AI in educational contexts

This is my argumentative essay for the writing assignment in 6VWO. Although I've already written about this topic at length on this blog, I wanted to upload this anyway. I'm proud, and I think it also introduces some fresh perspectives.

Anyway, enjoy!

A little more than two years ago, ChatGPT was released to the general public (Schulman et al., 2022). Since then, the chatbot has skyrocketed in popularity, being particularly popular among students. In the U.S., 26% of teens aged 13 to 17 use it for their schoolwork, as reported by Sidoti et al. (2025). Furthermore, according to a survey conducted among the students and personnel of Utrecht University, 80% of students have tried the chatbot at least once, with 31% of students reporting they use it several times a week (Hesselink, 2024). The introduction and rapid growth of this new technology raises questions regarding its ethics and use in educational contexts. The use of generative AI in high school should be banned entirely, as it is unreliable, untrustworthy and forms a severe threat for the development of cognitive abilities.

Foremost, generative AI is unreliable and biased. Contrary to popular belief, ChatGPT does not search for information on the web for you. The technology underpinning the chatbot is GPT (short for Generative Pretrained Transformer), a large autoregressive language model (LLM). Language models like these function by predicting the likelihood of a word occurring, given the preceding context of its training data and the prompt written by the user (Bender et al., 2021). This is comparable to the autocomplete widely available on iPhone keyboards, but on a much larger scale. This means that inherently, output generated by a language model cannot be trusted to be factually correct, as they are only "haphazardly stitching together sequences of linguistic forms, according to probabilistic information on how they combine." (Bender et al., 2021). In essence, all a language model does is guess the 'correct' answer to a question based on probabilities observed in its training data. This also means chatbots like ChatGPT are unable to properly credit sources. Furthermore, models such as GPT are trained on vast datasets derived from the web. The size of these datasets enables them to fluently construct sentences, but also causes numerous "stereotypical and derogatory associations along gender, race, ethnicity, and disability status" (Bender et al., 2021) to be hardwired into the model, as the models are trained on unfiltered data including hate speech, fake news and other misinformation. Thus, output generated by these models is both unreliable and biased and should therefore not be used in educational contexts.

Moreover, the use of chatbots capable of generating entire high school essays forms a severe threat to the development of cognitive abilities. According to Klimova (2012), writing is vital in the development of a range of cognitive functions, as it teaches students to plan ahead, set objectives and think of structure. It trains critical thinking by analyzing sources for their credibility. It forces you to think clearly and synthesize your thoughts into meaningful writing. Furthermore, writing also practices other important language skills, like listening, reading and speaking (Klimova, 2012). If generative AI were more widely adopted in schools, this would pose a great risk to the acquisition of writing skills, and thereby the development of these crucial cognitive abilities.

One could argue that by utilizing language models on a smaller scale, it could have a profoundly positive effect on the development of writing skills. For example, they could be used as a tool to paraphrase sentences, easily find synonyms or help tidy up the structure of an essay. Certainly, using generative AI as a writing tool, similar to how we already use dictionaries and spell check for writing and calculators for mathematics, could be a great way to implement the use of language models at school. However, this would assume that students have the discipline to use generative AI as a tool, rather than a cheat code. To illustrate this, we could look at a different demographic. According to the survey previously mentioned, around 19% of teachers at Utrecht University utilize ChatGPT for creating teaching materials, and a staggering 7% use it in their grading (Hesselink, 2024). As explained, language models like GPT function by probabilistic principles (Bender et al., 2021) and therefore, by definition, cannot logically reason. This makes them completely unfit for grading students. Furthermore, teaching materials should be of high quality, as students see them as an objective source of truth; biased and unreliable output has no place in that. University teachers should be capable of recognizing the downsides of using advanced text prediction algorithms to grade their students or generate teaching materials, yet they are apparently unable to make the right call. If university teachers choose short-term benefits over quality, we cannot reasonably expect high school students to make a disciplined decision regarding this technology.

In conclusion, language models underpinning chatbots like ChatGPT are based on text prediction algorithms that generate text based on probabilistic principles. This means output generated by generative AI is unreliable, cannot cite sources, and cannot be guaranteed to be factually correct. Encoded misinformation from the web makes these models biased. Furthermore, using language models as a cheat code poses a severe threat to the development of cognitive abilities. And while the impact of language models on the development of writing skills could be profoundly positive when utilized as a tool, we cannot expect students to make that disciplined decision. Therefore, the use of generative AI in educational settings such as high school should be completely banned.

Reference list

Bender, E. M., Gebru, T. McMillan-Major, A. & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be to big? In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency (pp. 610-623). https://doi.org/10.1145/3442188.3445922

Hesselink, I. (2024, April 15). Nearly four in ten students use ChatGPT often. DUB. https://dub.uu.nl/en/news/nearly-four-ten-students-use-chatgpt-often

Klimova, B. F. (2021). The importance of writing. Paripex-Indian Journal of Research, 2(1), 9-11. https://doi.org/10.15373/22501991/JAN2013/4

Schulman, J., Zoph, B., Kim, C., Hilton, J., Menick, J., Weng, J., ... Hesse, C. (2022, November 20).Introducing ChatGPT. OpenAI. https://openai.com/index/chatgpt

Sidoti, O., Park, E. & Gottfried, J. (2025, January 21). About a quarter of U.S. teens have used ChatGPT for schoolwork — double the share in 2023. Pew Research Center. https://www.pewresearch.org/short-reads/2025/01/15/about-a-quarter-of-us-teens-have-used-chatgpt-for-schoolwork-double-the-share-in-2023/