← Cyprus & AI

Cyprus & AI · Research 03

Beyond the Enterprise Survey

AI in Cyprus's Education System: What the Evidence Says About Learning

Cyprus has documented AI use among teachers, introduced institutional guidance and examined selected AI-supported learning interventions. What do these findings establish about learning, and how far can they support conclusions about lasting skills?

As part of the Cyprus & AI research series, this article extends the investigation beyond the businesses covered by Europe's harmonised enterprise technology surveys. Education also raises questions about what people learn, how their abilities are assessed and which capabilities they may later bring to professional work.

The change of subject also changes the questions the evidence must answer. A survey can establish how teachers report using AI in their work. Institutional guidance can show how schools and universities intend to govern its use. Learning studies can examine what happens when students work with the technology under particular conditions. These sources describe connected developments, although they measure different things.

A study involving dental students at European University Cyprus provides an example. The group assigned to ChatGPT-supported preparation achieved a higher mean score on the immediate individual knowledge test than the literature-research group among the students who were tested. The researchers reported a short-term group difference, subject to methodological limitations including unequal test completion.

The experiment raises a question extending beyond that particular assignment. Which conclusions about learning can reasonably follow from the results that educational research measures? Several Cyprus-specific initiatives and educational studies now provide material for examining that question.

Where AI Is Already Visible

OECD TALIS 2024 provides a standardised international survey of self-reported AI use among lower-secondary teachers in Cyprus.

According to OECD TALIS 2024, 31% of lower-secondary teachers in Cyprus reported using artificial intelligence in their work. The figure covers AI broadly and reflects teachers' self-reported use, rather than a measure of generative AI adoption across the education system. Source: OECD TALIS 2024, Cyprus country note.

The population is important to the interpretation. The survey concerns teachers working at the lower-secondary level, a defined part of the education system. It offers no equivalent measure of every primary-school teacher, university lecturer or student in Cyprus.

AI Use Among Lower-Secondary Teachers in Cyprus

31%Reported using artificial intelligence in their work

Lower-secondary teachers · OECD TALIS 2024

Self-reported professional use of artificial intelligence, defined broadly, by teachers at ISCED level 2. The figure is not an adoption rate for all teachers, students or the education system as a whole. Source: OECD TALIS 2024, Cyprus country note.

Reported professional use also covers activities with different educational purposes. The OECD country note records reported uses including generating lesson plans and activities, and learning about or summarising topics. A common technology label does not mean that the activities have identical implications for learning.

The TALIS result consequently establishes a defined observation of professional AI use. The survey cannot determine from that figure alone how students encounter the technology or whether their educational outcomes have improved.

Understanding the institutional response requires a different form of evidence. Published guidance provides information about the responsibilities and expectations that authorities and universities have begun to establish.

How Institutions Are Responding

Cyprus has introduced national guidance for the use of AI in schools.

On 26 November 2025, Cyprus's Ministry of Education circulated national policy guidance on the responsible and ethical use of AI in primary and secondary education. Its publication establishes a formal policy framework, while leaving the extent of implementation in individual schools to be assessed separately. Source: Cyprus Ministry of Education, circular of 26 November 2025.

The publication addresses an institutional responsibility that extends beyond the decisions of individual teachers. Questions of appropriate use, educational integrity and the conditions under which AI may assist learning need to be considered at the level of educational policy.

The national guidance documents an official response. Its effects on teaching practices, assessment or educational outcomes would require evidence from implementation and evaluation.

TALIS collected the teacher responses in 2024, before Cyprus's Ministry of Education circulated its national school AI guidance on 26 November 2025. The later guidance therefore cannot account for the AI use reported in that earlier survey.

At two universities examined here, published policies describe formal approaches to AI use in higher education.

The University of Cyprus adopted AI guidelines through a Senate decision on 11 June 2025. The University of Nicosia has also published an institution-wide AI policy, referring to a Senate resolution from March 2023. Its policy commits the university to reviewing assessment methods and gives oral examinations, presentations and practical demonstrations as possible approaches. These documents establish institutional policy commitments, without showing how consistently they have been implemented or whether they have improved learning outcomes. Institutional sources: University of Cyprus AI Guidelines and University of Nicosia AI Policy.

The University of Nicosia's stated approach to assessment is relevant to the wider educational debate. Oral examinations and practical demonstrations may allow institutions to observe aspects of student performance under conditions that differ from those of a written assignment prepared with AI assistance.

Whether any particular assessment approach is effective depends on its design and intended purpose. Published institutional commitments do not establish how extensively revised assessments have been introduced or what their results may have been.

The school guidance and university policies document decisions about how AI should be governed. The available learning studies address a further question by examining outcomes in particular educational settings.

What Cyprus Classrooms Have Tested

The European AI4EDU project examined AI-supported education in classroom environments. Its published evaluation includes cases from Cyprus involving physics and chemistry instruction, alongside activities conducted elsewhere in Europe.

The AI4EDU project reported differing short-term results across a small number of Cyprus classroom pilots in physics and chemistry. The existing-class designs and differing assessments restrict the conclusions that can be drawn for Cyprus schools more widely. Source: AI4EDU, D6.2, sections 4.1.1–4.1.3.

The project illustrates a practical challenge in educational evaluation. Classrooms are existing social and instructional environments. Differences in pupils' prior knowledge, classroom arrangements, teaching support and other conditions may influence the results observed during an intervention.

The assessment methods also differ across the reviewed classroom cases. Some comparisons concern performance at the end of instruction, while others examine changes between earlier and later tests. Each design provides information about a particular outcome under the conditions in which it was measured.

A study reporting a statistically detectable difference offers evidence relevant to its own comparison. An inconclusive statistical test cannot establish that the intervention had no learning effect. Sample size, the variation in measured performance and the organisation of the classroom investigation all influence the interpretation.

The AI4EDU material also requires attention to the relationship between its publications. The gravitational-force intervention associated with Study Buddy belongs to the same underlying research family when it is described in related project materials. Repeated publication of that case does not establish an independent replication. Related study: Study Buddy, same AI4EDU gravitational-force case.

The chemistry pilot presents a separate reporting limitation. The participant population described in the project overview differs from the sample analysed in the reported results. Without a verified account of that difference, the two populations should remain distinct. Source: AI4EDU D6.2, pilot overview and chemistry case tables.

These constraints do not make the classroom observations irrelevant. The studies provide evidence about particular learning interventions and show how the conditions of an evaluation shape the conclusions that researchers can draw.

Their short instructional periods and selected assessments also leave further educational questions open. Understanding whether pupils retain knowledge after a substantial interval, or can apply it independently to a different problem, would require corresponding observations.

The classroom cases are most informative when treated as defined studies of specific educational activities. A university experiment offers a complementary example involving an individual knowledge examination after AI-supported work.

What a University Learning Trial Shows

Researchers at European University Cyprus investigated ChatGPT in a learning assignment involving second-year dental students. Their work examined how students performed on an individual test after preparing an assignment with different research methods.

In the study, the ChatGPT-supported learning activity involved group preparation. Students subsequently completed an individual examination at the beginning of the scheduled presentation session, after their assignments had been submitted. The examination therefore assessed knowledge at a particular point following collaborative work.

A study involving second-year dental students at European University Cyprus reported higher scores on an unannounced individual knowledge test after ChatGPT-supported group assignments. Of 77 students allocated to the two conditions, 70 completed the test. The ChatGPT group scored an average of 7.54 out of 10, compared with 6.94 in the literature-research group (reported p = 0.045). All seven missing test participants came from the comparison group, limiting confidence in the result. The study did not examine whether the knowledge was retained over time. Source: Kavadella et al., JMIR Medical Education (2024).

Immediate Individual Knowledge Test

European University Cyprus · Second-year dental students

77 students were allocated; 70 completed the test. All seven missing tests were in the comparison group. The test measured immediate knowledge, with no delayed follow-up.

Reported p = 0.045

Mean scores among the students tested, on a shared scale from 0 to 10. Source: Kavadella et al., JMIR Medical Education (2024).

The result concerns an individual knowledge test, making it more informative about immediate learning than an evaluation based solely on the quality of the groups' completed presentations.

The test took place after students had worked on the assignment and before the scheduled presentations. Their preparation was collaborative, while their test responses were individual. These conditions are relevant to the particular outcome that was measured.

The unequal completion of the examination also requires care. Because all seven missing tests were from the comparison group, the observed groups may differ from those originally allocated. The reported statistical result should therefore be interpreted alongside that asymmetry.

The study provides evidence from a specific university cohort, subject and educational activity. It offers no direct measurement of how long the demonstrated knowledge remained available or whether students could transfer it to substantially different situations.

An immediate knowledge result can nevertheless be educationally relevant. Its value depends on the learning question being investigated. If the aim is to assess understanding shortly after an assignment, an appropriately designed immediate test can provide meaningful evidence.

A claim about durable knowledge or independently available professional judgement would require further assessment. Those capabilities concern performance under conditions different from those measured in this particular experiment.

The dental study and the AI4EDU classroom cases consequently provide different forms of information about AI-supported learning. Interpreting them together requires closer attention to the educational outcomes they actually measure.

What the Results Can Tell Us About Learning

The reviewed studies examine learning in particular contexts, using assessments intended to capture specific results. The meaning of those results depends on the relationship between the intended educational objective and the outcome that was observed.

For the purposes of this analysis, five categories help distinguish the evidence.

AI-assisted output concerns work completed with the support of an AI system. A well-prepared presentation or a more comprehensive written explanation may be valuable educational work. Its quality alone cannot establish how much of the underlying knowledge the individual acquired.

Immediate learning concerns performance measured soon after an educational intervention. The dental study's individual examination is an example. Such a measure describes what students demonstrated at that moment, subject to the study design. Establishing a causal learning gain would require additional evidence.

Self-reported capabilities and AI literacy concern participants' reported knowledge, confidence, attitudes or experiences of working with AI. These observations can be relevant to educational aims, while remaining distinct from a direct demonstration of particular skills.

Independently demonstrated capability concerns performance under conditions in which the relevant AI assistance is unavailable. Such an assessment may be necessary when the educational objective includes knowledge or judgement that a learner should be able to apply without that assistance.

Retention and transfer concern what learners can demonstrate after time has passed and, where appropriate, how they perform when facing different tasks. These outcomes require assessments designed to investigate persistence and application beyond the original learning exercise.

The five categories answer different educational questions. They do not provide a universal ranking of what constitutes good learning.

Competent use of AI can itself be a legitimate educational objective. A student may need to learn how to question an AI-generated explanation, identify a weakness in its reasoning or use the technology responsibly in a professional task. The assessment of such capabilities should reflect the work the student is expected to perform.

Other educational objectives involve knowledge or judgement that must remain available when the technology is absent. An assessment intended to examine those capabilities requires suitable conditions for individual demonstration.

The Cyprus-specific classroom and university studies discussed here provide observations about short-term outcomes. Their designs do not establish the longer-term retention and transfer of independently demonstrated skills.

Among the publicly accessible Cyprus-specific studies examined for this article up to 7 November 2026, the evidence remains too limited to establish general conclusions about skills learners can independently demonstrate and retain after AI-supported instruction. This describes the scope of the reviewed research, not an absence of such effects or of unpublished evaluations. Research scope: AI4EDU original report and Kavadella et al.; bounded source selection through 7 November 2026.

This finding is an interpretation of a scoped research corpus. It cannot establish that AI-assisted learning fails to produce durable capabilities, and it does not imply that every educational intervention should be evaluated through the same measures.

The educational objective determines what a useful assessment should investigate. The examples from Cyprus show how the findings of particular studies become more informative when their outcome measures and methodological limits are kept visible.

What Better Evaluation Would Require

The research examined for this article includes official teacher-survey evidence, published institutional policies and selected studies of educational interventions. These sources make it possible to describe particular developments while recognising the limits of the conclusions they support.

The public sources reviewed for this article document national school AI guidance and policies at individual universities, alongside selected education initiatives. They provide a more limited basis for assessing longer-term, independently demonstrated learning outcomes. This is an assessment of the available public evidence; it does not rule out institutional evaluations that have not been published. Sources: Ministry of Education, University of Cyprus, University of Nicosia.

That assessment concerns the reviewed public material. It should not be interpreted as a comprehensive inventory of every school, university or training provider in Cyprus. The various education populations and institutional settings also require separate treatment.

A practical implication concerns how institutions design evaluations of AI-supported learning.

I would assess AI-supported learning through separate measures of work produced with AI, performance students can demonstrate unaided, and what they retain or transfer to new tasks after an interval. These measures address different educational questions and should be interpreted alongside the aims and conditions of each course.

The appropriate combination depends on the educational objective. A programme intended to teach responsible AI use needs an assessment of that ability. A programme making claims about lasting independent knowledge needs evidence collected under conditions that can support those claims.

For schools and universities, the choice of assessment affects how confidently the results of an AI initiative can be interpreted. For policymakers, the same principle applies when deciding what can be learned from published guidance, pilot projects and evaluations.

Employers have a related interest in what qualifications and observed performance reveal about people's capabilities. Where AI is used in professional work, the conditions under which relevant capabilities are developed and assessed may inform how educational achievement is interpreted.

The dentistry study offers a concrete illustration. Its higher immediate test score is worth reporting, together with the study conditions and the unequal number of completed examinations. Questions about how the same students would perform months later remain outside the observed results.

An institution wishing to establish that further outcome would need to investigate it. The strength of future conclusions about AI-supported education in Cyprus will depend on how closely evaluations match the knowledge and capabilities they are intended to assess.

Apparatus

Sources and research scope

The official statistical basis is the OECD's Results from TALIS 2024: Cyprus Country Note. National guidance is documented in the Ministry of Education's circular of 26 November 2025.

The institutional sources are the University of Cyprus AI Guidelines and the University of Nicosia's official AI policy.

The educational studies discussed are the AI4EDU Evaluation of Educational Impact Results, D6.2 and Kavadella et al., Evaluation of ChatGPT's Real-Life Implementation in Undergraduate Dental Education: Mixed Methods Study, published in JMIR Medical Education in 2024.

The article draws on publicly accessible Cyprus-specific evidence examined through 7 November 2026. The underlying research is a scoped review with documented exclusions. It is neither a comprehensive inventory of the education system nor a systematic review of all published and unpublished educational evaluations.

Correspondence

If this piece reflects a question you are weighing, a short note is a good place to begin.