Insider’s View: how generative AI could make scientific publishing fairer, and more competitive

Scientists from around the world are using generative artificial intelligence tools to write papers in English, and it’s already altering the publishing landscape.

A new study from the University of Basel has found that papers by scientists from countries where English is not the primary language have become “measurably” closer to a US benchmark since 2022, when ChatGPT, the world’s most used generative AI tool, launched.

This convergence effect has been strongest in papers from countries linguistically distant to English. While papers from countries such as Saudi Arabia and South Korea suggest a high adoption of AI tools for writing, those from countries that are closer to English linguistically, such as Germany and Sweden, show lower levels. Adoption appears to be lowest in English-speaking countries.

And while the data only covers a few years, it shows the trend is intensifying. This effect is both highly statistically significant and increasing in magnitude: by 0.15% in 2023 and by 0.4% in 2024, relative to the 2022 baseline. The study found no convergence effect before 2022.

The scientists believe this effect could give a leg up to researchers who do not speak English as a first language, which is “a disadvantage, because it’s not only the content [of a paper] that matters but also how you write it down,” said Christian Rutzer, one of the study authors, speaking to Science|Business.

If so, this trend may have a significant impact on the economics of publishing. Borrowing ideas from trade studies, the researchers say levelling the playing field may intensify competition, reducing the success of “weaker” researchers from English-speaking countries and driving “stronger” researchers to produce better papers.

“There were offers that maybe in the past could compete with non-native English speakers’ offers, but because they were better in the language but not in the content,” said Rutzer, suggesting those researchers may now have lower chances of getting their work published.

In the best-case scenario, this competition would not only boost researcher productivity and publication quality but also bring fresh ideas from other parts of the world into the scientific discourse.

Rutzer, an economist, gives labour market studies as an example. A large portion of the literature on the topic deals with the uniquely flexible US labour market, and its tendency to experience huge unemployment in times of crisis, but this is only because the study of economics is so US-centric.

“There are more heterogeneous views from all over the world. Fusing the different experiences and cultures into the scientific system [. . .] could bring up new ideas and have a positive spillover to native language [i.e. English] offers because they may also get new ideas,” said Rutzer.

Capturing the benefit

But even if this boost is possible, it is not a given. Journals, reviewers and policymakers will need to respond correctly to the rise of AI.

The researchers give three recommendations if we are to get the best results out of the use of AI for refining scientific papers. First, instruct scientists to disclose what language assistance tools they used to write a paper, to preserve trust.

Related articles

• Horizon Europe applications are skyrocketing. Is AI to blame?
• The EU’s AI in science plan gets a mixed reception

Second, support equitable access and training on generative AI for under-resourced institutions to create a fairer playing field.

Third, update editorial and reviewer guidance to treat declared language support as independent of scientific merit. Right now, some report reviewers are biased against those that disclose they have used AI tools.

But making rules to get the best from new AI tools, especially for the entire scientific publishing ecosystem, is not an easy task, Rutzer said. He believes the key goal with any new policy should be reaping the benefits AI can bring by levelling the playing field and improving science.

“[Policy] should not be punishing persons who use generative AI tools, because then we would again have the fact that those who need them most because of the linguistic disadvantage, are punished,” he said.