91�鶹������Ʒ����

By Kenneth Y. T. Lim, Ahmed Hazyl Hilmy, and Bryan Kuok Zi Wei

Let us imagine a conversation between two high school students in the city-state of Singapore in the year 2035:

“Whoa, is that GPTutor Pro?” Chen Wei asks, pointing at the PRO icon at the corner of his friend’s tablet screen. “How did you convince your parents to pay the $200 a month subscription?”

“Oh they would never!” Adam exclaims. “My school got some kind of sponsorship deal with Edu AI. Someone from the company is coming down to make a speech next week.”

The two sixteen-year-olds sit across each other in a food court, tablet screens propped on their table next to half-finished bubble teas. 

“Ugh, I’m stuck with GPTutor Basic,” Chen Wei sighed, tapping his own tablet. “It crashes whenever I ask it anything complicated, and the token limit is so low. I need to wait… eight more hours before I can finish my geography assignment.”

He brings up a dense table of weather data. “How am I supposed to analyse this myself? The metrics dashboard tells me my ‘manual analysis efficiency’ is below-standard.”

“That’s rough,” Adam replies. “If it makes you feel better, even GPTutor Pro has usage limits too. That’s why some of my classmates got their parents to upgrade to the Max subscription, it costs $2000 a month but they say it’s totally worth it, because they can basically use it 24/7, no limits on image, video or sound generation either!”

“That’s insane!” yells Chen Wei. “They’ll have such an advantage on the upcoming ‘AI literacy’ exams. Ugh, what are we graded on again?”

“Let me check…” Adam taps his screen. “Prompt optimization and response evaluation.”

“Haha, they expect us to evaluate GPTutor’s responses? Why? It’s basically smarter than us,” replies Chen Wei.

“Well, there was that network outage last month.” Adam says. “I was in class when it happened. My teacher asked us to continue analysing the themes in some short story manually, she said all we needed was pen, paper and our brainpower. Half the class just took a nap.”

“Same here” laughs Chen Wei. “Mr Lazar just said we needed to ‘think for ourselves,’ and went on and on about how he didn’t have AI tutors when he was in school. At first it was funny, but I got jealous when he told us about his learning metrics.”

“Oh what did they use back then?” asks Adam.

“Nothing!” says Chen Wei, taking a sip of his bubble tea. “No metrics at all! No penalties for ‘spending excessive time solving single problems’ or anything.” 

“Well, that was a different world,” Adam frowns. “A slower world, ya? It’s different now. We don’t need to understand things in detail, AI can do that better than us. We just need to learn how to work with AI effectively. That’s what my parents always tell me.”

“Yeah, I guess that makes sense.” Chen Wei looks down at his tablet and sees the notification: 

Reminder: your GPTutor Basic access will refresh in 7 hours 50 minutes. Consider upgrading to Pro ($200/month) for an extra 4 hours of access time per day.

Recent research findings on the impact of AI technologies on learning provide insights into the risks of uncritical AI adoption. Lee et al. (2025) found that generative AI use significantly reduces users’ perceived cognitive effort in tasks requiring critical thinking, which is key to meaningful learning. Zhai and Wibowo’s (2024) systematic review identified several studies which highlight the risk of overreliance on AI systems, and how this can undermine students’ critical thinking and problem-solving skills and reduce their media literacy skills. Mei et al.’s (2025) study on Generative AI use in creative writing found that AI assistance can lead to a homogenization of ideas, potentially limiting diversity of views in educational settings.

Balzan et al.’s (2025) concept of the ‘autonomy budget’ offers a useful framework. Drawing on self-determination theory, they posit that within any human-AI system, there is a finite ‘budget’ of decision-making power. When AI systems are granted more autonomy over decisions, human autonomy decreases proportionally, unless the technology is deliberately designed to enhance rather than diminish human agency. This suggests that the way forward is to carefully consider how educational technologies should be designed to augment rather than supplant human agency.

This approach is supported by UNESCO’s 2024 AI Competency Frameworks for teachers and students. Rather than training efficient AI consumers, these frameworks emphasize the development of critical understanding and creative agency. The student framework in particular outlines 12 competencies across four dimensions (human-centred mindset, ethics of AI, AI techniques and applications, and AI system design) with three progression levels (Understand, Apply, Create) which offer a pathway to a more genuine AI literacy rather than mere AI dependency.

Building on UNESCO’s frameworks and a more equitable distribution of the ‘autonomy budget,’ we can imagine a different future, in which students are creators rather than consumers of technology.

“How’s your community AI project coming along?” Adam asked, taking a sip of his bubble tea.

“Here, let me show you,” Chen Wei grins, swiping his tablet to show lines of code. “My team managed to customise an open-source model to translate old Hokkien-language radio broadcasts. We’re applying for a grant from the National Archives to create a data repository which can preserve this part of our language heritage.”

“Cool!” Adam exclaims. “Most commercial models ignore this kind of ‘economically insignificant’ languages, ya? I’m impressed you and your team managed this!”

“Yeah, we couldn’t have done it without the help of family members who still speak the language. Their input was essential.” Chen Wei replies. “How about your big urban ecology project?”

“So we’ve set up these acoustic sensors in green spaces across the city, right?” Adam brings up a map of Singapore on his tablet, with points scattered across. “We’re using a community-trained AI model which can identify local bird calls to keep track of which birds our sensors detect in the different spaces. It’s part of a citizen science initiative to monitor local biodiversity, see how it changes over time due to climate change and urban development.” 

“Wow, that’s really cool! Could I set up sensors in my school garden? I hear bird calls there every morning, I just have no idea what kind of bird.” Chen Wei laughs.

“Sure, the source code and sensor schematics are available here, I’ll send you the link.” Adam replies.

“Nice, I’ll share this with my school’s AI Tinkering Club,” says Chen Wei. “Oh, check this out. Last week, our teacher showed us how to create our own custom language models. We modified the dataset to introduce biases and see how it affected the outputs. Guess what we did with this one!”

“Oh, it really likes ancient memes, huh?” Adam smiles. “Send me the dataset sometime. I’ll browse it when I’m bored. Oh I just remembered, there’s a community Hackathon this weekend! Have you signed up?”

“The one about coming up with solutions to help the elderly?” says Chen Wei. “Not yet, can I join your team?”

“Yeah, of course!” smiles Adam.

“Nice. I love working on projects like this. It’s so cool to use tech to make something which can help the community,” says Chen Wei.

“Couldn’t agree more,” says Adam. 

Where the first scenario created passive operators dependent on commercial AI systems, this alternative vision nurtures active creators who harness technology for community benefit.

To realize a vision like this for AI in education, it is crucial to move beyond the current chatbot-centric model towards approaches which support metacognition, facilitate deeper engagement, and foster critical thinking while respecting ethical principles (including environmental sustainability.) 

By distributing the “autonomy budget” in ways that empower rather than diminish the cognitive capacities of human learners, we can create learning environments in which technology serves rather than shapes educational goals. This vision of AI in education, in which students are creators rather than consumers, is intended to be a hopeful alternative to current trajectories, an ideal to strive for as we navigate the integration of AI technologies into education systems across the globe.