One troubling side effect of the rise of generative AI is its increased use in academic research and education.
For educators, it is often difficult to spot and narrow down. However, a recent paper by Ph.D. students has highlighted useful methodology for recognizing AI-generated text and emphasised the potential dangers such an issue presents.
Published by Cornell University in March 2024, the paper Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews by Liang et al. “present[s] an approach for estimating the fraction of text in a large corpus which is likely to be substantially modified or produced by a large language model (LLM)” (Liang et al., 2024).
Academia forms the foundation of society’s higher understanding of its past, present, and future. It is supposed to be a rigorous enterprise conducted by experts and professionals, not something that generative AI has any part in. It is essential that this reputation is maintained and that academic institutions provide and be given the necessary support to prevent the spread of generative AI in academia.
The findings
The paper highlights several interesting findings about the potential overuse of generative AI tools in academia. Among the main findings, the most noteworthy are the ‘deadline effect’, the ‘reference effect’, the ‘homogenization correlation’, and the ‘low confidence correlation’.
- The ‘deadline effect’ suggests that “[e]stimated ChatGPT usage in reviews spikes significantly within 3 days of review deadlines.”
- The ‘reference effect’ shows that “[r]eviews containing scholarly citations are less likely to be AI modified or generated than those lacking such citations.”
- The ‘homogenization correlation’ finds that “[h]igher estimated AI modifications are correlated with homogenization of review content in text embedding space”, meaning that AI-modified papers display a notable amount of similarities to each other.”
- The ‘low confidence correlation’ proposes that “[l]ow self-reported confidence in reviews are associated with an increase of ChatGPT usage.”
Liang et al. made particular note of certain adjectives that they found to have significantly increased in occurrence, including “commendable”, “innovative”, “meticulous”, “intricate”, “notable”, and “versatile”. The significant increases in the use of these adjectives all, crucially, begin in 2023. This, of course, isn’t an accident; ChatGPT was released in November 2022 and has enjoyed a meteoric rise in popularity since then.
How well do you really know your competitors?
Access the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain competitive edge.
Thank you!
Your download email will arrive shortly
Not ready to buy yet? Download a free sample
We are confident about the unique quality of our Company Profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the below form
By GlobalDataAcademic rigor in context
Even to those of us who don’t assign great value to academia—to the peer review process and universities generally—the trends highlighted in the paper should cause some alarm. Though the paper only looked at papers submitted to AI conferences for review, we can assume that the findings would be similar in other academic areas. This is concerning when put into context.
In the UK, the political and financial situation that universities find themselves in is dire. The funding model that incentivises higher intakes of students from overseas is at odds with the government’s headline policy of reducing immigration. As all other methods have failed, the government’s move to no longer allow those on student visas to get visas for dependents has contributed to a 44% fall in overseas student enrolments in 2023. Meanwhile, of course, the government asserts that immigration figures remain intolerably high.
This builds upon longstanding financial issues plaguing many universities. Strikes over pay, working conditions, and pensions have been ongoing for several years, exacerbated by the reduction in overseas student intake. With staff being laid off, the remaining academic staff are having to work more in conditions that they are increasingly unhappy with, pressuring academics to resort to whatever timesaving measures they can, including ChatGPT, to meet deadlines.
All of this lowers the standard of academic output and teaching, impacting the world-leading reputation of UK universities. This, in turn, will prompt the government to reduce support to universities, especially ones offering degrees they think to not be useful (remember the British tabloids’ war on “Mickey Mouse degrees”?).
In this context, it is important not to entirely blame academics for their use of ChatGPT to meet deadlines. While these individuals have indeed sacrificed their professional integrity for the sake of a deadline, the solution is not to berate them or even ban ChatGPT.
The phenomenon observed by Liang et al. can only be resolved by a systematic reinvestment into higher education that supports academic staff, thereby improving student outcomes and society’s collective pursuit of knowledge and understanding.
