I know that students have pretty strong feelings about artificial intelligence.
“FUCK AI”. “Just admit you can’t think for yourself, you smooth-brained twat”. “Never, AI should be destroyed [...] AI should be banned, all information about how it can be built should be destroyed, all of the heads and engineers for AI companies should get life in prison for, among other things, copyright and IP theft, and the incalculable social harms caused by AI. People who outsource their thinking to machines should be sent to inpatient psychiatric facilities until they are judged capable to return to society.”
Those were all responses Critic received in the 2026 Critic Census this year. Despite those big feelings, 74.4% of students who filled out the Census actually do use artificial intelligence (AI), which was an increase from 65.6% in 2024. So, no matter if you use it or not, artificial intelligence is here. It sits next to you at the library, behind your lecture slides and within your essays. AI exists, it’s impressive, and it’s a bit fucking scary.
When things are scary, I think it’s natural we turn to some kind of flight, fight, or freeze. Staring down the barrel of innovation is terrifying, and I sometimes think that AI might be the bullet to end it all. You could run from it, you could fight it, or you could just stare. But none of those three options will make AI go away, so I guess the next move is to figure out where AI fits. If there is one way to conquer a crippling fear of the unknown, it is to learn about it.
Unfortunately, there are so many controversies and issues surrounding AI that I think it’s become incredibly difficult to work out the actual facts. I’m stressed, so to help me out with breaking all of this down, I’ve called in some backup. But instead of asking ChatGPT for any of this information, I just brought along some water bottles to the interviews I conducted with some very informed people and went crazy.
What Actually Is AI?
I’ll choose a definition that’s directly relevant to students. The University of Otago’s AI Governance Policy describes artificial intelligence as “[a]ny technology or tool that uses data to make inferences and generate outputs such as predictions, recommendations, decisions, or content (text, images, video, code) with a degree of autonomy.”
Basically, the most accessible forms of AI (such as ChatGPT), are large language models (LLMs), a form of generative AI. These get trained on a whole lot of text, which they break down into small words (called tokens). The models then analyse the patterns present in these tokens, and when prompted, can provide a response that's just the most statistically likely pattern of tokens.
In order to do this, AI needs a fuckton of this training data. A model’s effectiveness depends on the quality and breadth of the training data it feeds on. This concept is pretty intuitive – someone who picks up a Critic every week would be better at guessing our content for next week than someone who only just picked an issue up for the first time.
There are, of course, other forms of AI, many of which were commonplace before ChatGPT was around – such as more traditional and analytic AI tools, like Grammarly’s writing assistant. We will focus on the more mainstream generative AI, that seems to have really exploded in just a few years.
Academics
AI in academics is probably one of the most topical applications of AI for students. There are a tonne of established problems with using AI in this area, and even more grey areas. As such, we approached Dr Gleb Tsipursky, a behavioural scientist and author of the peer-reviewed book, The Psychology of AI Adoption at Work: From Resistance to Results, published by Georgetown University Press. His work and commentary have appeared regularly in newspapers including The New York Times, the Toronto Star, the New York Daily News, and The Plain Dealer in Cleveland. And now Critic Te Ārohi! He should be extra proud of that one.
Gleb expects AI to become a normal intellectual tool within the future of academia. “But that raises the value of the human capabilities around the tool,” he added. Students will need to get better at framing questions, evaluating evidence, identifying weak assumptions, checking sources, making ethical judgments, and knowing when an answer deserves trust. “Universities should teach both AI-enabled performance and tool-off competence,” Gleb explained. “Someone who can produce better work with AI but cannot detect a confident error is not yet competent in the way an employer, patient, client, or citizen will need them to be.”
The first problem with applying AI competently in academic settings is their tendency to hallucinate – presenting false or misleading information as fact. We’ve all seen examples of this: Reels depicting AI models answering “how many r’s in ‘strawberry?’” wrong, AI offering up citations that don’t exist, or, much more concerningly, real life lawyers presenting real life court filings with fabricated case law.
This is due to the predictive nature of LLMs – if they’re always working off patterns and the most likely response, they’re going to fuck up sometimes. This is undoubtedly a problem, especially when using AI to explain learning concepts without providing sources. LLMs rely on complex math and unfathomable amounts of training data to produce fluent, plausible answers. However, according to Gleb, that process is undertaken “without any built-in guarantee that the claims are true”. In this way, AI is a bit of a black box. It won’t necessarily tell us why it answered the way it did, and that makes it hard to identify and correct its mistakes for both users and programmers. But, even if we did understand the internal mechanics of an AI perfectly, we would still need to verify what it tells us against reliable evidence.
The second problem is that AI is generally really good at what it does. In Adults (the best show ever), a teacher is marking essays and remarks that “they’re all AI. It's just how good the AI is that I’m grading.” That’s real. AI is good, and it’s getting better faster than any AI-checker or plagiarism system can keep up with. At the start of the COVID lockdown in 2020, around when AI first started coming out, everyone could cheat remotely with ease. AI checkers emerged, and they were alright-ish, but AI models got better, faster. Current students watched AI evolve unbelievably quickly, from being fairly obvious and often shit at answering questions or completing assignments, to being pretty much unidentifiable. Here’s where the academic problem lies.
Personally, I haven’t written a single essay this year. That’s pretty unusual coming from someone studying Politics, Philosophy and Economics and Psychology. I’ve seen a lot of different ways lecturers have been trying to deal with this whole AI problem. I’ve had an assignment that encouraged me to prompt an LLM to write a lab report, and then critique it. I’ve had a midterm essay changed to a twenty minute one-on-one chat with my lecturer. Sometimes I feel like we’re all just as confused as each other when it comes to if and how AI can be responsibly applied in academics.
According to the Census, the most common uses for AI include learning concepts or studying (57.8%), planning, writing, and/or editing assignments (39.2%), as a replacement for Google or everyday use (17.2%), or as a friend or for advice (11.8%). If used right, AI is useful as fuck – a personalised tutor with infinite time for questions, a sounding board for essay structures or a way to help find studies to reference.
In Gleb’s opinion, students should start by considering the purpose of the assignment before they use any AI. “If the assignment is meant to build a skill, students should not hand that skill over to AI before they have practiced it themselves,” he explained. Where AI use is permitted in assignment, you could use it to explain a difficult concept in another way, testing an argument, generating questions, suggesting an outline, or giving feedback on a draft. “But students should verify important claims against course material or primary sources, make the final intellectual decisions themselves, and be able to explain and defend what they submit without relying on the tool,” Gleb stressed.
According to Gleb, a simple habit for students to ask themselves is what was the most important contribution AI made, how you checked it, what you changed after checking it and whether you could explain the reasoning behind it if the AI tool disappeared. “That turns verification into part of learning rather than an extra compliance exercise,” Gleb noted.
I think you can identify a clear tension here. On one hand, learning needs to be protected – AI can all too easily be used in place of genuine effort, creativity and critical thinking. On the other hand, AI, when responsibly applied, could help you learn better and more in depth. Against that tension, the University’s official policy for student use is fucking confusing. It essentially outlines a case-by-case situation, noting that different papers, and even different assessments, have different guidelines. Students should check outputs they receive from LLMs, as they’re responsible for any errors or fabrications in generated material. Students should also be aware that “Gen-AIs have different capacities and limitations” and that responses may be biased as the models are trained on a monoculture of knowledge. LLMs also have a tendency to positively reinforce the user, providing skewed answers that are essentially what you want to hear.
The case-by-case thing might feel like a bit of a cop out, but it's more likely due to the simple fact that no one really knows what the answer to AI is. AI technology is evolving so fast that lecturers can’t know what the best thing is. Students are so paranoid about AI that a lot of us don’t even admit using it on AI disclaimers for assignments, even if they did use it completely appropriately. But again, fleeing and hiding from a scary thing won’t make it go away. AI is here to stay. Like with any controversial thing (drugs, alcohol, sex), AI should be regulated for harm reduction. The AI horse has bolted, and now the AI horsie needs credit on your essay.
“I think institutions need much clearer assessment-level rules,” Gleb suggested. “Otago's current guidance sensibly recognises that different papers and assessments may need different approaches, but the variation becomes frustrating when students cannot easily tell what applies.” Critically, Gleb dreams of a world where every assessment had accompanying guidance from lecturers that said whether AI is permitted, limited, or prohibited, what uses fall into that category, and why. “The reason matters, because students need to understand which capability the assessment is trying to develop.”
In terms of the AI paranoia some students might be feeling, Gleb posited that universities should rely less on trying to catch students out for AI use after the fact and more on assessment design. “If independent mastery matters, use some combination of oral defense, in-class work, staged drafts, or follow-up questions that require students to explain their reasoning,” he suggested. If AI use is allowed, Gleb would ensure that disclosure is straightforward and psychologically safe. “Students who believe any admission of AI use will make them look suspicious have an incentive to hide even legitimate use, which makes responsible-use policies harder to enforce.”
The University also has an AI Governance Policy, which lays out their own commitments as an institution. In short, these include a lot of references to ethical use, especially in terms of upholding Te Tiriti o Waitangi obligations, and sustainability targets. The University “acknowledges the inherent risks associated with AI” and commits to managing these, as well as ensuring that their policies evolve as the technology advances. Furthermore, while the University recognises that AI technologies present “significant opportunities to advance its mission, enhance operational excellence, and support innovation in research, teaching, and service delivery”, AI systems must be used to augment human capability, with “human judgment and oversight remaining central”. The full policy can be found on the University website.
Overall, Gleb wants students to aim for a future of “dual competence”: students understand how to use AI to extend their capabilities and can still “understand the underlying domain well enough to challenge it”. For Gleb, that latter point is especially important. “AI can remove some of the routine work through which novices used to build judgment,” he explained. True – using AI will save you from trawling the Internet for hours looking for a source to back up what you think is common knowledge, but that builds character. “Universities will need to redesign that learning deliberately, rather than assume expertise will emerge automatically.”
Employment
At this point, I’m being told once a week that I’m never going to get a grad role, or even a career, because of AI. To console me, I contacted Professor Emeritus James Maclaurin.
James is currently helping the University “think about AI [...] How we're using it and what it means for us.” James trained as a philosopher, and now works in AI policy, which is a pretty awesome pipeline. He is the Director of the Centre for Artificial Intelligence and Public Policy (CAIPP) at the University of Otago – a group of interdisciplinary academics exploring “policy, regulation, ethics and governance associated with the implementation of artificial intelligence.”
According to James, one thing about technologies is that they typically start out expensive, inefficient, and a bit shit. But, as we make more and develop them, people find easier, cheaper, and quicker ways to do so. This is known as Wright’s Law (yes, the airplane guy). For a species that fears what it does not know, this isn’t great for the psyche: Wright’s Law makes it “very difficult to know where [AI progress] is going to land”.
Similarly, James referenced the Jevons Paradox, which emerged from observations of coal use during the Industrial Revolution. As steam engines became more and more coal efficient, people assumed less coal would be used overall. Instead, this increased resource efficiency meant using steam engines became cheaper, so they were used more, and so the overall use of coal actually increased. In short, learning to use a resource more efficiently actually increases consumption. According to James, this applies to people as well. “Iif you make people much more efficient at doing something, then that thing gets cheaper. And then people want more of it.” And so we end up using more of the resource – people.
However, James conceded that things like checkout machines in supermarkets have cost people their jobs. “So, sometimes automation replaces jobs. Sometimes it replaces bits of jobs. When it replaces bits of jobs, it tends to make the rest of the job more valuable because you're producing more. And that can be good for employment.” He emphasised that we don’t know how often this will be this case, or in which domains, so he wouldn’t say that’s a “magic bullet”. I’m pretty sure a magic bullet is something else entirely, but his point stands.
James does agree there are domains where employment “seems to be going in the wrong direction”, where we see employers hiring fewer graduates. He said it’s hard to say at this stage whether this is due to companies choosing not to hire anyone in case they can replace that role with AI within the year, or simply a lack of information about how different types of work are going to be impacted. Each generation has had a big job loss scare – from factory automation to the internet, and now AI. The important thing to remember is that while junior-level employees and grad roles could theoretically be replaced with AI, this fucks an organisation’s talent pipeline. If you have no juniors, then that calls into question who will hold the senior roles in the future.
On top of all this, AI will likely be a job sector that many students will end up employed in, especially if it continues to grow at the rate it currently is. Especially with something like AI that is so applicable to every field, students should look at AI as another key skill that they can learn.
Policy and Regulation
When faced with things that are spiraling wildly out of control, regulation is one way to try and curb the wildfire. However, the toughest part with regulation is how slow policy is developed compared to how fast AI evolves. According to James, by the time a policy has undergone consultations, and passed through ministers and public discussions over the course of a year, “the technology is ten times more powerful [...] than it was at the start of the year.”
AI concentrates power: once your company has a model, you’re in. And you’re fucking powerful. That’s just capitalism in general, though – the system rewards early adopters. Whoever is luckiest at the beginning of Monopoly is most likely to win. These AI companies are the fastest growing in human history, and the global economy over the last three years has been almost entirely built around the potential AI has for economic growth. Implementing policy will decentralise this, as well as slow the rapid advancement of this technology and give us some breathing room to actually develop some understanding.
James compared AI to the Industrial Revolution, explaining how that was a transition of physical power – from human or beastly muscle to steam engines – and now we are living through a transition of intelligence. While the “original computing revolution [...] helped us with our thinking, it didn't change the rate of thought, or the amount that we could think. Not in the way that the combustion engine suddenly changed everything.” The computer was just a rumble in the distance, and AI is the real storm.
It took us a long time for policy to catch up with the huge advancements society made during the Industrial Revolution, but we got there after a hundred years or so with the buzzy invention of things like the weekend and child labour laws. But this AI revolution is moving a whole lot faster already. “And so our policy response has got to be much better,” James insisted. He added that we undoubtedly know a lot more about technology and how it works than we did back then, “so the news isn’t all bad.” We might be a wee bit quicker off the mark this time.
Water
Everyone’s read an article (watched a TikTok) about how bad AI is in terms of water use. According to OpenAI CEO Sam Altman, one ChatGPT search uses 0.32mL (1/15th of a teaspoon) of water. However, researchers that factored in water consumed for electricity generation and onsite computer cooling estimated one query using between 10mL and 50mL. For reference, the tequila shot you bought on Saturday for twelve whole dollars was 30mL. When you think about how many times you’d prompt an LLM in one session, and then how often you have a session, and then multiply that by, say, the 74% of Otago students who use AI (15,000, give or take), let alone by usage across the whole world, those mLs start to be a pretty big number.
But forget about ChatGPT for a second, and think about burgers. According to one fairly conservative study, the water footprint of producing a 150g beef burger patty is 35.5 litres. And, according to the Water Footprint Calculator, it’s about 2300 litres.
Ignore the huge differences between these estimates for a moment, and let's look at the lower estimate for both. Producing one ChatGPT search is around 0.32mL, and one beef patty is around 35,500mL. That’s about 111,000 ChatGPT queries to equal the cost of eating one burger (not including the bun), and over 222,000 queries to equal a post-town Cheesy Pleaser from Maccas. Even using the higher end of the other estimate for ChatGPT water use, it would still take at least 700 prompts to balance the cost of one burger.
Of course, this isn’t quite that clearcut of an issue: these are difficult estimates to make. The huge variation is due to how much of the process you factor into each calculation, and the differences in water. For example, beef uses a lot more recycled “green” water in production, whereas AI uses treated drinking “blue” water. A lot more water is also used in training AI than in prompting it, which makes it difficult to bring these numbers into an individual usage perspective.
Basically, it’s easy to rag on the amount of water AI uses. While it is a lot, it's important to look at it in context of the rest of our daily lives, and consider how we can reduce our impact on the environment in general.
The Arts
Artists are scared. As a writer, I’m scared, and AI is already having some effects on how I write. For example, Critic continues to use grammatically wrong em dashes to avoid AI allegations (which we still get – fuck you guys). In terms of people rejecting AI, art has been one of the most intense examples. But, as James pointed out, this discourse still “isn’t on a scale that's affecting the model makers.”
AI art is already pretty common. From the AI-made flyers in windows of small businesses all down George Street, to an AI-generated poster for DJ Pint Night, people are opting for an easy alternative to coaxing someone to fuck about in Canva. And people are upset about this.
We’ve already talked about how AI is trained on a whole lot of data, and this has led to debates around the world about copyright issues, and what compensations authors should receive if their work is used for AI training. The New Zealand Minister of Commerce and Consumer Affairs is working on a potential copyright framework for generative AI, and is set to report back on this by the end of March in 2027. Again – policy is slow. James raised cases in courts of law around fair use, and the argument that if we get to read books, why shouldn’t AI? He thought that to be especially true if a model can remember and learn from the books better than we can. It’s complicated and confusing, and there are no right answers.
On the other side of the argument though, technology has always influenced art and music, creating new styles and shifting the fields in interesting and exciting ways. On top of that, it’s arguable that AI could actually make art more accessible for people to explore their passions, decreasing barriers associated with things like film making. Plus, tech is still keeping up with AI. Various platforms such as Nightshade allow artists to “poison” the metadata of the art they produce, meaning AI can’t see what human eyes can see. Pretty neat.
Answers (Or Lack Thereof)
As well as all of this, AI has a whole lot of other pretty incredible applications that I haven’t even touched on. There’s AlphaFold, an AI system that can predict the 3D structure of a protein from its amino acid sequence that won the Nobel Prize in Chemistry in 2024, or Anthropic’s mission to use AI to cure cancer. There are also a tonne of other debates, ranging from the psychological impacts and uses of AI, to the data centre being built in Southland. There’s just too much to get into here (though stay tuned for more on the data centre). In short, this technology has real potential for both good and harm.
James admitted that he wished he could offer some better answers to all this. While he himself has signed a few petitions asking that we slow down on development and implementation of AI until we better understand the technology, he added that those petitions “don’t have an effect on industries that are just driven by vast amounts of money.”
James believes that “the answer in the face of radical change is always information, always to know more,” and that this is especially important for people who don’t like AI. He advised that we maintain skills that we don’t want to be replaced by AI – writing, art, coding – whatever it may be for you. And he advised that any young people nervous about AI should use AI. “Because no matter what happens, you're going to be affected by it. And even if that's a negative thing, the more you know, the better you'll be.” Knowledge is power.
If you went to the first OUSA Leaders Debate, you will have seen Penny Simmonds answer questions on AI literacy and data centre regulation with vague comments about how we all watch movies and use internet banking, and we don’t want to go back to the “dark-ages”, or have “our data being processed in the Middle East”. If this doesn’t highlight the complete lack of understanding around AI, I don’t know what does. Fact-check yourself, and try to understand viewpoints outside of your own algorithmic echo chambers.
So I’ll say it again. AI is a dual-use, general purpose technology. James even compared it to electricity, emphasising that it “touches everybody and everything”. Importantly, all this also means you can’t actually just lay out the benefits and risks of AI. That’s something I wish I knew eight drafts ago. Instead, every area has some positive opportunities to use AI, and then it causes issues. We could keep listing both sides of these forever. No one knows where this is going to take us, and so the most important thing for us to do right now is to work to understand it.
