Over the past three years, generative artificial intelligence (AI) has had a profound impact on society. The impact of AI on human writing is particularly significant.
The large language models that power AI tools like ChatGPT are trained on a variety of text data and can now generate unique, complex, high-quality text.
Most importantly, the proliferation of AI tools has led to an overproduction of so-called “AI slop.” This means low-quality output produced by AI with minimal or no human effort.
Much has been said about what AI writing means for education, work, and culture. But what about science? Will AI improve academic writing, or will it just create a “scientific AI lag”?
A new study published in the journal Science by researchers at the University of California, Berkeley and Cornell University says slop is winning.
Generative AI improves academic productivity
Researchers analyzed the abstracts of more than 1 million preprint papers (published papers that have not yet been peer-reviewed) published between 2018 and 2024.
They investigated whether the use of AI was linked to increased academic productivity, improved manuscript quality, and access to more diverse literature.
The number of preprints an author produced was a measure of their productivity, and their final publication in a journal was a measure of the quality of the paper.
The study found that once authors started using AI, the number of preprints produced increased dramatically. Depending on the preprint platform, the total number of articles published by authors per month increased by 36.2% to 59.8% after AI implementation.
This increase was greatest among non-native English speakers, with Asian authors in particular ranging from 43% to 89.3%. For authors from English-speaking institutions with “white” names, the increase was even slower, ranging from 23.7% to 46.2%.
These results suggest that AI is frequently used by non-native speakers to improve their English writing.
How is the quality of the articles?
The study found that articles written using AI use, on average, more complex language than articles written without AI.
However, in the articles written, without it AI, with more complex language, is more likely to be published.
This suggests that more complex and higher quality texts are perceived as having more scientific merit.
However, when it comes to articles written with AI support, this relationship reverses: the more complex the language, the less likely the article will be published. This suggests that the complex AI-generated language was used to hide the poor quality of academic research.
AI increases the diversity of academic sources
The study also investigated differences in article downloads from Google and Microsoft search platforms.
Microsoft's Bing search engine introduced the AI-powered Bing Chat feature in February 2023. This allowed researchers to compare what types of articles were recommended by AI-enhanced search and regular search engines.
Interestingly, Bing users are exposed to a wider variety of sources and more recent publications than Google users. This may be due to a technology used in Bing Chat called search enhancement generation, which combines search results with AI prompts.
In any case, concerns that AI search would become a “dead end” recommending older, widely used sources were unwarranted.
move forward
AI is having a major impact on scientific writing and academic publishing. For many scientists, especially non-native speakers, academic writing has become and continues to be an integral part of the language.
AI is becoming embedded in many applications such as word processors, email apps, and spreadsheets, and it will soon become impossible not to use it, whether we like it or not.
Most importantly for science, AI challenges the use of complex, high-quality language as indicators of academic achievement. Rapid screening and evaluation of articles based on language quality is becoming increasingly unreliable, and better methods are urgently needed.
Critical and detailed evaluation of research methodologies and contributions during peer review is essential, as complex language is increasingly used to mask weak academic contributions.
One approach is to “fight fire with fire” and use AI review tools, such as the one recently published by Andrew Ng of Stanford University. Given the ever-increasing number of submitted manuscripts and the already heavy workload of journal editors, such an approach may be the only viable option.
