20/05/2026
AI Can Write Fluently — But Can It Be Trusted?
Artificial intelligence has transformed academic writing more quickly than most universities anticipated. Students, researchers, and professionals are increasingly using generative AI tools to summarise articles, structure arguments, improve grammar, generate ideas, and even draft entire sections of academic work.
The productivity gains are undeniable. Yet beneath the fluency and sophistication of AI-generated text lies a growing concern that higher education can no longer ignore: epistemic reliability.
Large language models such as ChatGPT do not “know” information in the human sense. They do not reason through evidence, verify sources, or assess truth claims. Instead, these systems generate responses by predicting the most statistically likely sequence of words, based on patterns learned from massive datasets. As Bender et al. (2021) argue, these models rely on probabilistic language generation rather than genuine understanding.
This distinction is critically important in academic contexts. One of the most concerning phenomena associated with generative AI is the emergence of what researchers call “hallucinations”, outputs that appear convincing and authoritative yet contain fabricated references, inaccurate claims, distorted summaries, or entirely false information. Ji et al. (2023) identify hallucinations as one of the central epistemic risks of natural language generation systems because the fluency of the language often conceals the unreliability of the content.
In practice, this means a student may receive a perfectly written paragraph with references that do not exist. A researcher may encounter a plausible explanation that subtly misrepresents a theory. An academic article may contain polished but factually incorrect statements that are difficult to detect without careful verification.
The challenge is therefore no longer simply about plagiarism or academic misconduct. It concerns the evolving nature of authorship and responsibility in an AI-mediated knowledge environment. Traditionally, academic writing required the author to gather evidence, synthesise information, evaluate competing perspectives, and construct arguments grounded in verified sources. AI is changing this process fundamentally. The writer increasingly becomes an evaluator of machine-generated knowledge rather than the sole producer of text. In effect, the academic writer must now act as both author and epistemic gatekeeper.
This shift places far greater emphasis on evaluative judgement, critical thinking, and source verification. The ability to question outputs, validate evidence, detect inconsistencies, and critically assess AI-generated claims may become more important than the mechanical production of text.
Importantly, this does not mean AI should be rejected from higher education. Generative AI offers significant opportunities for personalised learning, writing support, multilingual assistance, and research productivity. However, integrating it into academic environments requires a parallel strengthening of AI literacy and epistemic awareness.
The future of academic integrity may therefore depend less on preventing AI use and more on teaching students and researchers to interrogate AI-generated knowledge responsibly. In a world where machines can produce fluent language instantly, the true value of higher education may increasingly lie in the human capacity to evaluate what is true.
Refer list
Bender, E.M., Gebru, T., McMillan-Major, A. and Shmitchell, S. (2021) ‘On the dangers of stochastic parrots: Can language models be too big?’, in Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21). New York: ACM, pp. 610–623. https://doi.org/10.1145/3442188.3445922
Farquhar, S., Kossen, J., Kuhn, L., Gal, Y. and Rainforth, T. (2024) ‘Detecting hallucinations in large language models using semantic entropy’, Nature, 630, pp. 625–630. https://doi.org/10.1038/s41586-024-07421-0
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y., Madotto, A. and Fung, P. (2023) ‘Survey of hallucination in natural language generation’, ACM Computing Surveys, 55(12), pp. 1–38. https://doi.org/10.1145/3571730
By Wynand Goosen - Appetd Board Member