What do Pakistani university teachers think about AI?
A peer-reviewed survey of 270 teachers at seven campuses of one private university found cautious optimism: 53.3% agreed AI could widen the digital divide, even as majorities saw benefits for teaching, assessment and interactivity. The convenience sample measures perceptions, not classroom results or national opinion.
The short version
- What happened
- The cross-sectional online survey gathered 270 valid responses from teachers at seven Riphah International University campuses between 1 February and 15 April 2026; recruitment used convenience sampling.
- Who it affects
- Teachers need practical support for evidence checking, privacy, assessment design and student disclosure—not a generic AI mandate.
- What remains unproven
- Convenience sampling from one private university cannot estimate national teacher opinion in Pakistan.
Keep in mind
Convenience sampling from one private university cannot estimate national teacher opinion in Pakistan.
The sample was 86.7% male, 94.4% MPhil-qualified and heavily early-career, limiting subgroup comparisons and generalisability.
Self-reported perceptions do not measure AI literacy, classroom use, student learning, workload or the actual digital divide.
The questionnaire had no pilot study, duplicated one interactivity item and has not undergone confirmatory factor analysis in an independent sample.
Cross-sectional associations cannot show whether AI experience caused any attitude or whether training would improve practice.
Source analysis · 1 linked source. Primary analysed source date: 9 October 2026.
Read the full report · 8 min The evidence, methods, implications and limitations
The direct answer: teachers saw useful tools and unequal conditions at the same time
Teachers in this seven-campus sample were moderately positive about AI in education, while also expressing concrete worries about access and professional consequences. On the study’s five-point scale, the overall composite mean was 3.40. The highest-rated statements concerned making teaching more interactive, student enjoyment and help with assessment. Yet 53.3% agreed or strongly agreed that AI integration could widen the digital divide. That is not a contradiction: a tool can appear educationally useful while the conditions for using it remain unequal.
The finding should not be promoted as a national verdict from Pakistani teachers. Every respondent worked for Riphah International University, a private institution, and recruitment was by convenience rather than probability sampling. The survey captures what 270 respondents said during February to April 2026. It does not show how often they used AI, whether their students learned more, whether marking improved, or whether infrastructure gaps actually widened after adoption.[1]
Who answered the survey
The researchers collected valid responses from 270 current teachers across seven campuses: 58 at the main I-14 campus in Islamabad, 45 at Raiwind in Lahore, 42 at Al-Mizan in Rawalpindi, 35 in Faisalabad, 32 in Malakand, 30 at Gulberg Green in Islamabad and 28 in Peshawar. Participants completed an online questionnaire distributed through institutional communication channels. Participation was voluntary, consent was collected electronically and the reported completion time was eight to ten minutes.
The sample was heavily male: 234 respondents, or 86.7%, compared with 36 women. Nearly half were under 29, and 142 respondents—52.6%—had two years or less teaching experience. Most held an MPhil qualification, while 15 held a PhD. Teaching levels ranged from undergraduate to doctoral programmes. Those details matter because a young, early-career and male-skewed group from one employer cannot stand in for the composition of university teaching across Pakistan.[1]
What the questionnaire measured—and what it did not
The 14-item instrument asked respondents to rate statements from strongly disagree to strongly agree. Ten items covered perceived benefits such as help with teaching and assessment, interactivity, interdisciplinary work, student enjoyment, inclusion and real-life applicability. Four negatively framed items covered opposition to compulsory AI instruction, restricting it to higher classes, the digital divide and replacement of human teachers. Two content experts reviewed the items, but the researchers did not run a pilot study before fielding them.
This is a perception instrument, not a test of AI literacy or teaching practice. Agreement that AI helps assessment does not mean respondents could identify hallucinated feedback, protect student data or design a valid AI-assisted task. Agreement that students enjoy AI does not establish engagement, learning or retention. The article’s proposed ‘dual-lens’ framework organises perceived benefit and risk as separate dimensions; it is a useful hypothesis, but the authors explicitly say it is not yet a formally validated structural model.[1]
The clearest descriptive results
For teaching support, 67.0% agreed or strongly agreed that AI helps in teaching; 65.2% said it helps assessment; and 68.2% agreed with the strongest of two similarly worded interactivity items. On the statement that students enjoy learning with AI, 66.3% agreed or strongly agreed. Those are substantial majorities within this sample, but the duplicated interactivity wording may inflate internal consistency and the authors acknowledge that one item should be rewritten in a future instrument.
Views on replacement were divided rather than uniformly alarmed. The item stating that mechanical teachers will replace human teachers received the lowest mean, 2.96: 35.2% agreed or strongly agreed, 35.6% disagreed or strongly disagreed and 29.3% were neutral. The digital-divide statement produced a clearer warning: 53.3% agreed or strongly agreed, 36.3% were neutral and 10.4% disagreed. For university leaders, the practical signal is that enthusiasm for classroom functions does not erase concern about who has devices, connectivity and support.[1]
Differences by age and teaching level were small
Composite perceptions differed statistically by age and level taught, but the reported effect sizes were small. The age-group analysis produced partial eta-squared of 0.046. Teachers under 29 had a mean of 3.29, compared with 3.56 among those aged 29 to 36 and 3.52 among those aged 43 to 50. Variances were unequal, the oldest group contained only nine people and multiple subgroup comparisons increase the chance of unstable findings, so the pattern should guide questions rather than determine policy.
Undergraduate instructors reported a mean composite score of 3.27, below 3.57 for MPhil instructors and 3.53 for PhD instructors; the level-taught effect size was 0.038. Gender, qualification and teaching experience were not statistically associated with the composite. Those non-significant results are not proof that the characteristics never matter: the female and PhD groups were small, the study was not nationally representative and its broad composite may hide differences in specific skills or concerns.[1]
Why the benefit–risk correlation needs careful language
Perceived benefit and perceived risk were positively correlated at r = 0.537. In plain language, respondents who endorsed more benefits also tended to endorse more risks. The authors present this as support for treating optimism and concern as coexisting rather than opposite. That interpretation is plausible and useful for training: teachers should not have to choose between being labelled pro-AI or anti-AI before discussing where a tool helps and where it creates new burdens.
The statistical result is still preliminary. Exploratory factor analysis suggested two factors explaining 63.6% of variance, but two items cross-loaded and the same sample was used both to build and explore the structure. No independent confirmatory sample tested whether the framework holds. Two interactivity items were identical, and self-report data can be shaped by social desirability. A correlation among questionnaire scores also says nothing causal about whether experience with useful AI makes teachers more alert to risk, or the reverse.[1]
What this means for teachers and institutions
The defensible response is targeted support, not a blanket mandate. Teachers working with undergraduate students and those early in their careers may need concrete examples tied to foundational courses: how to design an assignment that requires evidence checking, how to disclose permitted use, how to protect student information and how to assess learning when a model can draft an answer. Training should include supervised practice and accessible alternatives, then measure whether it changes confidence and teaching quality rather than treating attendance as success.
Infrastructure belongs inside the same decision. If a course assumes paid tools, fast connections or personal laptops, students without them may face a new barrier even when the lesson is well designed. Institutions should inventory device and connectivity access, provide supported options, publish privacy rules and offer human routes for feedback and appeal. The survey does not prove these interventions work; it explains why a rollout that celebrates features without measuring access would ignore the most widely endorsed risk in this sample.[1]
Funding, disclosure and what would change the assessment
The authors report no financial support for the work or publication. One co-author was employed by Saudi Sicli Company; the remaining authors declared no commercial or financial conflicts. The paper says Claude Opus 4 was used for literature synthesis and manuscript revision and Grammarly for language and readability, with authors reviewing the output. Riphah International University approved the research, participants consented electronically, and an anonymised dataset is provided as supplementary material, supporting scrutiny of the reported analysis.
Confidence would rise with a probability-based sample spanning public and private institutions, balanced participation across genders and regions, an independently validated questionnaire, and preregistered analysis. Longitudinal or intervention studies should then connect perceptions to observed practice: verified AI literacy, lesson design, workload, student attainment, academic integrity, privacy events and access by income, disability, language and location. Until then, the strongest conclusion is local and practical: these teachers could recognise benefits without treating the risks as abstract, and institutional policy should be capable of doing the same.[1]