One in five articles published in journals may contain faked data

Independent investigators who evaluate scientific fraud mostly analyze the content of papers and look, for example, for manipulated images and implausible genetic sequences.

The German researchers approached this task differently:  identifying simple “red flag” indicators that do not require detailed examination of the paper itself, such as the use of private rather than institutional email addresses.

Through this approach, they found that the number of red flag publications across biomedicine rose from 16 percent in 2010 to 28 percent in 2020. Sabel estimated that around 300,000 papers a year were fake.

The researchers further investigated the techniques used by a sector whose annual revenues were estimated at $3bn-$4bn. 

Sophisticated AI

“They typically appear to use sophisticated AI-supported text generation, data and statistical manipulation and fabrication technologies, image and text pirating,” they said.

“It will be a race between the paper mills and those of us who try to detect them, with both sides using AI,” told the Financial Times Professor Gerd Gigerenzer of the University of Potsdam, a psychologist and co-author of the paper.

Ultimately, Gigerenzer explained the best solution was to reduce the pressure to publish, particularly in China. 

The paper has been posted as a preprint on MedRxiv.

Study abstract:

Integrity of academic publishing is increasingly undermined by fake science publications massively produced by commercial “editing services” (so-called “paper mills”). They use AI-supported, automated production techniques at scale and sell fake publications to students, scientists, and physicians under pressure to advance their careers. Because the scale of fake publications in biomedicine is unknown, we developed a simple method to red-flag them and estimate their number.