

Image: Unsplash

AI is no longer simply speeding up science; it is reshaping where scientific attention goes and how research questions are chosen. For institutions navigating this frontier, a central question emerges: are we evolving our research practices—or stepping into a new paradigm altogether?

I grappled with this question in a recent conference speech in Colombia. By the end of this newsletter you’ll know where I came out!

Like much of the rest of humanity, scientific researchers are increasingly using AI to speed every step in the research process — from getting grants to submitting papers. And this goes well beyond LLMs.

But let’s start there. ChatGPT and its competitors are now pretty effective research assistants. They can conduct systematic reviews of past literature in a small fraction of the time we used to take (though we still need to check for hallucinated references!). They can also analyse and prepare documents, dramatically reducing the time spent on the knowledge-management steps in research. Estimates suggest time savings of 30% to 60% across the chain of research activities — with AI even acting as an overall project manager to keep track of tasks and outputs.

When it comes to the experiments themselves, and the analysis of data, AI tools run the gamut. They can simulate experiments using generative adversarial networks, explore potential designs with Design-of-Experiment software, conduct time-series analysis and Bayesian optimisation, and capture and associate findings through knowledge graphs and clustering programs. In some cases — as in Insilico Medicine’s new lab — AI can run the whole show, from target selection to lead-molecule optimisation.

So we can see that AI will reduce both the time and the cost of the journey: from grant application to research question to conducting experiments to writing papers for publication. However, researchers must pay close attention to guidelines for the submission of both grant applications and draft papers. Funding agencies and scientific journals are becoming increasingly specific and stringent on the use they allow of AI in document preparation. To quote the NIH: 

“Applications substantially developed by AI will not be considered original ideas and may be deemed non-compliant…”  The use of AI that results in plagiarism or other misconduct may lead to serious consequences. And "NIH will use AI-detection technology to identify AI-generated content in applications.”  Likewise, the scientific community must be alert to the use of AI for so-called p-hacking: manipulating data artificially to achieve greater levels of statistical significance. This is one of several misapplications of AI that researchers must guard against.

Most fundamentally, I believe that — in addition to speeding the linear process as we know it today — AI can help us behave non-linearly: scouring the literature for knowledge gaps in addressing tough questions, helping us focus our work on filling those gaps, reassessing the overall knowledge landscape to guide us to next steps, and becoming more solution-led rather than piecemeal discovery-led, as is currently the case in the life sciences.

So my conclusion, not surprisingly, is that AI represents revolution, not evolution, in scientific research. This is not to diminish human agency in setting the problems and generating creative hypotheses to pursue, but AI–human partnership can accelerate our progress substantially.

When finalising my presentation, I came across another useful AI capability — checking our work! I input my original draft to Claude, which replied: “Not bad… I give it B+… but to score higher you need to fix the following problems.” One of these was in a small footnote on a slide — exactly the sort of thing a human assistant might overlook!

As always, I welcome your push-back and critique. If your institution, research group, or organisation is navigating these questions, I’d be glad to exchange perspectives — this frontier is too important to explore in silos.