Study predicts AI tools will boost paper counts but lower research quality
A new modelling paper forecasts that large language models will let scientists publish more papers, but the output will be less refined due to existing pressure for quantity.
Researchers have created a model, shared on the arXiv preprint server, to anticipate how large language models will affect scientific productivity. By breaking the workflow into discovery, required development and discretionary development, and applying optimal-foraging theory, the model treats LLMs as cheap, fast and accurate. Simulations indicate that AI can speed up all phases, enabling scientists to generate manuscripts more quickly.
Yet, because academic rewards prioritize the number of publications, the extra speed is expected to be used for producing more, less-refined papers rather than improving existing ones. Carl Bergstrom of the University of Washington argues that LLMs highlight, not cause, the problem of a quantity-driven publishing culture. The findings have not yet been peer-reviewed.
Why it matters
It warns that AI could amplify existing incentives that favor publishing many low-quality papers, shaping future research standards.
In this story