Scientific fraud is rising, and automated systems won’t stop it. We need research detectives

The latest idea among academic publishers is to use automated tools to screen all papers submitted to scientific journals for telltale signs.

Fraud in science is alarmingly common. Sometimes researchers lie about results and invent data to win funding and prestige. Other times, researchers might pay to stage and publish entirely bogus studies to win an undeserved pay rise – fuelling a “paper mill” industry worth an estimated €1 billion a year.

Some of this rubbish can be easily spotted by peer reviewers, but the peer review system has become badly stretched by ever-rising paper numbers. And there’s a new threat, as more sophisticated AI is able to generate plausible scientific data.

Also Read | Scientist who reported room-temperature superconductivity faces more controversy 

The latest idea among academic publishers is to use automated tools to screen all papers submitted to scientific journals for telltale signs. However, some of these tools are easy to fool.

I am part of a group of multidisciplinary scientists working to tackle research fraud and poor practice using metascience or the “science of science”. Ours is a new field, but we already have our own society and our members have worked with funders and publishers to investigate improvements to research practice.

The problems with automated screening are highlighted by a new screening tool publicised last month. The tool suggested around one in three neuroscience papers might be fraudulent.

However, this tool detects suspected fraud simply by flagging authors with a non-institutional email (such as gmail.com) and with a hospital affiliation. While this could catch some fraud, it will also flag many honest researchers, and the tool flagged a whopping 44% of genuine papers as potentially fake.