New Delhi: Generative AI has become part of academic research faster than researchers have worked out how openly to use it. A new international study of 1,138 communication scholars across 77 countries found that 69.2 per cent use generative AI for at least one research-related task. But many remain concerned about its risks; some worry that admitting to using it could carry a professional stigma.
The researchers, from Taiwan’s National Yang Ming Chiao Tung University, the University of Amsterdam, and Utrecht University, refer to this stigma as “AI-shame” — a fear that disclosing AI use could affect how a scholar’s work is received or damage their professional reputation.
Published in the journal Information, Communication & Society on 25 September, the study shows how communication scholars are using generative AI across the research process, from writing and proofreading to literature reviews, coding, data visualisation, and developing research ideas.
Its findings point to a contradiction. AI is already embedded in researchers’ workflows, but little agreement exists on acceptable use, disclosure levels, or the support researchers should receive from their institutions.
The study found that universities, journals, and researchers themselves often have different expectations about AI, leaving scholars to navigate the boundaries largely on their own. The authors of the study argue that this makes transparency particularly important: without knowing when and how researchers are using AI, it becomes harder for the academic community to develop practical rules for using it responsibly.
Security, privacy, biases
The spread of AI has changed who can access computational tools. Earlier forms of AI-assisted research often required programming skills. But generative AI allows researchers to describe what they want in ordinary language and receive code, text or other outputs in return.
“Recent advances in generative AI have extended the scope of application and accessibility of these methods. Crucially, genAI systems operate through natural-language prompts, allowing users without technical expertise to generate complex outputs. This has democratised the use of AI systems, allowing a wider range of researchers to use them throughout all stages of the research process, such as for automated annotation, creating experiment stimuli, or writing,” read the study.
Researchers in the focus groups described using AI to take on repetitive or labour-intensive work. Participants also described using AI for tasks requiring more creativity, including brainstorming titles or developing research material.
The concerns were not limited to academic integrity.
Researchers spoke about AI systems producing inaccurate information and failing to follow detailed instructions. Some worried about what happens when sensitive information is uploaded to commercial AI systems.
“Very worried about security,” said one participant. They added that they used paid versions of AI tools because they offered “some sort of privacy.”
Participants also raised questions about the electricity and water required to run large AI systems. Some researchers said they were trying to limit their use because they were unsure about its environmental footprint.
They also found that the question of AI use is shaped by language and geography. Researchers who do not work primarily in English may have particular reasons for using AI-based language tools.
“However, due to the nature of the model development process, where most training data were taken from American and European sources, using generative AI tools could inadvertently introduce cultural or linguistic biases into their work,” the study read.
Also read: US could face shortage of nearly 28,000 surgeons by 2038, study finds
A transparency problem
Universities have responded to generative AI with policies, guidelines, and informational resources, but researchers say institutional support does not always meet their technical, legal or methodological needs.
That leaves individual researchers trying to work out their own rules.
One participant described this as “delegation”. Giving an entire task to AI is different from using AI as a tool.
Participants preferred the second approach, where humans stay involved and check the AI’s output.
But there is another pressure as well. Researchers who avoid AI altogether may feel they are falling behind colleagues who use it to save time. Those who use it may worry that admitting it could damage how their work is viewed by journals or other scholars.
This creates a transparency problem. Scholars may believe they should disclose AI use while simultaneously fearing the consequences of doing so.
The authors argue that the focus should be on transparency about how AI is used in research. Attention should be given to what the researcher used it for, how much control they retained, whether the output was checked properly and whether the process of using AI is transparent.
“In terms of best practice guidelines, participants emphasized a preference for guidance over strict rules, favoring best practice approaches that provide more flexible frameworks rather than static rules. One participant suggested that universities should establish guidelines ‘on how to utilise this tool smartly to make your work more effective and also ethical,’ supported by a committee at the university level,” said the study.
(Edited by Aamaan Alam Khan)