A new theoretical paper argues that AI language models could make scientific research worse, not better, even if the technology worked flawlessly. Researchers from Princeton, the University of Washington, and other institutions built a mathematical model showing that when AI saves time on routine tasks, scientists don't invest that time into deeper analysis - they use it to start more projects, spreading their effort thinner across each one.
The paper, grounded in optimal foraging theory from behavioral ecology, simulates how researchers allocate effort across competing projects. In the model, each research project has two phases: an initial viability check, then a decision to abandon or push forward. The push-forward phase splits into mandatory work like formatting and submission, and voluntary work like extra experiments or deeper analysis. That voluntary work is what gets sacrificed when time becomes more valuable.
"As a labor-augmenting technology, LLMs increase the opportunity cost of our time, impelling us to do more, less well-rather than the same amount, better," the authors write.
Three scenarios, two bad outcomes
The researchers deliberately idealized LLMs for their analysis, treating them as tools that cut time costs without introducing errors and at negligible financial cost. That setup isolates the pure effect of time savings from the technology's known weaknesses.
The model produces three scenarios depending on where AI gets applied. In the first, AI helps evaluate early ideas. Researchers become pickier because starting over is cheaper, so only the most promising projects move forward - but those projects get less thorough treatment, since saved time is better spent launching something new. The authors say this pattern is typical of technical fields.
In the second scenario, AI speeds up publishing by accelerating writing, formatting, and analysis. Because getting a paper out takes less effort, weaker projects become worth pursuing. More papers enter circulation, but each one ends up shallower. This pattern is typical of fieldwork-based disciplines.
Only in the third scenario does AI actually improve quality. Here, it speeds up the voluntary deep-dive phase - extra experiments, more careful analysis. Because AI targets the exact stage where researchers have always cut corners due to time pressure, the time savings translate into more thorough work.
Real-world signs of the dynamic
What the model describes in theory is already showing up in practice. A field report from OpenAI covering eight scientific case studies found up to 60x speedups when rewriting research software, but the bottleneck shifted from coding to validation and long-term maintenance. The perceived time savings don't even have to be real to change behavior: a METR study found experienced open-source developers using AI tools took 19 percent longer to finish tasks, even though they felt 24 percent faster.
The friction is visible in the publication system. In fields where LLMs speed up writing, submissions are climbing fast and straining the already overloaded peer review system. Sakana AI's "AI Scientist-v2" pushed a fully AI-generated paper through an ICLR workshop, citation errors and all. Arxiv responded with tougher penalties, threatening a one-year submission ban for hallucinated sources or AI meta-commentary left in the text.
The paper argues that institutional responses need to be discipline-specific, because AI's effect on research isn't a uniform acceleration. It depends on which phase of the process gets sped up. A recent study on software development describes a similar dynamic as a tragedy of the commons, where individual productivity gains come at the expense of the people who have to review and maintain the output later.
Why this matters for researchers
The practical takeaway: time savings from AI don't automatically translate into better science. If you're a researcher, the tool's value depends entirely on which part of your workflow it accelerates. Speeding up the tedious but mandatory parts - formatting, submission, routine analysis - frees time that the model predicts you'll spend on the next project, not on polishing the current one. If you want AI to improve quality, it needs to target the voluntary deep-dive phase: extra experiments, more careful analysis, deeper validation. That's the only scenario where the technology's time savings flow back into the work itself rather than into a wider, shallower portfolio of projects.
For professionals navigating this shift, the distinction matters when choosing tools and setting expectations. The same LLM that speeds up your writing could be quietly making your research more superficial, depending on where you apply it. Understanding which phase of your work AI accelerates - and consciously protecting time for the deep-dive phase - is the difference between using AI to do more work less well and using it to do the same amount better. Courses focused on AI for Science & Research and AI Research Courses can help researchers identify where AI fits productively into their workflows.
Your membership also unlocks:
