From 2024 to 2026, the proportion of post-secondary students in the United Kingdom using generative artificial intelligence “in some way in their studies” jumped to a staggering 95 per cent from 66 per cent, according to the UK’s Higher Education Policy Institute (HEPI).
The trend is hardly limited to the U.K. In Canada, the percentages are also rising sharply, from 59 per cent in 2024 to 73 per cent last year, according to a survey by KPMG Canada. What’s more, nearly half of students surveyed said their critical thinking skills have deteriorated since adopting AI.
What was once a fringe tool is now used by the vast majority of students, said Afia Abedinin a recent presentation alongside Soroush Razavi at the Alberta Machine Intelligence Institute (Amii) annual Upper Bound conference last month. Both are machine learning educators with Amii.
“Students are not just using it for generating text; they’re using it for research, for generating ideas and approaching assignments,” says Abedin. “It’s becoming part of students’ workflow.”
Ubiquitous but uncertain
Despite the ubiquity of generative AI — such as OpenAI’s ChatGPT, Google’s Gemini and Microsoft’s Copilot — many students feel universities have been slow to embrace the inevitable, says Abedin, too often framing the tool only in terms of academic misconduct, rather than accepting it as a powerful new research aid that’s here to stay.
“There is a lot of uncertainty among students about whether they are using it appropriately or not. Many worry they are cheating and fear they might be caught if using these AI tools.”
According to the KPMG, 77 per cent of postsecondary students want clear direction on AI literacy, including courses that provide examples of how it can be used responsibly and ethically. Only one university surveyed explicitly mentioned critical thinking in the role AI can play in research.
“This is something we all need to address, because while these new technologies are great, we still want students to have critical thinking, so they’re able to ask questions of AI outputs, verify them and be confident when citing anything from generative AI,” says Abedin.
In Canada, the University of Alberta has begun to integrate its use in the classroom, declaring that it is “committed to the ethical and responsible use of AI.” Last year, the university introduced Gemini and NotebookLM to its institutional Google workspace. The custom versions include the use of “Gems” — sets of prompts users can set permanently in Gemini to guide searches.
A few premade Gems include one for brainstorming, one for breaking down complex topics into simple concepts, and a writing editor for fixing grammar, structure and tone. Users can also create their own Gem — perhaps naming it “fact checker” — with explicit, strict instructions like “find only trusted, peer-reviewed sources” or “consult only sources provided” when translating or summarizing.
The U of A’s approach is part of what’s called a “supervised adoption model,” aimed at encouraging innovation while mitigating risks such as data leaks, academic dishonesty and algorithmic bias.
The university’s Framework for the Responsible Use of AI includes references to transparency, mindfulness, accountability and human oversight. Three years ago, it launched an online introductory course called “Artificial Intelligence Everywhere,” accessible to all U of A undergraduates.
This is something we all need to address, because while these new technologies are great, we still want students to have critical thinking, so they’re able to ask questions of AI outputs, verify them and be confident when citing anything from generative AI.
Afia Abedin
Abedin and Razavi recommend what they call the “walled garden,” a digital research environment mostly protected from misinformation and personal data mining. They point out that developers are also working on tools to support knowledge workers who value thinking.
“That is now being recognized by universities, and they are taking meaningful first steps,” says Abedin. “They are creating resources, learning materials to support their instructors and students so they can adopt AI.”
Razavi advises using AI to isolate and summarize salient passages in sources, always returning to those sources to verify accuracy. Learning to question, detect misinformation, draw insights and connect the dots among patterns of data are the very critical thinking skills best left to humans.
AI skills in demand
Today’s students know that a growing number of employers are demanding proficiency with the technology. Some companies have even introduced “tokenmaxxing,” a productivity tool that measures in syllables how much generative AI employees use. Meta and Spotify value it so highly that they’ve introduced it into performance reviews.
The federal government also announced its new AI strategyin early June. It’s aimed at scaling up the technology’s adoption by businesses — from 12 per cent today to 60 per cent by 2034 — and providing access to free AI literacy training for all Canadians, including one million entry-level post-secondary students. The government also says it will create 90,000 AI-related jobs and work opportunities for young Canadians and another 250,000 jobs through AI adoption by 2031.
For Francerlândio Macena da Cruz, a graduating master’s student in digital humanities, discovering generative AI was a godsend. Shortly after arriving at the U of A from Brazil to begin his degree, he was diagnosed with ADHD, struggling with a form of mental paralysis called “executive dysfunction.” It left him unable to proceed with his academic work.
That was in 2022, around the time ChatGPT was released to the public. It helped him overcome his inertia, suggesting outlines for assignments and helping clarify his ideas.
“It helped me immensely,” he says. “Without it, I couldn’t have done 10 per cent of what I have done.”
Cruz ended up creating an app called NeuroBridge for his master’s capstone project, closely modelled after his own research journey and designed to help neurodivergent students overcome many of the same challenges.
It helped me immensely. Without it, I couldn’t have done 10 per cent of what I have done.
Francerlândio Macena da Cruz
“It’s designed outwards from the needs of neurodiverse learners,” says Kenzie Gordon, Cruz’s project supervisor. “It fits a variety of learning needs and addresses academic integrity concerns in a way that many university accommodations tools fail to do.”
He describes NeuroBridge as “an AI-powered, single-pane-of-glass web application that unifies the entire academic workflow into one neuro-affirming interface,” working as an “active executive scaffold rather than a ghostwriter.”
His prototype creates his own walled garden connected to the U of A Library and its research environment, so the bot wouldn’t have free rein to scour the internet for information. The app would also impose guardrails to prevent cheating.
NeuroBridge would also contain a “transparency ledger” to log human-AI interactions and generate a compliant and transparent “statement of AI use,” says Cruz. And though it is specifically developed for neurodivergent students, the tool could prove useful to anyone, he notes.
Cruz cautions that any use of chatbots must include rigorous fact checking and corroboration. One way to avoid hallucinations and misinformation is to limit a search to research material provided, so the bot doesn’t include questionable content from the internet.
“It works with the documents you feed it, and if you don’t have the materials yet, it will help search for those in your university library.”
Digital dialogue
One of Cruz’s instructors,Dr. Geoffrey Rockwell, a Canada CIFAR AI Chair and Amii Fellow, uses generative AI in the classroom as a “Socratic” tutor. Students are encouraged to create a fictional character withCharacter.AI and engage in conversation with a chatbot using Rockwell’s own tool, calledThe Ethics Professor.
Now, with the rise of chatbots, dialogue’s time has come around again. I suggest that we can make a virtue of the availability of these chattering machines.
Geoffrey Rockwell
“There is a long tradition in philosophy of thinking through difficult topics with dialogue,” Rockwell wrote inThe Conversation. “Dialogue is a paradigm for teaching, inquiry and a genre of writing that can represent enlightened conversation.”
“Now, with the rise of chatbots, dialogue’s time has come around again. I suggest that we can make a virtue of the availability of these chattering machines.”
Cutting through the noise
Using generative AI outside of a closed system can be risky, saysDr. Timothy Caulfield, the U of A’s leading expert on misinformation. A healthy dose of skepticism in any search is essential, and part of using AI well is scoping out the pitfalls.
In arecent study, Caulfield’s team found that while chatbots have been “rapidly adopted across research, education, marketing and medicine … most interactions come from non-experts using chatbots like search engines, including for everyday health and medical queries.”
Half of the answers to queries Caulfield’s team tested contained errors, misinformation or hallucinations, he says.
“For some things, like describing a disease or procedure, it can perform well. But for contested topics often associated with misinformation — unproven treatments, nutrition, emerging wellness trends — it seems to perform particularly poorly.”
Don’t let the algorithms push the information to you. Go find it.
Timothy Caulfield
Caulfield advises seeking out information from trusted sources, such as universities, public health agencies and research institutions that aggregate science in a responsible and transparent manner.
“Don’t let the algorithms push the information to you. Go find it.”
But even that is becoming precarious, he says. Journal editors and peer reviewers now find themselvesoverwhelmed with AI-generated papers that are getting harder to detect all the time, and not enough peer reviewers to separate fact from bunk.
“To be honest, I do think it is going to get more and more difficult to cut through the noise,” says Caulfield. “Expert voices — those who know the literature in an area — are more important now than ever, at the exact moment experts are being demonized and questioned.”
It’s clear the rush is on to get our relationship with generative AI right, says Abedin. And it doesn’t end after graduation.
“It’s already shaping how employees are working, how they’re making decisions and analyzing their work. We want students to have that critical thinking, so they’re able to ask questions of AI outputs, verify them, be confident and be that crucial human in the loop.”
Related Stories:
- Leveraging AI to enhance research support
- Six tips for using generative AI
- Leading researcher aims to create a meeting of minds between human and artificial intelligence
