Learn to ‘struggle’ with AI to fight cognitive decline

AI generated

SINGAPORE – After tackling a practice problem, Puah Joon Leng photographs his working and asks artificial intelligence to identify his mistakes. It is a deliberate ritual: he wants to have already wrestled with the problem before the answer arrives.

“When the AI schools me, it will be more impactful,” said the 23-year-old Singapore University of Technology and Design (SUTD) engineering student.

On open-ended assignments, AI acts as a teammate that helps him map out the field, locate research papers, and stress-test his arguments.

Puah feeds AI tools like Perplexity, ChatGPT and Gemini with research papers, GitHub projects, builder blogs and LinkedIn posts to help him update his knowledge in fast-moving fields like AI and robotics. He does this by asking the AI tools to identify what is new or worth paying attention to and whether new research represents a genuine advance or merely hype.

“What I find most valuable is not just having articles summarised but having AI connect the dots.”

However, Puah stresses that the synthesis and final judgments remain entirely his own – yet another deliberate attempt to ensure he retains ownership of his critical thinking.

This willingness to struggle with a problem, rather than turning to AI for easy shortcuts, is a trait researchers and educators now consider increasingly precious. Bypassing this struggle, they warn, could cost students their critical thinking skills.

This concern over the technology’s long-term cognitive impact is the second wave of the AI debate taking hold in the education sector.

OpenAI’s 2022 release of ChatGPT ignited the first AI wave, which quickly triggered a massive arms race. Today’s generative-AI landscape is crowded with powerful alternatives – from Google’s Gemini and Meta’s Llama to models such as Anthropic’s Claude, and DeepSeek.

Immediately following the first AI wave, schools were less focused on the psychological nuances of productive struggle and more concerned with an immediate, existential threat to academic integrity.

The United States’ two largest public school systems, New York City and Los Angeles, banned ChatGPT shortly after its launch. Across the globe, universities from Paris to Bengaluru enacted similar bans, citing threats to academic integrity and plagiarism.

Then, Singapore’s Ministry of Education (MOE) did not issue a blanket ban but instead committed a $1.8 million fund to boost AI skills and literacy among students in 2022. In 2023, MOE launched the EdTech Masterplan 2030 and started to introduce AI-enabled tools on the student learning platform.

Public universities here adopted a similar stance. Students are allowed to use AI tools to complete their assignments, but they cannot flout rules on things like plagiarism.

These rules were put to the test in 2025 when three students from Nanyang Technological University got zero marks for an assignment after they were found to have used Gen AI tools in their work. NTU upheld the zero mark given to one of the students penalised as it was found that the student had 14 instances of false citations or data in her essays, the university said.

The Singapore narrative

Since 2023, Singapore’s education sector has matured significantly, shifting towards a more measured and proactive approach.

For the 33,000 teachers in local public schools, MOE has progressively rolled out more AI-assisted teaching and marking tools. They include Markly, which helps teachers grade essays, and Authoring Copilot, which generates lesson ideas and activities.

MOE is also promoting the use of AI-enabled learning companions for school-going children as young as 10 years old, sparking a heated debate among parents about whether early exposure short-circuits the development of foundational thinking. These AI tools, available on the MOE-backed Student Learning Space, act as digital tutors that provide feedback on writing or maths problems.

Amplifying this shift is Singapore’s Personal Learning Device initiative, which guarantees secondary school students digital access, making AI tools more readily available.

Singapore’s trajectory stands apart from its neighbours’, not just in pace, but also in philosophy. For countries with less developed education infrastructure – where an AI-enabled device might be the only available teacher in a remote province – AI is primarily a tool for basic access.

For Singapore – where every student in the public school system already has a device, a qualified teacher and a curriculum-aligned AI platform – the conversation has moved on. The question is no longer about access, but about consequence: what exactly are students gaining access to, and at what hidden cost?

Venky Shankararaman, who is vice-provost of education at Singapore Management University (SMU), put the risk plainly: “We will not graduate thinkers.”

He describes a growing pattern at SMU where students are losing confidence not because they are failing, but because they have begun to believe AI does everything better than they can.

The danger is not that students are using AI, it is that they are using it in a way that quietly bypasses the cognitive work that produces real understanding.

“Learning happens in that friction,” Shankararaman said.

“When we write something, we write a few pages, look at it, and it is crappy, and then we redo it. That struggle is very important in learning. If you do not go through that struggle and AI produces a nice output straight away, there is danger.”

This phenomenon has a name: cognitive offloading, and it is not merely theoretical.

A 2025 MIT Media Lab study using EEG (electroencephalogram) measurements found that individuals who wrote essays with AI models showed measurably lower brain engagement than those writing independently, and performed worse at recalling what they had just “written” later.

EEG is a test that measures the electrical activity in one’s brain using small sensors attached to the scalp. The MIT study was conducted on 54 individuals aged 18 to 39 while they wrote essays either using their own thinking, a search engine, or AI assistance.

The results revealed that heavy reliance could lead to cognitive offloading, where the brain delegates planning and reasoning to the tool without processing it.

Furthermore, a study of 1,000 high school students in Turkey revealed that while students who used AI tools submitted better homework, they scored 17 per cent lower on supervised examinations with the tools removed.

Researchers found that students relied too heavily on the AI to bypass the “effortful learning” phase, leading to cognitive surrender rather than actual skill building.

An Bo, head of the division of artificial intelligence at NTU, said: “My biggest concern is that we may be raising a generation that never learns to struggle productively, and struggle is where learning actually happens.”

Deliberate use of AI in education is key – such as using it to stress-test a fully formed argument or to accelerate the brainstorming process, said Bo.

“The goal of education was never just to deliver correct answers. It was to build people capable of thinking for themselves,” he said. “AI is a mirror, it reflects how clearly we still understand that distinction. If we hold the line, AI will make education better. If we don’t, we may lose the very thing education was for.”

Designing workflow for productive struggles

Educators are now actively designing learning environments that force students back into that discomfort zone.

Sumbul Khan, a senior lecturer at SUTD, said: “Learning, from my perspective, is never comfortable. You have to go outside of your comfort zone to learn.

“When you encounter new information, and learn when you can apply it and use it – it is that learning and that kind of discomfort that is the productive struggle.”

In her course, students document their entire design journey on a shared digital platform in class.

On this platform, students report every decision, every use of AI, every moment they reject AI’s output in favour of their own judgment. Students indicate whether they used AI as a tool or a teammate, or when AI is not used for a particular task.

“This makes students more responsible about what they are doing,” Khan said.

A study, published in the proceedings for the 2024 Conference of the North American Chapter of the Association for Computational Linguistics, looked at how large language models can help novice educators in remediating maths mistakes.

The study revealed that novice educators, frequently under time pressure, default to giving struggling students the answer rather than nudging them through the difficulty. It is a habit that the researchers noted misses prime learning opportunities.

“These educators were not sure how to nudge students and come up with the right thing to say on the spot. Without guidance on how to effectively foster productive struggle, they frequently defaulted to providing the solution to the students,” researchers said.

Chew Han Ei, an adjunct senior research fellow at the Institute of Policy Studies and a member of the Digital for Life Fund Steering Committee, agreed.

Referring to how a young journalist learns, he said: “It is only after you have gone through a few editing rounds with the editor that you learn – through the hard way – what is needed to produce a good story.”

Digital divide redefined

As the technology becomes ubiquitous, experts warn that the digital divide in Singapore is no longer defined by who has access to a device, but by who has been taught to use these tools critically.

Chew notes that while public schools act as a leveller, the real “supercharging” happens outside the classroom. “Outside of schools, the different social capital and learning environments, family environments kick into place.”

The next big leap, he said, comes with equipping parents, not just teachers, to support children to become discerning AI users.

Ultimately, the long-term success of the system rests on strategic choices made over the next three to five years.

Singapore is at a “genuine inflection point” in education, NTU’s Bo said, as he remains cautiously optimistic about the technology.

While Singapore’s existing initiatives – including the National AI Strategy 2.0 and the EdTech Masterplan – are “genuinely strong” by international standards, current frameworks are not yet complete.

He said Singapore’s success depends on how it hands out the AI tools to students.

“If AI is used as a thinking partner, students get sharper, because AI raises the ceiling of what they can take on,” Bo said. “If AI becomes a way to produce work without real understanding, the same technology hollows out the very skills education is meant to build.”