THE Higher Education Commission issued its notification in February 2026 and confirmed it at its 47th Commission meeting in July: from this autumn, every student in a Pakistani university, whatever the discipline or institution, must complete three credit hours of artificial intelligence. The instinct is sound. Nations slow to build computer literacy in the 1990s spent two decades importing skills they could have developed at home. The difficulty lies below instinct, in the machinery that must carry instruction into classrooms.
A three-credit-hour course requires a competent instructor. In July 2025, the Ministry of Information Technology and Telecommunication, with the HEC, the National Computing Education Accreditation Council, Meta and atomcamp, announced a two-track programme: soft-skills training for more than a thousand non-technical faculty and hard-skills certification, including Meta’s Llama credentials through Coursera, for 250 to 500 technical faculty. This cohort was never meant to be the whole teaching pool. Pakistan already has computing departments and qualified staff. Yet much of that staff is concentrated in established departments and carries responsibility for existing programmes, while the mandate reaches every discipline in every university.
The task is measured not by institutions but by sections. The Economic Survey records higher-education enrolment at about 1.96 million. Across four-year degrees, annual intake runs to several hundred thousand students. At forty to a class, compliance requires roughly ten thousand sections each year. Against that requirement, the planned technical-certification cohort is only 250 to 500. Even if existing computing faculty carry most of the load, a shortfall on this scale is not one that programme was designed to close. The programme’s answer is a cascade: master trainers first, faculty workshops afterwards and national rollout later. But a cascade is a promise about years and the classroom opens in weeks.
A practical AI course also needs machines. Household internet access rose from 34 per cent in 2018–19 to about 69 per cent in 2024–25, while internet use among people aged ten and above increased from 17 to 57 per cent. Yet Gallup Pakistan’s analysis of HIES data records only 8.7 per cent of households as owning a desktop or laptop, rising to 10.1 per cent when tablets are included. Connectivity has doubled; productive devices reach only one household in ten. Household ownership does not directly measure university laboratories, but it determines whether students can practise beyond scheduled classes, repeat exercises independently or build the portfolios employers increasingly expect from graduates today. A Smartphone opens a chatbot, but it cannot substitute for sustained practical work. In under-resourced universities, AI may therefore be discussed rather than operated: students may learn the definition of machine learning without training a model.
The gap is unequal. Urban device ownership stands at 14 per cent and rural ownership at five. One curriculum will be delivered into two materially different countries. A graduate of a well-endowed Lahore university may finish with a portfolio; a student in Turbat, where the university cannot supply what the home does not, may finish with a notebook. The same credit hours will certify different competencies and the labour market will price the difference before the transcript acknowledges it.
The fiscal constraint is discussed least and binds hardest. The proposed National AI Ecosystem Development Programme carries a projected outlay of Rs283 billion between 2026 and 2031, but its initial proposed allocation for 2026–27 was only Rs185 million. In February, the Prime Minister separately committed the government to investing $1 billion in AI by 2030 and announced plans for AI curricula in schools. Unless a defined share reaches universities for laboratories, computing capacity and faculty development, the mandate will precede its means.
The regulatory framework carries another tension. The HEC’s 2025 draft framework for ethical generative-AI use in higher education calls for transparent attribution and holds authors accountable for submitted content. But attribution rules collapse against examinations that reward recall. A system measuring retrieval of memorized material is uniquely vulnerable to technology that retrieves it instantly. Universities will teach AI in one room and penalize its traces in another. The policy is not wrong; the examination beneath it is.
The strongest counterargument deserves a hearing. In Pakistan, phased rollout has repeatedly meant abandoned reform: capacity assessments become permanent, pilots continue without scaling and institutions most in need are served last. Waiting for readiness can mean waiting indefinitely, whereas a deadline forces vice-chancellors to find instructors, reallocate laboratories and request funds. Mandates have built capacity before. But a mandate builds capacity only where money follows notification. A deadline met by relabelling an existing course does not force capacity into being. It forces documentation.
The answer is not to withdraw the mandate but to reorder its parts. The faculty programme should become a national certification standard. No institution should offer the course with fewer certified instructors than its planned sections require—a ratio published, audited and enforced rather than assumed. Institutions should be assessed for laboratory space, bandwidth and faculty depth, with phased rollout permitted where capacity falls short.
Assessment must be rebuilt alongside instruction. A law student, agronomist and physician need different examinations: supervised analysis, viva voce and portfolio work in which AI use is declared rather than hidden. Only examinations rewarding judgment can sustain an attribution policy. Pakistan has adopted many curricula and implemented rather fewer. The distance between notification and classroom will decide this reform. Close it and a generation of graduates will understand how these systems work. Leave it open and they will be trained only to depend on them—and the transcript will not record the difference.
—The writer is Commoner from 44th Common Educationist — Founder of WHI Institute.based in Sargodha.
