Bangladesh’s universities face major gaps in AI education, including outdated curricula, limited computing resources, inadequate faculty training and little formal instruction on AI ethics, according to a study examining AI readiness across the country’s higher education sector.

The ongoing qualitative study reviewed 35 computer science and related programmes and interviewed 59 students, educators and industry professionals to assess how universities are preparing students for an AI-driven workforce.

The researchers found that AI readiness in Bangladesh is constrained not only by a lack of hardware but also by institutional practices governing funding, curriculum development, mentorship and access to research opportunities.

The study noted that Bangladesh ranked 82nd in the 2023 Oxford Insights AI Readiness Index, reflecting strong policy ambitions alongside continuing weaknesses in technological capacity and research ecosystems.

35 programmes reviewed, 59 stakeholders interviewed

Researchers systematically reviewed AI-related course syllabi, university websites, faculty research profiles, event announcements and other institutional materials from 35 prominent computer science and related programmes.

The institutions covered both public and private universities with varying levels of funding and infrastructure.

The researchers also conducted semi-structured interviews with 59 participants. These included 25 students, nine educators and 25 industry professionals.

Students and educators were drawn from five major AI-teaching programmes at four universities, including Buet, the University of Dhaka’s Department of Computer Science and Engineering, the University of Dhaka’s Institute of Information Technology, North South University and BRAC University.

The researchers also examined faculty publication records and participation in major AI research venues to understand institutional differences in research capacity and opportunities.

Students know the theory but struggle with practice

A major finding was a disconnect between theoretical AI education and practical experience.

Students interviewed said they often learned mathematical foundations and theoretical concepts without getting sufficient opportunities to train models, work with real datasets, debug systems or deploy AI applications.

One student described completing undergraduate courses covering neural networks and backpropagation but said they had rarely trained models or worked with tools such as TensorFlow and PyTorch.

Industry participants reported seeing a similar gap among graduates, particularly in problem formulation and decision-making.

One industry professional said a graduate knew how to use AI tools but struggled to explain why AI was necessary for a particular problem.

According to the researchers, this suggests that students may graduate with technical vocabulary and theoretical knowledge but lack the practical intuition required to determine when and how AI should be applied.

Lack of computing resources remains a barrier

Limited access to high-performance computing also emerged as a major obstacle.

Students and faculty reported shortages of GPUs, outdated computers, insufficient storage and unreliable access to computing resources.

The researchers found that even comparatively well-resourced universities faced infrastructure constraints.

The problem, they argued, is not simply a question of buying more computers. Funding priorities, administrative decision-making, laboratory space and departmental politics also determine which students and researchers can access AI infrastructure.

One faculty member interviewed said limited funding for GPUs and outdated computers prevented students from working with modern deep-learning systems.

The study therefore characterises AI readiness as a “layered sociotechnical infrastructure”, where technological limitations are closely connected to institutional governance and resource allocation.

Informal networks shape access to AI opportunities

The study also found that access to mentorship, research groups and hands-on projects can depend heavily on informal networks.

Students reported that opportunities were sometimes influenced by GPA requirements, personal connections and timing rather than transparent recruitment processes.

This can leave students outside established networks with fewer opportunities to gain practical AI experience.

The researchers argue that such practices effectively “blackbox” AI education, making important learning opportunities less visible and less accessible.

They also found limited routine engagement between universities, alumni and industry, reducing students’ exposure to contemporary AI tools and workplace expectations.

AI ethics largely missing from curricula

Perhaps the most significant finding was the limited presence of Responsible AI education in formal university curricula.

The researchers found gaps in areas including AI ethics, privacy, fairness, human-centred design and responsible deployment when comparing Bangladeshi programmes with global reference programmes.

Faculty members reportedly recognised the need to introduce these topics but faced lengthy curriculum approval processes.

One faculty member said curriculum changes can require multiple approvals and take months or even years, meaning new content may become outdated before it is formally introduced.

Industry participants highlighted the potential consequences.

One senior industry professional described a case in which a technically functioning AI system used student data without proper consent and failed to account for fairness, ultimately disadvantaging rural students and exposing personal data.

The researchers argue that Responsible AI should therefore not be treated as an optional addition to technical education.

Policy ambitions not translating into classrooms

The study identifies a disconnect between Bangladesh’s national AI policy ambitions and what students experience in university classrooms.

Although Bangladesh has engaged with international AI frameworks and developed national strategies, the researchers found that these commitments have not necessarily translated into changes in curricula, teaching practices or student training.

The authors argue that this challenges the idea that AI readiness can be achieved primarily through policy development or investment in technology.

Instead, they propose understanding readiness as a combination of infrastructure, curriculum, faculty capability, mentorship, governance and industry engagement.

Researchers call for locally grounded AI education

The study recommends pilot Responsible AI courses, participatory curriculum development, stronger university-industry collaboration and greater investment in faculty training and computing infrastructure.

The researchers also argue that Bangladesh should avoid treating AI readiness simply as a race to match standards established elsewhere.

Instead, they call for locally grounded approaches that incorporate Bangladesh’s own educational, institutional and social conditions.

The study has several limitations. Its sample was primarily drawn from urban, research-oriented universities and industry practitioners, limiting the generalisability of the findings to regional and vocational institutions.

The researchers also noted that Bangladesh’s AI education and governance landscape is changing rapidly, meaning the findings represent a snapshot of a rapidly evolving environment.

They describe the work as an ongoing study and call for future comparative, longitudinal and intervention-based research to examine how AI education in Bangladesh develops over time.