Transcript: Demystifying AI in Writing
Introduction
Yukta: Hi everyone, welcome back to this week's podcast episode: Demystifying AI in Writing. I am Yukta and I am with Cynthia.
Cynthia: Hi, I’m Cynthia, it’s good to see you guys again.
Yukta: I’m glad that you are here. Before we dive in, I have something to ask you. Have you ever intentionally avoided specific words or punctuation, like emdash, because they sound AI-generated?
Cynthia: Hmm… Yes actually. And it’s kind of funny, because so many people around me do the same thing. It’s like we’ve all started to recognize the “AI voice”...and now we are actively trying not to sound like it. This is kinda strange cuz it means that AI has been unconsciously shaping not just what we write… but how we write.
Yukta: Exactly. And I think that’s where things start to get really interesting….and a little concerning. If we’re already changing our writing to match or avoid AI, then it’s not just a tool anymore. It’s influencing our thinking, our voice, and even our sense of originality. That’s precisely what we want to put into discussion today: writing with generative artificial intelligence in higher education.
Cynthia: I think it is really important to emphasise that there are many factors in this conversation, it isn’t a simple ‘AI is good’ or ‘AI is bad’ conversation. Indeed, there are various perspectives to consider, like that of, students, instructors, or even university administrators. However, for today, we will focus the primary discussion on students’ point of view. OK, are you ready to dive-in?
Yukta: Yes. I am ready!!!
The Cognitive Processes Revolving Around AI
Cynthia: Getting started, I am curious about what your standpoint is on this topic, can you share more about your thoughts?
Yukta: I think what’s really interesting about AI in universities is that it’s not just a tool. It’s actually changing how students think and learn, especially in writing-heavy subjects. Like, writing used to be the process where you figure out what you think… but now AI can kind of do that thinking for you.
Cynthia: From my point of view, there is a general misconception about AI reducing the need for students to think. To some extent, the concern is actually valid. However, the problem here is that we might have ignored one important process: the interaction between humans and AI. In one of the studies in 2023, Scott Graham stated that “the output [of AI] must reflect the input”. For example, for AI to generate a satisfactory result, the requests or the so-called prompts have to be as precise and as detailed as possible. In this case, students do think. It’s just that they think in a different way with a different thought process.
Yukta: Specifically, what kind of thinking do you think writing with AI does actually involve?
Cynthia: That’s a good question. It is quite multi-layered, actually. To roughly simplify this, I would say there are two types of thinking in the process of human-machine engagement: pre-generation and post-generation. The former one happens when students write prompts to ask AI to generate something. When students write prompts, they must first of all clarify what they are trying to accomplish. This involves determining the final objective of the work and understanding how this objective relates to and how it can assist the outcome that AI is about to provide. An important thing that might be undermined is that students have to articulate definitions or relevant perspectives, given that there are different viewpoints and pathways to approach one topic. To know which path to choose is to provide a guideline for AI to follow and keep track of the outcomes. Furthermore, there has to be some sense of anticipation about how AI might interpret the request and accordingly tailor it to our wishes. This requires at least a surface-level understanding of the topic.
Yukta: That makes sense…I see what you mean. There’s already a lot of intentional thinking happening with prompting, even before the AI gives an output. But what happens after the AI generates something? Is there still the same level of thinking involved?
Cynthia: I would say the level of thinking will remain the same or even higher in the post-generation process. After receiving outputs from the machine, students often have to make decisions about choosing the outcomes. This phase requires careful evaluation, reflection, and judgment, given that there are risks of misinterpretation and misinformation from the results provided. To better accomplish this, students again need to know the minimum requirements and possess a decent level of understanding of the task and the topic itself. Scholarship frames this as human-machine collaboration [Yang (2026)], stating that AI is a collaborator or maybe an assistant in the writing process instead of a complete substitute. Therefore, far from what was anticipated, working with AI and using AI in writing does not isolate learners from the thinking process.
Yukta: Yeah, I actually think that’s a really helpful way to break it down into pre- and post-generation thinking. Because it shows that students are still engaging cognitively… but I guess my question is how deep that thinking really goes. There was a qualitative study with first-year students where many said AI helped with things like structuring essays, refining ideas, and even improving clarity. They felt it made them more efficient and sometimes even expanded their thinking. But what stood out to me was that those same students also reported constantly having to verify outputs, which creates a different kind of cognitive burden.
Cynthia: To answer the question about how deep the thinking can go, I would say that it varies depending on the assignments and how the students are graded. The assignments that give credit for thinking and reflection will ultimately encourage students to engage in the cognitive process. To mention your concern about cognitive burden due to constantly having to verify outputs, it is notable that this process is inevitable in the context of obtaining information. No matter what kinds of methods students decide to use: AI, search engines, or books, double-checking is always a need. It’s just that people have the tendency to neglect this process due to ongoing trust in non-AI methods. So, the problem of cognitive burden actually plays out in the same way for any approach. What is different here is that the AI outcomes can play the role of a roadmap for students to follow, upon extending their research. Having said that, isn’t AI also making things more efficient? Students can focus on higher-level ideas instead of struggling with structure.
Yukta: I think that's the intention, but studies in educational psychology show that deep learning happens when students struggle through forming arguments and structuring ideas…and AI eliminates that struggle. One of the biggest concerns here is something called cognitive offloading, basically, when we rely on external tools to reduce mental effort. And that’s not always bad, in everyday life, that’s fine, like using Google Maps instead of memorizing directions. But in educational contexts, especially in writing, effort is actually essential because that's where the learning happens. And when you look at broader data, it becomes even more interesting because around 53% of students are using AI to generate ideas in writing, and about 46% are using it to summarize academic literature. When students use AI to generate ideas, structure essays, or even refine arguments, they’re skipping the hardest and most important part of the process: figuring out what they actually think. So if a large proportion of students are outsourcing those key stages of writing, I think it raises a bigger question about cognitive offloading.
Cynthia: That’s surprising, 53% is actually a huge number. But, I wonder, couldn't AI actually enhance thinking by giving students a starting point, like if a student has many ideas it can help narrow down the scope?
Yukta: It can, especially in the early stages, but the risk is in over-reliance. Some research I read by Dr. Chanpradit highlights that students who depend heavily on AI can lose what’s called cognitive independence, meaning they struggle to construct arguments on their own without assistance, especially if they’ve frequently utilised AI to bypass the hard work of synthesis and argument construction.
AI Literacy and Integration
Yukta: But I think this actually leads into a bigger issue… because even if students are using AI, the question is - Do they actually understand how to use it responsibly?
Cynthia: That’s interesting. What do you mean by that? Understanding in what sense?
Yukta: I came across an interesting statistic while reading over this topic that I honestly found mind-blowing…We have almost 80% of students using AI… but less than 20% actually understand the rules around it. Isn’t that crazy?
Cynthia: So you're trying to say that nearly 80% of students are using AI, but only a quarter of those 80% actually understand it
Yukta: Yeah.. precisely so we’re seeing widespread use… without widespread understanding.
Cynthia: Oh I agree. One side of it stems from course policies and non-uniformed allowance of AI among different courses. Many studies have demonstrated that students are left confused from the contradicting restriction of AI in coursework. Imagine you start the term with one course allowing AI use with minimal limitation, and a moment later you step in another class which fully prohibits using generative AI. These can lead to the problem that learners can not fully understand what their instructors want.
Yukta: So what you’re saying is that students are basically navigating completely different expectations across courses, which makes it really difficult to form a consistent understanding of AI use. And that creates a situation where AI becomes something students use… but don’t really understand. Can we really expect ethical or responsible use of AI… if students were never taught how to engage with it critically in the first place?
Cynthia: Yes, I think that is when we need to talk about AI integration in the educational environment. Or to be more specific, AI literacy. As researchers frame this as a human-centered approach, AI literacy involves understanding what the technology is… how it functions conceptually… its potential capabilities… its inherent limitations and potential pitfalls. Therefore, to refrain from teaching students how to use AI is to limit students’ opportunities to learn how to use it properly. In the long run, learning about AI would provide students with a pathway to actually avoid the majority of the problems that we have raised. Thinking about this, there is a parallel comparison with AI integration. Do you know what it is?
Yukta: Not really, you say it.
Cynthia: It’s sex education.
Yukta: HUH… I would love to see where you're going with this
Cynthia: YEH!!!! For both cases, the ultimate goal overall is to inform decision making. The sooner students get educated about the topic, the more correct decisions they are going to make. At the end of the day, AI literacy is all about acknowledging reality. Students are already navigating the space and potential of AI. Avoidance and restrictions are only gonna worsen the on-going problem of AI usage.
Yukta:... Yeah, I actually love the comparison. You wouldn’t just tell people not to engage with it and expect everything to turn out fine.
Cynthia: And also, One thing I noticed is that world-wide corporations have started to implement AI in the operating process. Different fields, from business, medical to entertaining, are making AI a part of daily use, then why not education, why not writing? What makes writing and education different from the mentioned industry?
Yukta: [Hesitating] I actually think that’s the key issue… it’s not that AI exists, it’s that we haven’t adapted writing and pedagogy around it properly.Because right now, students are being given access to these tools… without being taught how to critically engage with them.
Cynthia: Absolutely agree.
Yukta: hmm…I mean this not just in a technical sense, such as how to prompt or get better answers while writing, but critical AI literacy is also being able to: Question AI outputs; Recognize bias; Understand its limitations. In some cases, studies have even found fabricated references in up to around half of outputs, which is honestly alarming in an academic setting. Because if students don’t have strong foundational knowledge, they might not even realize something is wrong.
Cynthia: Yeh, that makes sense. Now that you bring it up, AI literacy and integration are actually a realistic need. I remembered reading that around 75% of students have actually encountered mistakes in AI outputs.
Yukta: Exactly!
Cynthia: So if students don’t have the critical understanding, that could be a problem
Yukta: and that's what makes it more concerning. The issue isn’t just that AI makes mistakes, as you mentioned, technical issues are always being updated, AI output is getting better and better...its about how it presents information. It often sounds extremely confident, even when it’s wrong.
“Opportunity Cost” of AI Use
Cynthia: Now that we have talked about cognitive processes and AI literacy, there is a merging area in between that we can further discuss. I think we can frame this topic as the “opportunity cost” of AI use in writing.
Yukta: Can you elaborate on that? What do you mean by opportunity cost in this context?
Cynthia: Hmmm, it’s like people argued that AI use can not teach students sufficient skill sets. It’s like an exchange between being efficient with assignments but not truly learning anything during the writing process (cause people said it’s the AI work). To an extent this is why various instructors are hesitating about implementing this in higher education. However, from my standpoint, the efficiency created by using AI can facilitate the knowledge acquisition process or the writing process, to be more precise. That means the exchange, or the so-called opportunity cost, that people said doesn’t exist.
Yukta: That’s an interesting perspective. Don’t you think it creates an illusion of understanding, though? Because the depth of engagement with the knowledge changes…like you might believe that it is assisting you, but it might just be a crutch in disguise.
Cynthia: What exactly do you mean by saying that?
Yukta: What I mean to say is that AI is changing the way we write and engage with literature. For example, if I were to speak from an anti-AI educator’s perspective, introducing AI means we lose what we had before. We can never write and engage with scholarship the same way, with the same depth and level of understanding, because AI might just be hindering it, even though we can’t see it yet. We don’t know the long-term impacts of implementing this technology in education. Therefore, this forms an illusion because we think we understand it fully, but we don’t.
Cynthia: I would agree with you on the part that AI changes the way we write, but we all need to understand that just because something is traditional does not mean that it cannot be changed. There should be space for innovation, and we should open-mindedly welcome it. And the fact that AI influences writing does not mean that it is a bad influence and we can not conclude that AI doesn’t foster learning during the writing process. Similar to the point that I have made about cognitive processing and the thinking process of using AI, students still learn, acquire skills, and articulate their knowledge, just in a different way.
Yukta: That makes sense but if AI is generating ideas, refining arguments, and improving language… then we have to ask: who is actually responsible for the intellectual work? Did you know that over 40% of students are already using AI to complete assignments, and nearly 1 in 5 are even using it in exams.
Cynthia: Wow, these numbers are higher than expected.
Yukta: That introduced the idea of agency gap, which refers to when students lose ownership of their own work. And what’s interesting is that students themselves still describe AI as an efficient collaborator, not a replacement. But I think that the opportunity cost of using AI…and research really backs this up, studies highlight concerns around things like ghostwritten assignments and reduced critical engagement.
Cynthia: Actually, I don’t think it reduces critical engagement. Generally, there is a wide range of effort types in the process of writing. However, let's just focus on the two main types: (one) productive effort that supports learning and two extraneous efforts that consume cognitive resources without deepening understanding. In this case, AI helps eliminate the need for extraneous effort and further promotes productive efforts. Given that, most of the problems of writing occur on a surface level, like sentence formulation, grammatical correctness, and phrasing ideas they already understand. In this case, AI writing helps alleviate these pressure points and enable students to pay attention to deeper analysis, conceptual challenges and rhetorical obstacles.
Yukta: Right, so AI, in an ideal world, is basically focusing on the annoying parts so we can focus on the ‘smart’ bits. But I think the concern is that AI doesn’t always just remove unnecessary effort…sometimes it removes the productive struggle too. Realistically, do you think students are using that extra time to go deeper into their thinking…or are they just submitting assignments earlier and calling it a day?
Cynthia: Yes, this is exactly when we should bring up the idea of AI integration again. Through effective course designs, educators can prevent this from happening, and as a result, preserve the productive effort during AI use in writing.
Yukta: Yeah, I agree, when there is a will, there is a way. Research shows that over 90% of students find AI easy to use and time-saving, and around 70% say it provides instant feedback, which is actually really valuable in learning contexts.
Cynthia: Yeh true, AI can support the broader skills that writing is meant to teach. Upon reflection, in the context of university, AI in writing can be viewed as calculators in math, or legitimate cheat sheets in term tests. That said, it is more of a tool that assists learning.
AI, Biases and Equity
Yukta: There is, however, one issue we haven’t fully tackled yet, and honestly, it’s one of my biggest concerns with AI in writing: do you have guesses as to what it could be?
Cynthia: Listeners, what do you guys think?
Yukta: It is bias. AI is often framed as neutral or objective, but our research shows that’s not really the case.
Cynthia: Can you further elaborate on that?
Yukta: If AI reproduces dominant academic voices, then students who already feel marginalized, for example, multilingual students, first‑gen students, students whose writing doesn’t match that “standard” academic tone, would be further disadvantaged. Researcher Peter Cardon talks about how professional and academic communication already privileges certain generic rhetorical styles. AI might reinforce that digital divide even more.
Cynthia: That risk definitely exists. But I’d argue that AI also makes those biases visible in a way traditional tools don’t. There are many AI literacy frameworks that emphasize understanding AI’s limitations and biases as a core educational goal. When students analyze AI outputs critically, they can actually see which perspectives are missing or over‑represented.
Yukta: That sounds ideal in theory. But do you think students really have the background knowledge to recognize those biases? As mentioned prior, scholars warn that AI often presents information very confidently, even when it’s incomplete or skewed. For students who are still developing disciplinary knowledge, that confidence is misleading.
Cynthia: That’s fair…but that’s not unique to AI itself. Textbooks, academic articles, and even instructors carry biases, too. The difference is that AI gives us a chance to teach bias explicitly. Yang talks about AI as “human‑in‑the‑loop” collaboration, which means the student remains responsible for judgment. Don’t you think that if we teach students to question AI outputs, bias will eventually become a part of the learning conversation?
Yukta: Maybe, but the reality is that many students don’t question AI; they blindly trust it. AI literacy varies widely among students, often depending on varying levels of institutional support. Without clear instruction, students may internalize AI outputs as authoritative, especially when they’re under pressure.
Cynthia: That’s exactly why we should circle back to the conversation of integration. When AI is banned or avoided, students still use it, just without guidance. This is when bias goes unchecked. Teaching students how AI is biased actually empowers them to resist it.
Yukta: I get that, but I’m still scared about how subtle those biases can be. Okay, take, for example, in my experience, AI consistently frames arguments in a consensus‑driven way, students might stop taking intellectual risks and start believing there’s only one acceptable way to “sound” academic. Moreover, AI only widens the "digital divide," where students with better access to premium AI models or high-level AI literacy are better positioned to challenge AI bias, and have an unfair advantage, inequity, over those who might accept it because they simply don’t know any better.
Cynthia: That concern drives the focus towards education. Institutions need to step in and level the playing field. Self‑regulated, human‑centered AI instruction can support students in planning, monitoring, and evaluating AI use. Equity will be reinforced if education succeeds in teaching everyone how to use them responsibly.
Yukta: So from your perspective, the real danger isn’t biased AI, it’s biased AI education?
Cynthia: Exactly. When AI literacy is uneven or absent, bias becomes invisible. When it’s taught, AI becomes a powerful lens for understanding how knowledge and power actually work in academia.
Yukta: That reframes things for me. Instead of asking whether AI is biased, maybe the better question is: are we preparing students to recognize and challenge that bias?
Cynthia: Yes, and if writing is about learning how to engage critically with ideas, then confronting AI bias might be one of the most relevant writing skills students can develop in today’s context.
Potential solution
Yukta: Okay, so we’ve gone deep into the benefits, the risks, the existential crisis of “is this my essay or ChatGPT’s?”… but I feel like we need to land this plane.
Cynthia: Yeah, because right now it feels like we’re stuck in this weird middle ground or grey area…not banning it, not fully embracing it, just… side-eyeing it while secretly using it. So what do you think the solution is?
Yukta: Banning AI doesn’t seem realistic. Scholarship also backs this up; most students actually believe AI will be essential for their future careers, and want formal training, not restrictions. Therefore, banning AI is a big no-no.
Cynthia: One of the many solutions that could be implemented is redesigning assessments:
Yukta: Okay, but let’s be real….as long as grades exist, students will optimize AI utility.
Cynthia: To a certain extent, that is true. Traditional assessments are outdated.
Yukta: Yeah, so going back to our discussion, as you mentioned, just because something has always been done a certain way doesn't necessarily mean it is the best approach
Yukta: If an essay can be generated in 30 seconds…
Cynthia: The problem is the task.
Yukta: Now that we talk about it, research does suggest evolving assessments into two “lanes”:
Lane 1: Controlled, in-class, no-AI environments → testing raw thinking
Lane 2: Open, AI-integrated work → testing how well students use AI critically
Yukta: Precisely: And to be able to successfully implement this, we need to ensure that it is consistent across all disciplines. One really important finding is that one-off workshops don’t work; this has to be something students engage with continuously.
Conclusion
Cynthia: Okay, let’s wrap this up with some key takeaways. There are five main points that we would want to highlight. First of all, AI isn’t just a tool; it is shaping how we think and write. Writing is no longer an expression; it is a collaborative interaction.
Yukta: Second, the biggest issue isn’t AI itself; it’s the lack of AI literacy. Most students are already using it, but very few are actually taught how to question it, recognize bias, or use it responsibly.
Cynthia: Three, it clearly improves efficiency and writing quality, but also introduces risks of inaccuracy and over-reliance
Yukta: Taleway 4, efficiency comes with a trade-off. AI saves time but can also lead to cognitive offloading, where students skip the critical part: forming their own ideas. And most importantly…Writing isn’t just output; it’s thinking. And that’s what we need to protect.
Cynthia: Finally, the future isn’t banning AI, it is redesigning our education systems around it
Yukta: I think what this all comes down to now is a shift in power dynamics between students and educators.
Cynthia: Right now, the conversation around AI feels a bit… asymmetrical. It seems like students are taking more control; they know how to use these tools, how to bypass detection systems, while educators are still trying to catch up. Despite the fact that educators somewhat know how to solve the problem, it's obviously not that effective. Knowing that these three topics we presented exist, and looking at both the good and bad aspects, helps educators in pedagogy know what to improve and what to prevent and systematically solve problems consistent with that.
Yukta: Exactly. So instead of trying to control AI use, maybe the focus should be on understanding it. Because when we look at everything we’ve discussed, the cognitive impact, the efficiency gains, and the ethical concerns, it gives educators a clearer roadmap: what to integrate, what to monitor, and what to redesign
Cynthia: So it’s not about choosing between “AI is good” or “AI is bad”… it’s about finding a balanced, intentional way to use it.
Yukta: Exactly! A way that keeps the benefits…like accessibility, efficiency, and support, but still protects critical thinking, originality, and learning.
Cynthia: So, overall, would you say AI is more harmful than helpful in writing?
Yukta: I wouldn’t say it’s purely harmful…yes, it has its drawbacks, but it also clearly has benefits. But I think in writing-heavy disciplines, there’s a real risk that it shifts writing from a process of thinking to just a product to generate. And if students aren’t going through that thinking process…struggling with ideas, forming arguments, refining their voice, then they’re missing out on a core part of education.
Cynthia: I think what this conversation really shows is that AI in university writing isn’t just a simple yes-or-no issue. It’s nuanced. It has real benefits, but also some pretty significant challenges
Yukta: On that note, thank you so much for listening to this episode.
Cynthia: We really hoped you enjoyed it, and that it got you thinking
Yukta:...and maybe even reflecting on how you use AI in your own writing. And as always, we’ll see you in the next one. Follow for more such interesting conversations that break down the role of GenAI in pedagogy
Yukta & Cynthia: Byeeee!!!!!!!!!