Irrelevant AI-Based Science Is a Big Problem for Google Scholar, Study Shows

AI-powered scientific research is polluting online academic information, according to a worrying report printed in the Harvard Kennedy School Incorrect Feedback.

A group of researchers investigated the number of research articles and the evidence of fiction writing Google Scholaran academic search engine that makes it easy to search for research published historically in a wide range of academic journals.

The team specifically questioned the misuse of pre-trained transforms (or GPTs), a type of large-scale linguistic models (LLM) that have popular applications such as OpenAI’s ChatGPT. These models can quickly interpret textual content and generate answers quickly, in the form of figures, images, and long lines of text.

In this study, the team analyzed a sample of scientific papers found on Google Scholar with indicators using GPT. The selected papers contain one or two words that match each other discussion assistants (usually, chatbots) built by LLMs. The researchers investigated how these suspicious papers were distributed and processed online.

“The risk of what we call ‘evidence theft’ increases significantly when AI-based research becomes widespread in search engines,” said Björn Eksström, researcher at the Swedish School of Library and Information Science, and co-author of the paper, University of Borås. to release. “This can have significant consequences because negative results can be spread across multiple groups and across multiple sectors.”

The way Google Scholar pulls research from the Internet, according to the latest group, does not show papers whose authors have no scientific knowledge or peer reviews; the engine will pull in academic distractions—student papers, reports, transcripts, and more—along with research that’s passed multiple hurdles.

The team found that two-thirds of the papers they studied were at least partially produced using unknown GPTs. Of the papers produced by GPT, the researchers found that 14.5% were about health, 19.5% were about the environment, and 23% were about computers.

“Most of the papers produced by the GPT were found in peer-reviewed journals and working papers, but some included research published in major scientific journals and conference proceedings,” the team wrote.

The researchers described two risks that arise from this development. “First, the proliferation of fictitious ‘studies’ that permeate all areas of research threatens to disrupt the process of scholarly communication and endanger the integrity of the scientific record,” the panel wrote. “The second threat lies in the increasing number of seemingly scientific content that is artificially generated by AI tools and re-edited to be found by publicly available academic search engines, especially Google Scholar.”

Because Google Scholar is not an academic repository, it is easy for people to use it to search for scientific literature. It’s good. Unfortunately, it is difficult to separate the wheat from the chaff when it comes to popular magazines; although the difference between a peer-reviewed study and a working paper can be confusing. After all, AI-generated articles were found in other peer-reviewed and non-peer-reviewed publications, indicating that GPT’s work is damaging the watershed of online education content – not just work that exists outside of the mainstream mainstream.

“If we don’t trust the research we read, we risk making decisions based on bad information,” said Jutta Haider, a researcher at the Swedish School of Library and Information Science, in the same release. “But while this is a question of scientific error, it’s also a question of media and information.”

In recent years, publishers have failed to successfully publish a number of scientific articles that were not valid. In 2021, Springer Nature was forced to return more than 40 papers in Arabian Journal of Geoscienceswhich even the title of the magazine had a variety of topics, including sports, air pollution, and children’s medicine. In addition to being without a title, the articles were poorly written—to the point of being unintelligible—and the sentences often lacked direct meaning.

Artificial intelligence is expanding the story. Last February, publisher Frontiers was hit by flak to publish a paper in his journal A cell and Developmental Biology which included images generated by AI Midjourney software; especially, very much anatomically incorrect images of the reproductive tract and genitalia of rats. Limits he returned the paper a few days after publication.

AI models can be a help to science; systems can note the weak notes from the Roman Empire, find The famous Nazca Linesand revealing hidden information in dinosaur fossils. But the results of AI can be as good or bad as human action.

Peer-reviewed journals—and perhaps those hosted by searchers and researchers of scholarly journals—need safeguards to ensure that technology works for scientific research, not against it.