The AI skills divide in Europe and Central Asia: Who benefits and who gets left behind

As generative artificial intelligence (AI) rapidly enters classrooms across the Balkans and Türkiye, the policy conversation often focuses on tools—chatbots, adaptive platforms, automated grading systems.

Yet, emerging evidence from the World Bank-led Education AI Readiness Assessment — deployed in a first set of countries in Europe and Central Asia (ECA)–Bulgaria, Romania, and Türkiye — points to a more fundamental insight: The future of AI in education will be shaped less by access to technology and more by how education systems build the right skills to use it. At its core, AI is not skill-neutral. It rewards those who can use it critically and responsibly, and risks widening gaps for those who cannot.

AI is reshaping what it means to be “skilled,” but foundational skills remain essential

Across all three countries, AI is already transforming how students learn and how teachers teach. Students increasingly use AI as a personalized tutor, a homework assistant, and even a career advisor. On the ground, this is visible in the uptake of localized platforms:

• Turkish students are actively using the state-backed EBA Student Assistant for personalized remediation, while the AI Teacher Assistant provides integrated translation support for over one million school-age Syrian refugee children.
• In Bulgaria, students are using Ucha.se, a platform featuring over 27,000 gamified, curriculum-aligned video lessons and interactive quizzes, as well as BgGPT — the first large-scale generative AI model trained entirely on the Bulgarian language, removing the English-language barrier for local learners.
• Romanian students are using learning assistants like SARO to prepare for the high-stakes Baccalaureate exam, boosting homework completion rates by 28%.

Teachers too rely on AI to design lesson plans, generate content, and analyze student performance. This growing diversity of uses redefines what skills matter. For students, basic digital literacy is no longer sufficient: They must develop critical thinking, information literacy, and self-directed learning. Research suggests that students with strong foundational cognitive skills use AI more effectively, while those who treat it as a crutch risk falling behind (Oakley et al., forthcoming). For teachers, the bar has also risen — they must integrate AI meaningfully into pedagogy, design tasks that foster higher-order thinking, interpret AI-generated data on student learning, and navigate ethical issues such as data privacy and algorithmic bias.

The tools are arriving faster than the skills to use them

All three countries have made major investments in infrastructure, and their policy environments are increasingly ambitious. National strategies explicitly prioritize digital transformation and ethical AI use, including:

• Bulgaria’s National AI Concept (2020-2030),
• Romania’s National AI Strategy (2024–2027), and
• Türkiye’s AI in Education Policy Document and Action Plan (2025–2029)

Importantly, there is significant teacher interest. For instance, in Bulgaria over 70 percent of teachers are aware of AI tools, and around half have already experimented with them. This enthusiasm has driven high participation in national upskilling programs: Türkiye has trained over 157,000 educators via its centralized Teacher Information Network (ÖBA), while Romania has reskilled nearly 83,000 teachers—nearly half its active teaching workforce—through its National Recovery and Resilience Plan-funded “Digital Pedagogy” initiative.

Yet this momentum is constrained by a key challenge: insufficient human capacity. Teacher training in AI remains uneven and often theoretical rather than practical, and even where teachers are motivated, they lack systematic support to integrate AI effectively. Evidence from pre-service training programs in Türkiye suggests that future teachers’ confidence in using AI depends on their ability to understand and operate the technology, yet operating AI is only the first hurdle. The more decisive bottleneck is whether teachers can use these tools to develop students’ higher-order cognitive skills, such as critical thinking, and reshape their pedagogy accordingly. Similarly, digital and AI-related skills among students vary widely, with pronounced disparities between urban and rural areas and across socioeconomic groups, raising the risk that AI could reinforce existing inequalities.

AI can polarize further the distribution of skills

Looking across these three countries, consistent patterns emerge. AI tends to amplify existing inequalities in skills. Students who already have strong digital and critical thinking skills benefit more from AI tools, while others risk falling further behind.

Teachers are the pivotal actors. Their ability to guide, structure, and contextualize AI use determines whether it enhances or undermines learning. Where teachers are well-prepared, AI can support personalized, engaging, and inclusive education. Where they are not, the same tools can encourage superficial learning or academic dishonesty.

Meanwhile, systems are moving faster on technology deployment than on human capacity building: Infrastructure and policies are advancing rapidly, but teacher training and curriculum reform are lagging behind. Compounding this, national examinations in all three countries rarely assess the higher-order skills — critical thinking, information literacy, ethical reasoning — that AI use demands, reducing schools’ incentive to integrate AI meaningfully.

What this means for the future

The success of AI in education will depend on rebalancing investments toward skills development. But just as decisive as what we do is how fast we do it, and whether we have the capacity to move at scale. Priority actions should include:

• Embedding AI literacy across school curricula;
• Providing continuous, hands-on teacher training (pre-service and in-service) so educators can not only operate AI tools but use them to cultivate higher-order thinking;
• Designing curricula and assessments that explicitly measure critical thinking and agency among students; and
• Addressing inequities in access to both technology and skills.

Yet even these actions will only deliver results if two further enabling conditions are in place. First, governments must invest in building rigorous evidence before committing to national rollouts — through randomized controlled trials or equivalent designs that can distinguish what genuinely improves learning from what merely looks promising. Second, no education system can navigate this transformation alone: Structured public-private partnerships are essential to ensure that EdTech innovation serves pedagogical need rather than commercial interest. Türkiye’s ETKİM hub offers one model worth studying, bringing together government, industry, and educators to co-design solutions.

AI in education is not a guaranteed path to better learning. Deployed without the right conditions, it can widen the gap between advantaged and disadvantaged students, erode the deep-thinking skills it was meant to strengthen, and produce an illusion of learning — where students generate answers without ever building understanding. True readiness, then, is not a race to deploy more tools. It is the harder, slower work of building the institutional capacity to monitor what AI is actually doing in classrooms, track unintended consequences as they emerge, and take proactive corrective action before inequalities harden. Without that capacity, AI in education is less a solution than a very expensive gamble.

Source : World Bank