This AI tool flags predatory journals: how researchers built a “firewall for science” to protect research integrity

Researchers around the world are swamped with unsolicited emails promising fast publication in little-known journals.

There is a lot of scandalous activity on the dark web. Scams are everywhere, left, right, and center.

They appear in the unlikeliest of places as well. Even a budding scientist might get approached by a scientific journal and asked to pay a fee to publish their work.

Daniel Acuña, associate professor of computer science at the University of Boulder, Colorado, has revealed the nuisance of “predatory” publishers that charge a fee of hundreds or thousands of dollars to publish a scientist's work in PDF format.

Acuña receives these offers weekly and likens them to “whack-a-mole” in that they pop up one after the other.

Researchers from less privileged countries are also targeted, especially when their academic pedigree is less than that of the institutions of North America or Europe.

Often, the same company registers different names, one after the other, making it hard to track and trace.

Phony offers are hard to detect

The warning signs of predatory publishing were raised in 2009 when CU Denver librarian Jeffrey Beall coined the term.

The Directory of Open Access Journals has worked since 2003 to weed out suspicious titles based on strict standards.

Volunteers look for clear peer review policies, transparent editorial boards, and consistent publishing practices.

Luckily, Acuna and his team have been able to develop an AI tool that aims to highlight dubious sources.

Acting on the warning signs

The AI was fed a sample of 15,000 journals from the Directory of Open Access Journals.

It singled out around 1400 titles that looked fishy, and subsequently, human intervention raised the red flag on 1000 of these.

The model is eagle-eyed in the sense that it can denounce shoddy grammar, exaggerated publication figures, or high levels of self-citation.

“This should be used as a helper to prescreen large numbers of journals,” Acuña said.

“But human professionals should do the final analysis.”

Fragile practice needs firewalls

The researchers hope the tool can be adopted by universities and publishers as what Acuña calls a “firewall for science.”

The timing is apt as human-to-human peer review in science (and other fields) is a dying art, especially with the onset of machine intelligence.

Just weeks earlier, hidden AI prompts were uncovered in papers from leading universities – designed to trick AI-driven reviews into approving them.

Academic publishing is a fragile industry and is calling out for safeguarding measures such as this, though the progress will always be in flux.

“We should probably treat science like software,” Acuña said.

“Expect flaws – but also expect constant fixes.”