Research with integrity – GenAi, paper mills and inclusivity

Climbing one of the hills near the Acropolis on the first evening of the 8^th World Conference on Research Integrity (WCRI), I was treated to a wonderful view of Athens. Having taken the Metro from the airport, I’d only seen Athens at street level, but here it was… the whole city spread out.

This fresh perspective – seeing the big picture – made me think about the opportunities a conference like this offers to do the same. The space it can afford to think about what makes research tick, the different roles and activities supporting good research, and some of the bad practices affecting the quality of science.

It may strike you as unusual that there are conferences on research integrity, but over the years it has become an active area of research in its own right. The field of metascience, or research on research, is rapidly growing, evidenced by many presentations at this year’s conference about how to measure research integrity. How can we know if a new training actually works? What data do we have on equality and diversity that could help us understand whose voices are not being heard in research, academia and publishing? How many publications contain fabricated, falsified or plagiarised data?

These and many other questions were discussed over 4 days in sessions ranging from the philosophical to the highly specific and practical.

Intelligent approaches to Gen AI

Some of the big areas covered at the conference included paper mills and fake clinical trials; the implications of generative AI; and equity, diversity and inclusion.  It’s clear, however, these aspects are deeply interwoven.

For instance, generative AI tools (many of which are now embedded in commercial software such as Photoshop 24) can be used to create fake text and data for paper mills, but might also help screen for fraudulent activity, support researchers writing in English as a second language, and require good training and education for ethical use.

Much of the conversation around gen AI is focussed on creating guidelines that are flexible and values-based; the field is moving so fast that making them too specific risks guidance going out of date. Addressing gen AI through the lens of existing research integrity structures however makes for more generalisable support, as discussed previously.

Fake it ‘till you break it…

Paper mills were a recurring theme across many different topic areas. I came away with a strong sense that we need to raise awareness amongst researchers – there is a real risk that fake research is polluting the literature.

Paper mills produce fake research publications for profit. Whole networks exist purely to sell authorship online, creating fake data and text, using fake email addresses and creating fake academics – even taking over the whole editorial and peer review process to completely bypass scrutiny. It’s a shocking concept, but  evidence of the scale of the problem keeps accumulating. For instance, over 8000 papers were retracted last year from Hindawi journals, a subsidiary of Wiley, primarily due to paper mill activity.

However, the research integrity community is stepping up to counter this. Sleuths, like Elisabeth Bik and Jana Christopher, work to detect these activities, along with academics developing tools to screen publications, such as Jack Wilkinson from the University of Manchester who presented the INSPECT-SR tool aimed at weeding out fake clinical trials.

Research integrity teams at publishers try to verify authorship, screen papers, obtain raw data, and use tools like iThenticate to spot plagiarism and image alteration. However, those wily paper mills will often submit manuscripts to multiple journals at the same time, and so the Committee on Publication Ethics (COPE) and STM, the trade body for academic publishers, are working on ways for publishers to spot these multiple submissions.

But this is, clearly, not just harmless cat-and-mouse antics. The very real dangers of this fraudulent activity were brought home in a number of sessions. One speaker reported raising concerns about more than 900 articles related to studies about women’s health, and while this has led to 151 retractions and 75 expressions of concern, a large volume of problematic literature persists – and it can take an average of 3 years for journals to address concerns.

The pollution of scientific literature by fake studies leads to miscalculation of the size and importance of whole fields of research, meaning we can come to wrong conclusions on safety and efficacy, even from systematic reviews (the gold standard for evidence-based medicine, influencing medical practice and government decision-making processes), and drive researchers down wasteful and frustrating dead ends.