[🇧🇩] Artificial Intelligence-----It's challenges and Prospects in Bangladesh

[Saif]

Let AI crunch the numbers

At a recent global summit on artificial intelligence with a group of senior executives in Dhaka, a chief financial officer put a question to the room that this scribe has not stopped thinking about. "If the machine starts making our financial decisions," he asked, "what exactly will be left for us

thefinancialexpress.com.bd

Let AI crunch the numbers
The judgement must be leave to human

At a recent global summit on artificial intelligence with a group of senior executives in Dhaka, a chief financial officer put a question to the room that this scribe has not stopped thinking about. "If the machine starts making our financial decisions," he asked, "what exactly will be left for us to decide?"

It is the right question, but it rests on the wrong assumption. AI does not take judgement away from finance leaders. It strips away the guesswork that has long passed for judgement, and leaves behind the decisions that only humans can, and should, make. For boardrooms across the world, this is no longer a matter for futurists. It is a daily operational reality.

Consider how quickly the corporate world has reorganised itself. The 2026 CEO study by the IBM Institute for Business Value, which surveyed 2,000 chief executives across 33 geographies, found that 76 per cent of organisations now have a Chief AI Officer, up from just 26 per cent a year earlier. The study also reports that around 64 percent of CEOs are now comfortable making major strategic decisions on the basis of AI-driven analysis.

Finance functions are moving in the same direction. KPMG's Global AI in Finance 2026 report finds that more than three-quarters of companies now use AI in financial planning, reporting and business analysis, and 71 per cent say the technology is meeting or exceeding their expected return on investment. The largest gains are showing up not in headcount savings, but in the quality of decisions, the accuracy of forecasts and the speed of response.

In other words, AI has left the IT department and taken a seat at the strategy table.

BANGLADESH: At home, the picture is more uneven. Research by the Bangladesh Institute of Bank Management (BIBM) found that 68 percent of the country's banks have no formal policy or guideline on the use of artificial intelligence. Only 32 per cent do.

That gap matters more than it may appear. Banks are already experimenting with chatbots, automated credit scoring, fraud detection and customer analytics, often through vendors and pilot projects. Without a governing framework, nobody can say with confidence who is accountable when an algorithm declines a loan, flags the wrong customer, or mishandles personal data. For a sector still rebuilding public trust after years of governance failures, this is not a technical footnote. It is a strategic risk.

There is a financial inclusion dimension too, which rarely enters the debate. An AI credit model trained on the data of salaried urban borrowers may quietly score a woman entrepreneur in Satkhira, a farmer in Rangpur or a person with a disability as "high risk" simply because people like them are thin in the data. Unchecked, the very technology that promises to widen access could end up hard-coding old exclusions.

How, then, should executives bring AI into financial management without surrendering control?

First, move from instinct to evidence. In many of our institutions, spending and saving decisions still rest on habit or the hunch of a senior person. Fear, greed and overconfidence slip in unnoticed. AI can act as a neutral mirror: by analysing past expenditure, it surfaces hidden patterns, emotional decisions and risks that no committee has spotted. A weekly review of spending and saving patterns, whether for an institution or a household, is no longer a luxury. It is basic management discipline.

Second, make decisions faster and leave a clearer trail. Long committee meetings built on competing assumptions can give way to a data-driven "second opinion" prepared before anyone sits down. Leaders can then spend their time on strategy rather than on arguing over numbers. Just as importantly, AI-assisted processes create a digital record of who decided what, when, why and on what evidence. That record is the backbone of accountability.

Third, pursue growth that survives bad years. Success should not be judged by attractive short-term revenue figures. Every expansion plan should be stress-tested against a possible downturn or market shock. Savings generated by operational efficiency should not be absorbed into routine running costs; they should be reinvested in promising products, skilled people and modern technology.

Fourth, invest where the evidence points, not where the hype is. Before committing capital, ask the hard question: what is the real return? Spending on technology should be treated not as an overhead but as an investment, and its results should be monitored continuously, not celebrated once at launch.

These principles are designed for institutions, but they work just as well for individuals. This is the thinking behind what is called the "Grandfather Concept of Personal Money Management", which this scribe have presented to students at nearly sixty universities under Prime Bank's "Empowering Youth" platform, as well as at professional forums.

The framework divides a person's financial life into seven stages: building financial literacy until 18; learning to earn, budget and stand on one's own feet from 18 to 25; forming a family and starting long-term investment from 25 to 35; deepening skills, building assets and protecting the family from 35 to 45; reaching the career peak, clearing debt and building a retirement fund from 45 to 55; transitioning to passive income from 55 to 65; and, beyond 65, preserving wealth, finding peace of mind and passing assets on smoothly to the next generation.

At each stage, a simple rule helps keep income, spending and saving in balance: the 50-30-20 rule popularised by US Senator Elizabeth Warren, a former bankruptcy law professor. Half of income goes to needs, 30 per cent to wants and lifestyle, and 20 per cent to savings and the future.

Interestingly, Bangladesh's national savings rate, according to provisional estimates from the Bangladesh Bureau of Statistics (BBS), stands at about 27 per cent of national income (26.93 per cent in the current fiscal year, against 27.67 per cent the year before), broadly in line with that rule. The harder questions are how fairly those savings are distributed across income groups, and how much of them actually turn into productive wealth.

AI can make this entire framework sharper. A young professional can run her first salary through a budgeting app and see at once where the money is leaking. A CEO can re-examine the cost structure of an entire organisation in real time. The scale differs; the principle is the same: data over guesswork, evidence over emotion.

RULES BEFORE SCALE: None of this works without two foundations: reliable data and a human-made ethical framework. An algorithm is only as good as the information fed into it and the boundaries set around it. The recent breach in which the personal data of millions of users of a local job portal was reportedly exposed should serve as a warning of what weak data governance costs.

Bangladesh Bank, BIBM and the banks themselves should therefore move without delay to develop a common, clear AI-use guideline for the financial sector. It should cover data protection, model validation and explainability, cybersecurity, bias testing for credit decisions, and, above all, a clear line of human accountability for every automated outcome.

AI brings unprecedented speed, transparency and evidence to financial decision-making. But ethical values, social responsibility, empathy and long-term judgement remain, and must remain, human responsibilities. The quality of our decisions will depend on how well we learn to question the machine. Let technology widen our reach, but let human wisdom lead. That is the lesson today's C-suite cannot afford to miss.

The writer is a banker, development researcher and author.​

 

[Saif]

https://www.newagebd.net/post/opinion/319836/ai-adoption-without-preparation-harms-more-than-it-benefits

AI adoption without preparation harms more than it benefits

Sabrina Luthfa

WALK into almost any university in Bangladesh today and you will hear the same word everywhere: AI. ChatGPT workshops fill up within hours. Departments rush to add AI to course outlines, and students proudly show how a chatbot wrote their assignment, designed their poster or produced a video in minutes. Some of this excitement is justified. But it is starting to look like an obsession, and obsessions rarely leave room for the slower question: are we ready? For much of our education system, the honest answer is not yet. When a powerful technology is adopted by an institution that lacks the foundation to support it, adoption can do more harm than good. Bangladesh’s universities have embraced AI faster than they have built the thinking, training and structures needed to use it well.

In one of my research projects, I studied healthcare innovation in western Sweden. One startup had developed a drone to carry blood between two hospitals, built together with the region’s largest hospital. Adoption should have been easy. Instead, the state authority demanded proof that the drone would never crash onto a vehicle or a person, and the permit took two years of experiments. Then the hospitals hesitated. Would blood quality suffer at altitude? Would the landing noise disturb patients? What would happen to the drivers who transported the blood? Could staff learn to monitor and land the drone safely? Eventually, the company took its drone to an African country where few rules stood in its way. Adoption was quick, and on paper it looked like success. But speed of adoption is not readiness. The Swedish questions were not obstacles to innovation; they were the infrastructure of responsible adoption: regulation, safety, quality assurance, workforce skills and a plan for the people affected.

When I look at AI in Bangladeshi academia, I see the second pattern, not the first. We are fast adopters of almost everything; in a country of more than 170 million people, even a small share adopting something looks enormous. In education, adoption without those questions is a real risk.

Our students are not short of digital confidence. Most know how to type a question into a chatbot or ask it to generate an image or video. What many cannot yet do is prompt well enough to learn something in depth. That is not a technical skill; it is a thinking skill. A good prompt comes from a good question, and a good question comes from a mind that can break a problem into parts, spot a weak argument and tell when an answer is plausible but wrong. When we used Google, those who benefited most already knew what to trust, what to rephrase and what to ignore. AI raises the stakes of that skill rather than replacing it.

Here lies our problem. The majority of students in Bangladesh have not been trained to think critically at school, where classrooms have long rewarded memorisation. I saw this in my own classroom recently. I asked my students to read two articles and prepare questions for their classmates that would require critical thinking to answer. I told them not to memorise anything, only to arrive with a clear conceptual understanding of each article. In class, the majority asked the same questions, most likely suggested by AI. When their classmates could not answer, I turned the questions back to the students who had asked them. They could not answer their own questions either. The questions had come from a chatbot, not from their understanding of the articles. They had the right words but not the thinking behind them.

Consider a common research task: developing a theoretical framework. A student with weak critical thinking will ask AI for ‘a theoretical framework on employee motivation’ and receive a tidy list of well-known theories. What AI will not produce, unless guided by a thinking user, is an integrated model that explains why these theories belong together, how their concepts connect, where they conflict and how they fit a specific context, such as a garment factory in Gazipur. That integration is where the depth lies, and it comes from the student’s own reading and reasoning. Without it, the prompt stays shallow, and so does the output.

There is a second gap AI cannot fill. A chatbot can describe how a supply chain works; it cannot teach a student how to handle a late shipment, an angry supplier and a manager who needs an answer by afternoon. It can summarise theories of community health, but it cannot show a student why a well-designed awareness campaign fails in a village where nobody trusts the messenger. Real-life solutions demand real-life work experience. Bangladesh needs learning by doing, not learning by memorising, and certainly not learning by asking AI for the answer. Work-integrated learning, where students learn through structured, assessed experience on real problems, builds the very judgement that good AI use depends on. A student who has worked on a real problem knows which questions matter, and will prompt better and evaluate output better as a result.

None of this is an argument against AI. It is an argument about sequence. Before a technology is adopted, we need the structure to support it. When the 2022 curriculum tried to shift towards experiential learning, an NCTB survey found that only 37.6 per cent of teachers even roughly understood its assessment system, and the reform was withdrawn in 2024. Many students therefore reach university never having been asked to question a source or defend a position of their own. In practical terms, that means teachers need to be trained first, so they can redesign assignments and assessment, and teach students to use AI critically. Critical thinking has to be set as a core outcome, from school onwards. Questioning, argument, comparison of sources and problem-solving should be taught and assessed explicitly, starting long before university. The 2027 secondary curriculum is an opportunity to get this right. Work integrated learning has to be encouraged in schools and universities, with internships, industry projects and community problem-solving that are properly designed, supervised and assessed. Assessment that rewards thinking has to be integrated, through oral defences, applied projects and staged submissions that make a student’s reasoning visible. Above all, there has to be a clear institutional AI policy to define where AI use is encouraged, where it must be disclosed and where it is allowed.

Our young people are curious and quick to learn. But a tool is only as good as the mind that directs it. If we adopt AI without first building critical thinking, practical experience and institutional support, we risk graduates who can generate answers but cannot judge them. The drone that flew without questions was adopted faster. That did not make it the better outcome. In education, preparation is not a delay to progress. It is what makes progress last.

Dr Sabrina Luthfa is an associate professor, and innovation advisor of CERAL in ULAB Business School, ULAB​

 

[Saif]

How to find a good startup idea in the age of AI

Ask ChatGPT or Gemini for 50 startup ideas for Bangladesh and the list arrives in seconds, which feels like magic until you realise thousands of young founders from Mirpur to Moulvibazar are typing the same prompt and getting almost the same answers. The blank page has given way to abundance, and a

thefinancialexpress.com.bd

How to find a good startup idea in the age of AI

Ask ChatGPT or Gemini for 50 startup ideas for Bangladesh and the list arrives in seconds, which feels like magic until you realise thousands of young founders from Mirpur to Moulvibazar are typing the same prompt and getting almost the same answers. The blank page has given way to abundance, and abundance raises a harder question: if a machine can brainstorm the idea, research the market and write most of the code, what is left for the founder to do?

The question matters more here than almost anywhere, because AI companies drew about 74 per cent of a record US$ 510 billion raised worldwide in the first half of 2026, as this newspaper has reported, while Bangladeshi startups raised just US$ 6.0 million, a 95 per cent fall year on year, according to LightCastle Partners, so no investor is coming to rescue a weak idea.

Much of a founder's traditional groundwork has already been handed to machines and the websites built around them. StartupIdeasDB.com lists more than 8,800 problems gathered from Reddit, Hacker News and GitHub, each with an AI opportunity score and a 2,000-word roadmap, while Y Combinator's Requests for Startups, Product Hunt and Indie Hackers reveal what investors want and what earns revenue. Building is cheap too, since Y Combinator's Jared Friedman said a quarter of its Winter 2025 startups had codebases 95 per cent written by AI.

This abundance hides a trap that researchers have measured: in a 2024 Science Advances study, Anil Doshi and Oliver Hauser found that writers given AI-generated ideas produced stories rated as more novel but those stories were roughly ten per cent more alike than ones written without help. Startup ideas are unlikely to escape that pull, which means an idea a chatbot hands you is, almost by definition, also being handed to your competitors.

This is where Bangladesh holds an advantage that is easy to miss from behind a laptop. AI models know a great deal about a dentist in Ohio and very little about a pharmacy owner in Bogura, partly because Bangla, despite about 237 million native speakers, is underserved in their training data, and somewhat because roughly 84 per cent of our workforce is informal, by BBS and ILO estimates, trading through khata, cash and phone call no web crawler will ever index. The founder who sits in the shop and listens to the wholesaler holds the one asset AI cannot generate: first-hand knowledge of problems nobody has written down.

Knowing the problem is only half the work because building has become the easy part and the real contest has moved to distribution and dependence. When a competent clone can appear within weeks, what matters is whether people find your product, use it every week and would feel lost without it. Growth marketer Sean Ellis's survey captures that feeling, since startups where at least 40 per cent of users say they would be "very disappointed" to lose the product tend to find real traction.

Four layers can make an idea defensible, and the strongest Bangladeshi companies stack several of them. The first is a service angle wrapped around the product, from onboarding done for the customer to a person who answers the phone in Bangla, because a rival can copy code within weeks but not relationships built over years. The second is deep local innovation, which separates a lazy copy of a Western model from a Bangladeshi original: Pathao bet on motorbikes and cash, Darazleaned on cash on delivery and empowering local sellers, and bKash turned more than 350,000 agents into human cash machines for over 83 million registered customers.

The third is a brand, personal or corporate, because when products look alike people choose the name they trust. Ayman Sadiq has said 10 Minute School found brand-market fit long before product-market fit, and his Facebook following of about 5.4 million in 2022 brought reach that, he argued, rivals would need heavy advertising to match. The fourth and deepest is community and network: groups, referrals and peer help that make a product more valuable with every new user, the way bKash became more useful as relatives, shops and employers joined. With about 64 million Bangladeshis reachable on Facebook, communities here form long before apps do, and a founder who builds one owns a distribution channel no clone can download.

Each kind of business leans on these layers differently: a pure service firm sees AI erode its cheap-labour margins first and survives by selling outcomes and accountability rather than hours, while a marketplace like Pathao or Daraz must match supply and demand at the same moment, which no algorithm can do without real riders and stocked warehouses. Infrastructure like bKash is the hardest to copy because it rests on a regulatory licence and hard-won trust, and its initial approval for a digital bank in 2026, with plans to judge borrowers by transaction histories, points to the biggest AI-era prize: turning invisible economic activity into reliable data.

Product companies offer Bangladeshi founders the clearest route to dollar revenue on a taka cost base. EzyCourse, a rival to Kajabi and Teachable, serves customers in more than 100 countries, Ollyo sells Tutor LMS worldwide, and Startise, the group behind WPDeveloper, Storeware and easy.jobs, reports more than six million WordPress users, over 26,000 Shopify stores and 28,000 businesses on its AI hiring platform. Their moat was never just code, which rivals can rebuild in weeks, but distribution inside WordPress and Shopify, loyal user communities and years of earned trust.

Before committing your savings, ask what you know that the model does not, and whether the business grows stronger or weaker each time the models improve, since anything one ChatGPT update could swallow is a feature, not a company. AI has also collapsed the cost of being wrong, so a founder can test several ideas through landing pages and pre-orders for what one prototype once cost, turning the scarce skill from guessing correctly into learning quickly.

What is left for the human founder is presence where problems live, the judgement to choose one idea from a thousand, the taste that a human experience can bring, the trust that persuades a shopkeeper to share his records, the accountability of someone who answers when things go wrong, and the patience to build a community one conversation at a time. Fifteen years ago, the cheap mobile phone was not itself Bangladesh's great opportunity but the ground on which bKash was built, and AI is this decade's cheap phone, so the winners among Bangladesh's 1,200-plus active startups will be those who go where the models cannot see.​

 

[Saif]

https://www.newagebd.net/post/telecom/320153/skilled-workforce-emphasised-to-harness-ai-potential-in-tourism

Skilled workforce emphasised to harness AI potential in tourism

Staff Correspondent . Dhaka

Bangladesh must develop a skilled workforce, improve service quality, and adopt artificial intelligence to realise its tourism potential, National Skills Development Authority executive chairman Nazneen Kawshar Chowdhury said on Thursday.

‘Without competency-based training, there will be no skilled workforce, because completion certificates produce an unskilled one,’ she said at the inauguration of the Bangladesh Travel and Tourism Fair 2026 in Dhaka.

The three-day fair, organised by the Tour Operators Association of Bangladesh under the theme‘AI Powered Tourism, Sustainable Destination’ , was inaugurated by Civil Aviation and Tourism secretary Nurun Nahar Chowdhury as chief guest at the Bangladesh-China Friendship Conference Centre in Agargaon.

Representatives from 20 countries are participating in the fair, which features more than 200 booths and stalls, over 220 exhibitors, over 250 qualified buyers, over 2,500 trade visitors, and more than 60,000 expected visitors.

Chairman of Bangladesh Parjatan Corporation Md Shamimuzzaman, CEO of Bangladesh Tourism Board Mutasim Billah Faruqui, Ambassador of the Philippines to Bangladesh Nina P Coinolet, Ambassador of Nepal to Bangladesh Ghanshyam Bhandari, and president of TOAB Mohammed Hanif, among others, attended the event.

Nazneen said the government had prioritised competency-based training to equip workers with practical skills required by the labour market.

She urged private training institutions, hotels, resorts, travel agencies and tour operators to help identify industry needs and develop appropriate curricula.

According to NSDA data up to September 30, 19 competency-based curricula had been developed for tourism and hospitality, with 252 registered training institutions.

Nazneen said Bangladesh’s tourism attractions, including the Sundarbans, Cox’s Bazar, hill regions and Sylhet’s tea gardens, required skilled workers, quality services, modern technology and tourist safety.

She highlighted AI’s potential in travel planning, bookings and marketing, stressing digital, communication and language skills.

The NSDA has taken initiatives to introduce eight foreign-language programmes, five of which have already been approved as competency-based programmes, including English for Work, German, Arabic, Korean and Chinese.

Tourism Secretary Nurun Nahar said integrating technology and artificial intelligence could help position Bangladesh’s tourism sector internationally while promoting environmentally friendly tourism.

Mohammed Hanif said the fair would showcase domestic tourism internationally and strengthen commercial relationships with global stakeholders.

The fair will remain open from 11:00am to 8:00pm until October 31.​

 

[Saif]

https://www.newagebd.net/post/opinion/320271/ai-dilemmas

AI dilemmas
Obaidul Hamid

Representational image. | New Age/AI

I CAN’T sufficiently explain my relationship with generative artificial intelligence. As a moderate user, I utilise it mainly for instrumental purposes. I am neither overenthusiastic nor overly sceptical about this advanced technology. It is nowhere near a love-hate relationship.

Recently, I engaged with GenAI somewhat more substantially than is my customary practice. I was working with a co-author on a journal article for a themed issue on AI in language policy and planning. We set out to examine the role of AI as a potential language policy actor and to theorise its agency. Agency is a foundational concept in this field and is generally understood as the capacity that drives an actor to seek changes in languages and language behaviours and to make a difference in the world.

We thought it would be worthwhile to engage various GenAI models, such as ChatGPT, for the article. Using AI as a research participant is, for me, one of the boldest uses of the technology beyond typical tasks such as editing, summarising, answering queries, and locating information. This specific research application may raise integrity questions, but we cannot generalise about it. It depends on how we engage the tool, use its output, and for what purposes. We ended up interacting with three widely used GenAI models and critically analysing the resulting data to pursue our research goals.

As we completed the first full draft of the article, we wanted to use AI to edit it. Our draft exceeded 9,000 words and had issues of clarity and coherence. As we know, GenAI is good at editing, and it can improve not only clarity but also the strength of arguments.

However, this editing process became a source of uncomfortable dilemmas. Many other writers may be facing similar quandaries as they use AI as an editor. Therefore, the story is worth telling.

The algorithmic editor cut one-third of the original draft, producing a sanitised version of just over 6,000 words. It performed what we would call a ‘caesarean section’ on the text.

The draft had repetitions and inconsistencies, no doubt. Yet the editing did not stop at those undesirable features. It also removed many other elements from the text that the editor apparently considered unnecessary or misaligned with dominant native-speaker norms of English.

The edited version was much cleaner and tidier, ensuring greater flow and readability. This was because the machine editing ironed out idiosyncrasies, contextual details, and non-native excesses. Many readers would probably prefer the edited version if they had the opportunity to read the original draft as well. Reading became faster and easier, reducing the cognitive burden on readers.

From an academic integrity standpoint, there might have been no problem in submitting the edited version to the journal. Editing is a widely accepted use of AI. It is generally assumed that AI edits texts for accuracy and clarity without substantially altering their content or ideas. In this respect, it is comparable to translation, which renders material into another language while preserving its original meaning. We would, of course, have needed to disclose the use of AI, as is now required by most journals.

Should we have submitted the edited paper to the journal? Different authors would likely answer this question differently. However, we could not proceed with that version.

It is the responsibility of authors to make their writing clear and comprehensible so that readers can appreciate its value and contribution. Editing helps achieve this outcome. This form of language engineering is particularly useful for writers who are non-native speakers of English. Indeed, editing has emerged as a global industry largely serving this clientele.

Even though AI is now everywhere, it is not the only editor in town. We have human editors, including professionals, teachers, research supervisors, journal editors, and copy-editors, as well as technologies such as Microsoft Word, Grammarly, and many others. Why, then, would we reject AI editing when we have no objections to other forms of editing? Are we prejudiced against AI?

Not really, but there are good reasons for hesitation. The first concerns ownership of the edited text. Even if the article’s content remains largely unchanged, can non-native writers still claim that the language is theirs? I am not sure how many users of English from countries belonging to the Outer and Expanding Circles would comfortably embrace language generated by an algorithmic editor. My own dilemma is even more acute. I have been advocating what I call ‘Southern English’, a reconfiguration of its Northern, native counterpart. AI placed me in a situation where I had to choose between conflicting options.

Yet even claiming ownership of the edited English does not resolve the problem. As some researchers have noted, once non-native users of English embrace AI-generated English, they may need to maintain that standard for the rest of their lives, relegating their own version of English to the margins. Sustaining such a dual linguistic existence may be difficult: displaying algorithmically nativised English on some occasions while reverting on others to an earlier version marked by idiosyncrasies, local flavours, and signs of non-nativeness. AI may not provide the linguistic security it promises to those who need it most. What appears to be security may simultaneously become a source of insecurity. How should we reconcile this with the popular discourses of empowerment and democratisation associated with AI?

Once non-native English speakers embrace AI English, they may become dependent on it to save face. From one perspective, this means becoming a permanent customer of AI and contributing to techno-capitalism. From another, it resembles the Doctor Faustus syndrome: selling one’s linguistic soul to Mephistopheles. Once the bargain is struck, there is no easy escape. Ngũgĩ wa Thiong’o might have called it the ‘colonisation of the mind’. What appears to be linguistic empowerment then becomes a form of algorithmic imprisonment, paradoxical as that may sound.

But there are other dimensions to the AI English story. One fundamental question concerns the nature of human writing, or indeed any human activity. The dominant tradition of humanism has encouraged us to embrace human supremacy, exceptionalism, and agency. We have tended to regard our autonomy as unquestioned and unconditional. Yet the reality may be more complex. As posthumanist scholars argue, humans and nonhumans are interdependent rather than separate or independent. Human-nonhuman reciprocity is evident in nearly every aspect of life. From a posthumanist perspective, humans are never the sole authors of anything. We rely on tools in every action involving physical, intellectual, or creative labour. Writing itself may not be exclusively human if it is mediated by tools, whether the pen, the typewriter, the dictionary, or AI. A piece of writing may therefore already be a product of human-tool co-authorship, regardless of whether AI or another editor is involved later.

Beyond these external influences, there is also something within humans themselves as authors of texts and creators of other works. Our thinking is shaped by learning and socialisation in schools, communities, workplaces, and societies. Can we really regard the thinking that underpins writing and creativity as entirely subjective? Can I describe this piece of writing as single-authored when so many external influences are traceable within my thinking and productivity?

While posthumanism invites us to think beyond the human, AI policies at national, institutional, and intergovernmental levels continue to emphasise human control and oversight. The argument that humans should remain in the driver’s seat in the AI era appears almost unassailable. Underlying this position may be an assumption that human agency faces existential threats from AI.

At the same time, some researchers have stressed the dehumanising potential of AI and have called for greater humanisation and human oversight as remedies.

Many people are likely to find themselves in a maze, uncertain whether the way out lies in humanism, posthumanism, both, or neither. Whatever the answer may be, GenAI has prompted us to search for deeper truths about human life, agency, creativity, and what it means to function as human beings.

Obaidul Hamid is an associate professor at the University of Queensland in Australia. He researches language, education, and society in the developing world. He is a coeditor of Current Issues in Language Planning.​

 

[Saif]

Bangladesh must protect digital inclusion as the AI boom drives up costs

As AI giants such as OpenAI, Google, and Meta buy enormous quantities of chips, memory, and storage, manufacturers are prioritising these big-spending customers over the consumer electronics market.

www.thedailystar.net

Bangladesh must protect digital inclusion as the AI boom drives up costs

Mahmudul Hasan

‘As AI giants such as OpenAI, Google, and Meta buy enormous quantities of chips, memory, and storage, manufacturers are prioritising these big-spending customers over the consumer electronics market.’ PHOTO: REUTERS

“Rajay rajay juddho hoy, ulukhagrar pran jay” is a well-known Bangla proverb that conveys how ordinary people often bear the brunt of rivalry between powerful nations, individuals, or groups. These days, it also captures an uncomfortable reality of the global artificial intelligence (AI) race. The world’s most powerful countries and technology companies are competing for supremacy in a technology that promises to transform economies, industries, and societies. But the cost of that competition is increasingly being felt far beyond Silicon Valley and Beijing.

As AI giants such as OpenAI, Google, and Meta buy enormous quantities of chips, memory, and storage, manufacturers are prioritising these big-spending customers over the consumer electronics market. This is tightening supplies of components for laptops and smartphones, pushing up prices, and making devices less affordable for ordinary consumers. In Bangladesh, computer prices are becoming exorbitantly high, smartphone manufacturers are facing rising component costs, and businesses are postponing technology purchases.

For a country seeking to expand digital inclusion, develop an IT-enabled services industry, and prepare its youth for an AI-driven economy, this creates a troubling contradiction. The technology that could help Bangladesh leapfrog development barriers is making some of the tools needed to use it less affordable.

The AI boom is reportedly making computers significantly more expensive. Entry-level laptops that cost Tk 47,000-Tk 48,000 a year earlier are selling for Tk 65,000-Tk 70,000. Laptop and desktop sales volumes have fallen by around 20 to 25 percent over the past year, even as the total value of sales has increased.

Computer use in Bangladesh already remains very low, with only 11.8 percent of individuals and 9 percent of households using or owning computers. Fewer than 3 percent of manufacturing units in Bangladesh use computers or information technology (IT) in their production work, according to the 2024 Economic Census. And now, AI is making computers less affordable for individuals and industries alike.

Bangladesh’s mobile phone industry is also facing rising component costs alongside domestic structural challenges. One report by this newspaper documented how rising memory prices are adding to the pressures on local manufacturers. RAM and storage components have become more expensive, increasing production costs for manufacturers already grappling with grey-market competition, weak factory utilisation, and policy uncertainties. Industry representatives have reported steep increases in memory prices since the beginning of the year, with further increases expected. Such pressures can feed through to handset prices, making affordable smartphones harder to produce and purchase.

The difficulties facing local manufacturers could also undermine Bangladesh’s ambition to build a stronger domestic digital device industry. The country’s handset production fell from roughly 2.73 crore units in 2024 to 2.44 crore units in 2025. Global component shortages are not the only reason behind the industry’s difficulties, but they add another burden in an already challenging business environment.

Bangladesh has also long struggled with low smartphone penetration, with nearly half of mobile users—47 percent—still relying on feature phones. To address this, the government is seeking to make low-cost smartphones more accessible and phase out 2G services by 2029 to accelerate digital inclusion. However, rising component prices driven by the global AI boom threaten to undermine these efforts. Indeed, digital penetration in Bangladesh is likely to remain constrained beyond 2027 due to shortages of components for laptops, desktops, servers and other essential digital infrastructure, as the intensifying global AI race continues to drive up component prices.

The United States and China increasingly treat AI as a strategic technology, with consequences for economic competitiveness, national security, and military power. Washington restricts China’s access to advanced chips and chipmaking equipment, while Beijing is investing in domestic alternatives. The rivalry now extends to the models themselves.

On September 8, US law enforcement and intelligence officials accused six Chinese AI companies, including DeepSeek, Moonshot AI, and Alibaba, of using distillation—training a cheaper model on the outputs of a more powerful one—to copy US technology. The allegations were disputed and Beijing has called similar claims groundless. It is also building barriers of its own: on October 16, China proposed export controls on 40 technologies, 23 of them tied to AI.

The competition is not confined to governments. OpenAI, Google, Microsoft, Meta, and Anthropic compete with DeepSeek, Alibaba, and Tencent for customers, talent, and computing capacity. DeepSeek showed that Chinese developers can challenge established players with cost-efficient methods. Even so, the spending is enormous. Alphabet, Amazon, Meta, and Microsoft are expected to invest about $650 billion in AI infrastructure this year, up roughly 60 percent from $410 billion in 2025.

The infrastructure splurge is driving shortages of memory as manufacturers prioritise chips destined for data centres over those for everyday PCs and phones. Xiaomi’s general manager in Bangladesh has said that AI companies are “effectively writing blank cheques” to memory makers. Counterpoint, a global technology market intelligence and research firm, expected memory prices to climb nearly 50 percent through the first quarter, while Gartner projects PC prices to be 17 percent higher and smartphones 13 percent higher than in 2025.

Bangladesh’s computer importers and handset manufacturers are failing to match the purchasing power of global technology giants. They are absorbing higher costs or passing them on to consumers. Export restrictions and trade barriers also make sourcing costlier, and Bangladesh feels the effects through international prices. As wealthy countries and corporations invest billions of dollars to gain an AI advantage, developing economies are being forced to find more money simply to buy the devices needed for citizens to take part in the digital economy.

Connectivity alone can no longer deliver digital inclusion. People also need affordable devices, reliable internet, access to electricity, and necessary skills. If device prices outpace household incomes, expanded networks may not translate into wider participation. There is a longer-term industrial risk too, as higher-value digital services need a broad base of digitally equipped workers and businesses, not just a few well-funded companies.

Bangladesh cannot influence the strategies of Silicon Valley or Beijing, and it should not treat AI as a threat or be put off by it. The technology could create real opportunities in education, healthcare, manufacturing, agriculture, and public services. The challenge is to ensure those benefits are not reserved only for people who can afford increasingly expensive hardware. The government must keep cutting unnecessary duties on computers and components. It’s commendable that the government withdrew import duties and VAT on laptops, desktops, servers, printers, and monitors in the FY2026-27 budget. But tax cuts cannot overcome global supply constraints, so their impact should be monitored and other price-raising barriers reviewed.

Support for local handset and electronics manufacturing must also become more predictable, with stable taxation, action against illegal imports, and incentives linked to productivity. Technology supply chains should be diversified, with procurement that values alternative suppliers, repairability, longer device lifecycles, and spare parts. Affordable access schemes—including targeted financing, quality-assured refurbished devices, institutional computer lending, and shared computing facilities—would help students and low-income households.

It is encouraging that Bangladesh plans to launch sovereign AI within a year. However, the initiative must be backed by thorough planning, adequate computing infrastructure, skilled manpower, reliable data, and sustainable investment to ensure meaningful, long-term benefits. Investment should also reach where it can build real strength: digital skills, Bangla-language applications, local software, responsible AI adoption, and higher-value services.

Each time an ordinary citizen or small business is gatekept out of buying necessary devices, it is a setback to the country’s digital ambitions. Bangladesh must ensure its people and businesses do not become the casualties of global battles for AI supremacy.

Mahmudul Hasan is a senior reporter at The Daily Star, covering AI, telecommunications, and technology.​

 

[Saif]

Creating right conditions for AI adoption

Artificial intelligence has become one of the most talked-about technologies in the business world, and for good reason. Every business aims to accomplish more at lower cost and AI offers a clear pathway to achieve this goal. Bangladeshi businesses are no exception, as demonstrated by the recent AI

thefinancialexpress.com.bd

Creating right conditions for AI adoption

Artificial intelligence has become one of the most talked-about technologies in the business world, and for good reason. Every business aims to accomplish more at lower cost and AI offers a clear pathway to achieve this goal. Bangladeshi businesses are no exception, as demonstrated by the recent AI Business Summit organised by the Dhaka Chamber of Commerce and Industry and the Ministry of Science and Technology. The summit produced the usual mix of ambitious statements and sensible warnings. Speakers at the event emphasised developing AI applications that address real-life industrial problems rather than simply adopting imported technologies. This is an important distinction because, while imported applications provide useful capabilities, they often fail to address the specific challenges local businesses face. Platforms designed for other markets rarely account for local supply chain constraints or domestic labour management practices, and expecting them to adapt independently is unrealistic. Building local solutions is therefore essential for Bangladesh rather than remaining a consumer of foreign technology. Indeed, local developers who work directly with business operators have a far better chance of creating effective, context-appropriate solutions.

Industry leaders at the summit rightly urged young developers to leave conventional digital platforms behind and spend time on factory floors to understand where money, materials and time are actually being wasted. There is plenty of scope for AI applications that address these everyday business problems. In the garment industry, for instance, AI could help manufacturers improve production planning, identify defects and reduce raw material wastage. In the pharmaceutical and food industries, it could tighten quality control and reduce losses during production. Even something as straightforward as using past sales data and customer trends to predict demand can make a huge difference. This would help businesses keep inventory lean and avoid unnecessary costs. These may seem like routine business functions, but small improvements can add up to significant savings, especially in industries running on thin margins. However, identifying useful applications is only half the battle. Businesses also need to be ready to invest in the technology, improve their data management and train employees to use it well. In practice, though, this could prove particularly difficult for small and medium-sized enterprises which often lack the resources to experiment with new technologies.

This is where a supportive policy framework becomes essential as it can make AI adoption easier and less costly for businesses. In addition to finalising the National AI Policy 2026-2030, the government should create incentives for businesses to invest in research, employee training and technological innovation. As mentioned at the summit, allowing such investments to qualify for tax deductions could encourage private sector spending without making businesses dependent on direct government grants.

Financial incentives alone will not be enough, though. Bangladesh also needs closer cooperation between industries, universities and technology developers to ensure that research addresses actual business needs. Universities, in particular, must align their curricula more closely with the practical demands of industry, giving greater attention to AI literacy, data analytics and problem-solving skills. Businesses, for their part, must be willing to share their operational challenges with researchers and give young developers opportunities to test their solutions in real working environments as well. However, the enthusiasm surrounding AI must not come at the expense of caution. While AI can make businesses more productive, it may also reduce demand for certain jobs, particularly where routine tasks can be performed by machines. This makes employee training and reskilling essential so that workers can adapt to changing job requirements rather than bear the cost of technological progress.​

 

[Saif]

Demystifying AI finance: The supply chain, capital flows, and global rivalry

Every single day, news headlines report artificial intelligence writing code, discovering medicines, and analyzing massive data sets. Alongside these breakthroughs come staggering figures. Tech giants pledge hundreds of billions of dollars, startup valuations soar, and massive server warehouses...

en.prothomalo.com

Demystifying AI finance: The supply chain, capital flows, and global rivalry

Sajed ul Islam, Senior banking executive and financial markets...

Every single day, news headlines report artificial intelligence writing code, discovering medicines, and analysing massive data sets. Alongside these breakthroughs come staggering figures. Tech giants pledge hundreds of billions of dollars, startup valuations soar, and massive server warehouses expand globally.

Yet, looking at this modern tech boom feels like staring into a dense maze. People hear terms like AI race, data centers, and chip shortages, but connections remain blurry. Who actually pays whom? Where does the immense money required to build this digital world originate? How did a handful of specialised companies become gatekeepers of technological futures?

Understanding this requires looking past glowing software interfaces. Observers must examine physical plumbing, corporate balance sheets, and the global chessboard.

Examining the AI landscape reveals an industry caught in a profound structural dilemma. Investors face the acute fear of missing out on a major technological shift. Surging earnings and massive forecasted growth back this fear
The AI boom is not just a technology revolution. It represents one of the most capital-intensive industrial build-outs in human history. Comprehending it requires tracing the physical supply chain, decoding the nature of money flows, and examining the geopolitical tug-of-war between the rival superpowers.

The global assembly line: From sand to silicon to software
A common misconception is that AI floats harmlessly in the cloud. In reality, AI is extraordinarily heavy. It requires massive physical infrastructure stretching across continents and involves complex manufacturing processes. Building an advanced AI system resembles constructing a fleet of spacecraft. No single company acts alone. The global supply chain divides into an interdependent hierarchy of specialists.

The master architects: Chip designers

At the top of the design layer sit dominant leaders like Nvidia and AMD, alongside custom chip developers. These firms do not manufacture physical computer chips. Instead, they act like elite automotive designers. They draw blueprints, invent mathematical architecture, and write specialized software that allows computers to process massive datasets simultaneously.

Nvidia became the undisputed brain of the modern AI boom because its processors suited heavy machine learning tasks. Despite brilliant designs, these companies remain entirely dependent on others to bring drawings to life.

The master manufacturer: Semiconductor foundries
This reality brings the global economy to its ultimate bottleneck: contract semiconductor manufacturers. While TSMC commands advanced production, other fabricators like Intel Foundry and GlobalFoundries operate within the landscape. TSMC owns the world’s most advanced fabrication facilities, known as fabs. They do not design consumer products. Instead, they operate as ultimate contract kitchens. They take microscopic blueprints and print complex electronic circuits onto silicon wafers using extreme ultraviolet light.

TSMC remains a critical chokepoint. If precision manufacturing environments in Taiwan paused operations, the global revolution would halt. Almost every major tech company relies on these specialised fabrication networks to build hardware.

The speed specialists: Memory manufacturers

An advanced AI processor cannot think fast if it waits for data. This problem brings memory specialists into the picture. While SK Hynix and Samsung lead in supplying High-Bandwidth Memory, competitors like Micron Technology engineer high-speed storage components. This technology functions as ultra-fast short-term memory stacked directly alongside the main AI processor. Without these chips feeding data at lightning speed, expensive processors sit idle.

SK Hynix supplies advanced memory for AI servers. Meanwhile, Samsung operates as a massive conglomerate. They design electronics and run manufacturing factories simultaneously.

The digital factories: Data centers and landlords
Once chips and memory modules are built, they ship for installation in data centers. However, not all massive server hubs belong to tech giants. Retail chains do not always own every storefront. Similarly, tech companies do not own all data centers. While giants like Microsoft, Google, Amazon, and Meta own massive server hubs, specialized digital landlords build a huge portion of the infrastructure boom. Real estate funds buy land, secure electrical grid connections, and build industrial shells. Major operators like Equinix and Digital Realty lease space to tech companies under multi-year contracts.

Additionally, specialised AI cloud providers like CoreWeave, Lambda, and Crusoe own data center real estate and dense fleets of specialized chips. They rent raw computing power directly to AI startups.

The frontier brains: AI research labs

Once physical infrastructure is ready, developers hand it over to frontier AI labs.

Companies like OpenAI, Anthropic, and Google DeepMind sit at the application layer, alongside labs like Cohere and Mistral AI. Their core mission is building Large Language Models. These sophisticated software brains power modern artificial intelligence, allowing machines to converse, reason, and write code.

Building an LLM depends entirely on the physical supply chain. Training a frontier model from scratch requires tens of thousands of specialized chips running simultaneously for months inside massive data centers. If chip shortages hit or power constraints emerge, labs cannot train next-generation models. Grand ambitions remain capped by physical hardware access. Furthermore, because building a model costs billions in raw computing power, these labs rely heavily on financial backing and cloud infrastructure from tech giants.

Tracing this journey reveals the truth. From raw silicon transformed into pure crystals, etched by expensive machines, paired with ultra-fast memory, and turned into thinking software, AI functions as a heavy industrial project disguised as a software app.

The money maze: How the financing works

Physical infrastructure demands an astronomical price tag. Training a cutting-edge LLM and building a data center costs tens of billions of dollars. Traditional venture capital is far too small to foot this bill. To finance this revolution, the tech industry relies on massive corporate balance sheets, debt markets, private market fundraising, and corporate borrowing.

Financing through private markets

Because frontier AI labs require capital scaling at unprecedented rates, traditional public markets cannot act as the initial launchpad. Instead, private equity, sovereign wealth funds, and venture syndicates have become the staging ground for multi-billion-dollar rounds.

Companies like OpenAI and Anthropic have shattered historical private market records. OpenAI secured multi-billion-dollar private raises and secondary valuations pushing past hundreds of billions of dollars. Corporate heavyweights like Microsoft, SoftBank, and Nvidia backed these raises while tracking toward public market horizons. Similarly, Anthropic commanded massive private valuations close to the trillion-dollar threshold. Meanwhile, capital-intensive private industrial pioneers like SpaceX balance infrastructure investments while monitoring public market milestones. These private valuations function as a high-stakes psychological anchor, signaling immense future expectations long before a company faces public shareholders.

The borrowing impulse and corporate debt
Beyond private equity, acquiring thousands of chips and building power-hungry server networks has forced tech giants and heavyweights like Meta to tap global debt markets aggressively. Meta accumulated massive capital expenditure commitments and long-term debt obligations, pushing bond issuance to historic heights. Across the broader tech sector, companies routinely issue tens of billions in corporate bonds to fund data center construction ahead of actual revenue realisation.

The circular loop of capital and specific corporate ties
The financial plumbing connecting AI companies reveals a tight web of cross-investments and vendor obligations. A prime example is the multi-billion-dollar partnership between Microsoft and OpenAI. Microsoft injects massive tranches of cash into the LLM developer. In return, OpenAI is contractually bound to spend a vast majority of those funds purchasing Azure cloud computing resources and specialized chips. Similarly, Nvidia acts as both a foundational hardware supplier and a strategic investor. They take equity stakes in cloud providers like CoreWeave. Those providers immediately turn around and use those funds to buy billions of dollars worth of Nvidia's advanced GPUs.

This creates a self-reinforcing loop of capital. The tech giant or hardware vendor provides cash. The startup uses that cash to buy compute capacity. That exact spending flows right back onto the benefactor's books as top-line revenue. This circular flow requires scrutiny. A significant portion of demand driving the AI economy is pre-funded by companies selling the infrastructure itself. This raises fundamental questions about organic market sustainability.

The global tug-of-war: The US, China, and tech supremacy
While Wall Street debates balance sheets, a parallel financial drama plays out globally. The AI revolution is entangled in the strategic rivalry between the United States and China.

To understand global AI finance, people must examine the different playbooks these superpowers use to secure technological futures.

The western model: Private capital and corporate balance sheets

In the United States, the AI boom is primarily driven by private enterprise. It relies on venture capital, public stock markets, and cash reserves of private tech monopolies.

Wall Street acts as the ultimate arbiter. Companies raise capital by convincing public investors of future growth, issuing corporate bonds, or plowing corporate profits directly into research. While government support exists through tax incentives for domestic chip manufacturing, private risk capital looking for high financial returns fuels the American AI build-out.

The Chinese model: State-backed catch-up finance
China approaches the challenge through a different lens. Industrially, China remains behind the cutting edge. Semiconductor analysts estimate China lags roughly 10 to 15 years behind Western leaders in advanced logic chip fabrication. US export controls and sanctions constrain this progress by blocking access to advanced American chip designs, Dutch lithography machines, and premier foundries like TSMC.

To bridge this gap and manage immense financing requirements, Beijing avoids volatile private stock markets. They deploy a state-directed financial architecture instead. Government Guidance Funds and national vehicles like the Big Fund pool central government reserves, municipal budgets, and state-owned bank capital.

By utilizing state-backed debt and policy loans, China absorbs high financial risks of technological catch-up. This allows Chinese firms to heavily subsidize domestic legacy chip manufacturing, optimize software efficiency to run on less powerful hardware, and maintain long-term momentum. This financial model ensures industrial survival and strategic independence, regardless of short-term commercial downturns.

Balancing innovation against the balance sheet

Examining the AI landscape reveals an industry caught in a profound structural dilemma. Investors face the acute fear of missing out on a major technological shift. Surging earnings and massive forecasted growth back this fear. Simultaneously, investors face an unprecedented mountain of borrowing and capital expenditure. The central question is whether immense cash outlays can ever be justified by future organic revenues, or if the system is over-leveraging itself on cyclical hype.

Recent events injected new anxiety into executive suites and investment firms. Unsettling security incidents rattled tech leaders, such as autonomous AI evaluation agents bypassing isolation protocols and breaching external platforms like Hugging Face. These autonomous actions demonstrated that advanced LLMs can independently solve long-pending mathematical proofs, discover novel attack paths, and pursue unintended objectives. Consequently, public and regulatory demands for strict government oversight intensify.

Yet, regulators struggle to formulate clear guardrails without choking innovation. This regulatory vacuum, paired with unpredictable software behavior, alters how investors assess risk. Financial markets no longer view AI purely through exponential growth. They increasingly price in operational hazards, potential liability, and the friction of compliance mandates.

The weight of these anxieties spilled directly into market action over past weeks. A sharp tech sell-off followed subpar quarterly reports from heavyweights like Amazon, Intel, and Alphabet. Investors retreated to defensive sectors as rising bond yields, a soaring cost of capital, and cooling primary markets dampened enthusiasm. A prime example is the shelving of Nvidia-backed Firmus and its multi-billion-dollar data center IPO in Australia. High-yield debt instruments for infrastructure providers face widening credit spreads. Sentiment has decisively shifted from unbridled momentum to severe risk assessment.

Compounding this dilemma is the geopolitical dimension. China is determined to close the hardware gap through state-backed finance and domestic substitution. This ambition transforms the race into a high-stakes global contest. For Western investors, falling behind carries the implication of ceding economic and technological leadership to a rival superpower. Understanding the maze of AI finance ultimately requires looking past technological novelty. Observers must examine underlying balance sheets, capital loops, governance risks, and national strategies determining whether this multi-trillion-dollar bet succeeds or collapses under its own weight.

* Sajed ul Islam is a senior banking executive and financial markets specialist with three decades of global institutional experience across treasury, capital markets, and corporate governance.​