The other day, I called one of my college friends, Syed Alauddin Ahmed. He is a Dhaka-based school teacher who relies on private tutoring to support his family.
He was a leftist theorist in his youth. Now a bedridden father of three, Alauddin’s economic condition is worse than his physical condition.
He told me ChatGPT has eaten up a major portion of the private tutoring market in his area. Most students now use “AI” to generate their academic notes, presentations, and assignments with detailed infographics. A few years back, these tasks used to require significant time with a seasoned home tutor.
Large Language Model (LLM) based Artificial Intelligence tools, colloquially called “ChatGPT” (a proprietary eponym), have gained massive access to learners of all ages. Even my own child made a Gemini voice assistant his playmate at just five.
In Bangladesh, there is a need to connect these dots and conduct broader research on how AI genuinely impacts our academic and informal learning needs in depth. However, as a journalist seeking to understand at least the surface-level reality, I sought insights from teachers and students across engineering and social science disciplines. I wanted to see how AI is currently being integrated into informal and academic learning.
A large language model (LLM) is the building block of an AI entity. Think of an LLM as a vast network of words, trained on huge amounts of text to predict the next word in a sequence. By repeating this task across billions of examples, the model absorbs grammar, facts, reasoning patterns, and writing styles. But it does not know things the way a human does.
For brevity, an AI entity can be compared to Ghanada of Premendra Mitra’s stories. Ghanashyam Das, aka Ghanada, knows a lot. He can make an instant story by mixing context with his own knowledge, and it sounds true. But it is the audience who must decide what is fact and what is fiction.
An AI lacks a built-in fact-checker. It generates text from learned patterns, so it can state incorrect information with total confidence. This flaw is known as “hallucination.” Its knowledge is also frozen at the training date, so it cannot know about recent events without external help.
A plain model can only predict text. It cannot check a website, run a calculation, or send an email.
An AI agent is different. It is equipped with tools and memory, which let it take multi-step actions in the real world. The model acts as the “command centre” while the surrounding system provides the “workers.” An agent loop typically follows a repeating cycle: the AI observes the situation, thinks about what to do next, and acts by calling a tool, such as a search engine, calculator, or database. The tool returns a result, and the cycle repeats until the task is done.
When AI gets it wrong
AI can be a useful study companion. But it comes with risk points that any learner should watch for.
Say a student asks AI to help revise for an exam on a topic they only partly remember from class. The AI can produce a smooth, confident summary. But it may quietly add details that were never taught, attribute an idea to the wrong thinker, or invent a statistic that sounds plausible.
Before trusting such a summary, a learner should check six things:
- Unverified figures. Any number or statistic should be checked against the original source or textbook before it is memorised or repeated.
- Vague or invented sourcing. Phrases like “studies show” or “experts say” need a cross-check. AI sometimes uses these phrases without a real source behind them.
- Fabricated content. Watch for added commentary, theories, or claims that were not in the original material.
- Unverifiable claims. Statements presented as settled facts should actually be checked, not assumed correct.
- Missing context. AI may simplify a debate or theory in a way that drops important nuance, giving false confidence in an incomplete understanding.
- Missing original engagement. AI summaries are no substitute for reading the original text at least once. Relying only on the summary can leave gaps a teacher or exam will expose.
Sadman Sakib of Google Taiwan has identified “nonsensical and extremely robotic paragraph generation” as a rising issue in the LLM world. “It is particularly problematic for individuals who use LLMs not only for coding tasks,” he says. “But also heavily to read and understand different topics, (research) papers, and other (complex) materials.”
Sakib mentioned that some models can still be used well for multimodality, pure research and reading, and non-English content. Users should decide after trying several.
Core findings
AI has made both academic and informal learning faster, easier, and more interesting. Anyone with an internet connection and a digital device can learn almost anything. But the debate remains: how much can AI replace human wisdom?
I interviewed a university student, a developer, and two teachers, from engineering and humanities backgrounds. The findings were interestingly ambiguous.
The engineering duo mostly emphasised the productivity, speed, and administrative efficiency they get from AI. The humanities duo, however, expressed doubt over an illusion of mastery, effects on critical thinking, cultural and political bias, and emotional distance.
Both observations are valid. Software engineers work in agile environments. Humanities-based academic study is yet to answer many philosophical and political problems that rarely have a single deterministic answer.
In the pre-AI era, learning technical fundamentals relied on reading books, documentation, Stack Overflow posts, blogs, and YouTube videos. AI has shifted this workflow. It now summarises complex topics into simple, tailored answers within minutes, serving as a learning assistant.
According to Md Najmus Shakib, an Assistant Professor in World University of Bangladesh’s CSE Department, AI is a 24/7 “personal teacher” and “smart assistant” for both students and educators. He said AI functions as an always-available tutor in every student’s hands, making complex topics easy to understand with instant answers. It serves as a reliable companion across disciplines, helping with code debugging, language acquisition, summarising research papers, and step-by-step mathematical problem solving.
Shakib also observed that AI helps expand and develop students’ thinking and cognitive abilities.
For teachers, he said, AI simplifies administrative complexities. This allows educators to dedicate more attention to actual teaching, by handling lesson planning, quiz generation, and evaluation design aligned with Bloom’s Taxonomy. Shakib found that AI allows teachers to tailor lesson plans and curricula to meet the unique learning needs of individual students.
He noted that, when used ethically and correctly, AI is a transformative “game-changer” in education. It serves as a “powerful collaborator to human teachers,” not a replacement.
An interesting observation comes from Gono Bishwabidyalay’s Mohammad Asif Chowdhury, an Assistant Professor in the Department of Politics and Governance. He observed the amplification of Eurocentric and Western bias, and an illusion of mastery with lost realities.
Students of politics and governance deal with different types of theories and political analysis, where differences in culture, religion, and opinion exist. Most of the AI services we use are developed, maintained, and managed by countries with geopolitical dominance.
He observed: “Most of our texts relating to theory and framework are from the West and carry Western values by default. AI does not create entirely new bias but acts as an amplification of existing structural dependency within International Relations and Political Science curricula, which are already Eurocentric.”
When it comes to building academic or informal knowledge using AI tools, he warned that feeding books and media into AI without critical engagement reinforces a passive “banking model of pedagogy,” creating an illusion of mastery. AI-based learning misses vital qualitative elements, specifically institutional friction and lived political reality.
There are no shortcuts to education, Mohammad Asif observed. Students must consult expert human mentors rather than attempting to replace human teachers with AI. He also urged for clear policymaking regarding AI use in human dynamics.
When asked if he had found any qualitative change in students’ delivery, the teacher said, “The most interesting part is the Literature Review (LR) of any given research like assignment or credited thesis. AI generated LR is rarely original and mostly fake. The format is good, but the literature has no existence.”
He said students often use references produced by AI tools in their submissions. But when asked to recheck, they often fail to present the sources they used in their LR or assignments.
Asif urged for clear policymaking regarding AI use in human dynamics, with ethical and moral boundaries. He said, “AI detection systems could be bypassed by other AIs. It generates profit for AI companies only. We should use our brains and reasoning. Otherwise, robots will replace us.”
Software developer Waesh Ahmed, from Penta Global Limited, said a balance between expectations and outcomes should be maintained. Traditional technical learning relied on searching through books, Stack Overflow, blogs, and YouTube videos. AI now acts as an agentic learning assistant, summarising complex topics into simple, tailored answers in minutes. AI and agents help a developer’s productivity by rapidly retrieving targeted, to-the-point information.
But, Waesh Ahmed said, “Friction and struggle are essential elements of learning that solidify understanding. While using AI agents we can find the relevant and to the point information fast, it is also stripping down the effort it takes to learn something. Honestly, I don’t know if this is good or bad, but from my experience I would say, learning this way makes me less confident in the subject matter.”
Developers have tight deadlines to implement anything. They feed AI agents with long technical documents, and within minutes the documents are compiled and ready for implementation. “But the solid feelings that you have when you know something is missing here. If I want a solid understanding of something, I can talk to the AI for more clarification given that I have enough time. The thing is ‘enough time’. In my professional life, my seniors often expect that since we have AI, we do not need as much time researching something. I would not say that is a totally unrealistic expectation. But we often go with ‘whatever the AI says’ because of the time constraint,” Waesh observed.
He is, however, optimistic about the peer role of AI. Finding human study partners who share a mutual passion for technical topics is often difficult. AI successfully fills this gap, serving as an interactive companion available anytime to discuss concepts and clarify complex topics when time permits.
In spite of this human peer benefit, Muzeeb Mahatheer Ibn Eaunuss Sonnet, a student of Asia Pacific University’s English Department, sometimes feels deprived of the enjoyment of reading longer texts, as AI can summarise those in easy and expected ways.
“My class tests, mid-terms, and semester finals have a tight schedule. During such a rush, I use AI to summarise everything in any specific book to understand what the topic is all about,” he said. The student also finds an emotional distance from the answers he prepares with the help of AI, as he has little time to taste and explore the literary pieces.
When asked about the process he follows to deal with abstract poetry or prose with the help of AI, Muzeeb said, “AIs help me to get answers in minutes rather than thinking for a long time. This developed a tendency to get answers to complex questions quickly rather than thinking about it for a long time because of the tight exam schedules. So, I need to do so, for time management.”
Although AI gives him answers quickly, the student found a lack of emotional binding that he finds while discussing with a peer.
To understand the real impact of AI, we need organised and unbiased research activities on a massive scale. This means larger studies across more institutions, disciplines, and income groups, carried out without a fixed conclusion in mind.
There is also a human side to this story that policy makers cannot ignore. Teachers like Syed Alauddin Ahmed, perhaps, are losing income as AI changes how students learn.
Teachers like Syed Alauddin need training and support from the government to adapt their methods to the latest technology. Without that support, the burden of this transition falls unfairly on teachers who are already struggling.
