Stop treating ChatGPT like it knows anything

A critical look at the real threats ChatGPT poses to academic skills education

Jubilation and fear abound in the wake of the “AI fever” kick-started by the release of ChatGPT 3.5 in late November 2022. Both national media and specialist educational publications have been flooded by articles and opinion pieces on generative Large Language Models (LLMs) like ChatGPT, as well as detailed tutorials on using them for education and research. Within this news coverage, two major trends have become apparent. The first manifests in a spate of credulous articles extolling the virtues of a bright and fully AI automated future that will free researchers, educators, students alike from the tedium of processing large volumes of text and using their own brain power to transform these into a coherent arguments. The second trend, often concurrent with the first, promotes a near-apocalyptic vision of the future for education, in which AI will make plagiarism undetectable and render any teaching of academic skills obsolete.

While this media coverage highlights some important areas of concern for education, the implications of ChatGPT’s mass adoption are not as utopian nor dystopian as they are generally portrayed. In reality, while different levels and types of education will be differently affected, (academic skills) education will not be wholly supplanted by AI soon, if ever, for one simple reason: ChatGPT doesn’t “know” anything, but is rather a “stochastic parrot”, a machine built for mimicking existing texts without understanding their meaning (Bender et al., 2021). Of course, because this “stochastic parrot” is a highly skilled one, it poses a threat to academic integrity by facilitating largely undetectable plagiarism. However, as more level-headed experts and practitioners in the field of AI, computational linguistics and academic skills education have been pointing out for several years now, there are far more important and fundamental threats we must face, namely the uncritical replication of false information by students and researchers, and the undermining of fundamental academic skills such as reading comprehension, critical thinking, and composition.

He is a lecturer of English of Academic Purposes and English Linguistics at the University of Groningen, whre he teaches academic skills such as research methodology, writing, presentation and discussion, as well as entry-level constructionist English syntax. He recently co-authored a textbook on this topic: English Sentence Constructions (Verspoor, Kassenberg, Poarch & Keijzer, 2022). Additionally, he serves as the secretary of the Exam Board of the Faculty of Arts and as a member of the Exam Board Expert Team for the BA English Language & Culture, BA European Languages & Cultures, and MA Literary Studies. 

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? Conference on Fairness, Accountability, and Transparency (FAccT ’21)

Tim Kassenberg will join us at our Get-together on Saturday 1st of April in de Stadstuin in Groningen at one o’clock. Members are welcome to attend this lecture as well as the pub quiz & drinks!