If AI can scan thousands of articles, draft literature reviews, analyze datasets, and produce coherent manuscript sections, what exactly is the researcher contributing? That question has been circulating through academic departments for a while now, and most people are still trying to figure out where they stand.
Butson and Spronken-Smith (2024) take it on directly in a paper published in Higher Education Research & Development. Their approach is unusual. They structure the entire article as a dialogue between two fictional academics, Russell and Rachel, who disagree about how far AI should go in scholarly work. Russell sees AI as a cognitive amplifier that frees researchers from procedural labor. Rachel worries it’s eroding the intellectual identity of scholarship itself. Neither wins, and that’s deliberate.
I like the format. The question genuinely resists simple answers, and a dialogue holds space for the kind of ambiguity that a standard academic argument would try to resolve too fast. The authors frame the stakes clearly: “AI doesn’t just speed up the practice of doing research; it changes how we think about research problems, how we look at data, and even what we consider to be knowledge” (p. 565). That’s a much larger claim than “AI helps with lit reviews,” and it deserves the kind of sustained attention this paper tries to give it.
Russell’s Position and Its Blind Spot
Russell argues that if algorithms handle the scanning, pattern detection, and summarizing, researchers can redirect their energy toward what humans do best: synthesis, conceptual framing, original interpretation. AI takes on the procedural work. The scholar focuses on meaning.
There’s real truth in that. I use AI in my own research workflow, and it genuinely helps when I’m trying to identify connections across a large body of literature or test early-stage ideas against existing findings. AI can function as a useful thinking partner during that messy initial phase of writing, the stage where you’re still figuring out what you actually want to argue.
But Russell’s framing leans too hard on the efficiency argument. AI as time-saver. AI as productivity multiplier. I covered Ranganathan and Ye’s (2026) Harvard Business Review piece on whether AI actually increases productivity, and one of their key findings was that AI often generates more work, not less. The outputs need checking, the summaries need verifying, the drafts need rewriting. Running an AI-assisted literature review takes minutes. Making sure the results are accurate, complete, and properly contextualized can take longer than doing the review yourself. Efficiency only works if the quality holds up, and in research, that’s a big if.
Why Rachel’s Side of the Argument Matters
Rachel’s position resonated with me more than I expected it to. She accepts that AI can assist with searching and summarizing. But she draws a firm line at co-authoring thought. Academic writing, in her account, expresses intellectual identity. The work of putting ideas into language, restructuring an argument until it actually holds, revising a paragraph six times because the logic isn’t quite right, that process is how scholars develop and deepen their understanding. Hand too much of it to a machine and something essential gets lost.
I think she’s right to worry, and the research on student writing supports her instinct. Fan et al. (2025) found that AI improved essay quality but triggered metacognitive laziness. Students stopped monitoring their own reasoning because ChatGPT was handling the revision work for them. The essays got better bu the thinking didn’t. Kosmyna et al. (2025) found reduced neural engagement during AI-assisted writing tasks at MIT. Students who thought independently before turning to AI produced stronger work than those who started with the tool.
Now, these studies looked at students. But there’s no obvious reason why researchers would be immune to the same effects. The cognitive effort involved in writing is part of what makes the writing worth reading. If a scholar skips that effort and cleans up an AI draft, the output might look polished. The depth behind it may be thinner than it appears.
Rachel also flags peer review, and I wish the paper had spent more time there. Peer review is already strained. If AI accelerates manuscript production, the volume problem gets worse. And if reviewers start using AI to evaluate submissions, we land in a loop where machines generate text that machines assess. That’s not a thought experiment anymore. It’s already happening in parts of the publishing ecosystem, and the implications for research quality and trust in published findings deserve more attention than a paragraph.
Where the Paper Loses Momentum
The dialogue format is the paper’s greatest strength and also its most obvious limitation. It keeps the conversation open and exploratory. But by the time Russell and Rachel move through their fourth domain of research practice, literature review, writing, peer review, original thought, the structure starts to feel repetitive. Each section follows the same arc: AI offers speed and scale, humans contribute judgment and accountability. True, but after four rounds, I wanted the authors to commit to something more concrete. What does responsible AI use in a literature review actually look like? When does AI involvement cross from assistance into co-authorship? The paper raises these questions without ever quite answering them.
The authors seem aware of this. They write: “The pressing question now is not whether to integrate AI, but how to do it in a way that aligns with our core academic values and ethical commitments” (p. 574). I’ve been reading some version of that sentence in nearly every paper on AI and education for the past two years. At some point, the field needs to move from naming the question to proposing answers.
Some researchers have already done that. Cleland et al. (2025) developed concrete guidelines on AI disclosure in academic publishing through their AMEE guide. Eaton’s (2023) postplagiarism framework shifted the integrity conversation from detection to ethics of responsibility. Butson and Spronken-Smith’s paper occupies an earlier position in that trajectory. That was reasonable in 2024. The conversation has moved since then, and the next contributions need to offer frameworks, not just framing.
One observation in the paper deserves separate attention. The authors note that “the current state of AI algorithms, primarily designed for pattern recognition, are not equipped to discern causality” (p. 566). That’s an important limitation and one that gets glossed over in a lot of the enthusiasm around AI-assisted research.
AI identifies statistical associations across massive datasets. It finds correlations. But it doesn’t know why those correlations exist, and it can’t evaluate whether a pattern reflects a genuine causal mechanism or a coincidence amplified by sample size. Kalantzis and Cope (2025) made a parallel argument in their paper on AI and literacy. Human meaning-making is grammatical and purposeful. AI text production is statistical and associative. The machine generates language that looks meaningful without understanding any of it.
In a research context, that gap carries real weight. An AI can summarize 500 papers. It can’t tell you which findings are methodologically sound, which are outliers worth investigating, or which contradict each other in ways that reshape your argument. That judgment still belongs to the researcher. And it always will.
Reference
- Butson, R., & Spronken-Smith, R. (2024). AI and its implications for research in higher education: A critical dialogue. Higher Education Research & Development, 43(3), 563-577. https://doi.org/10.1080/07294360.2023.2280200
- Cleland, J., Driessen, E., Masters, K., Lingard, L., & Maggio, L. A. (2025). When and how to disclose AI use in academic publishing: AMEE Guide No. 192. Medical Teacher. https://doi.org/10.1080/0142159X.2025.2607513
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489–530. https://doi.org/10.1111/bjet.13544
- Kalantzis, M., & Cope, B. (2025). Literacy in the time of artificial intelligence. Reading Research Quarterly, 60, e591. https://doi.org/10.1002/rrq.591
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing tasks. MIT Media Lab. https://www.media.mit.edu/publications/your-brain-on-chatgpt/
- Ranganathan, A., & Ye, X. M. (2026, February 9). AI doesn’t reduce work—it intensifies it. Harvard Business Review. https://hbr.org/2026/02/ai-doesnt-reduce-work-it-intensifies-it