The research team, which included collaborators from institutions in Japan, Germany, Ireland, Australia, the Netherlands and Finland, deliberately built its evidence base from across the full spectrum of education stakeholders. The seventy participants represented academia, research, industry, policymaking and frontline teaching, a composition designed to prevent any single perspective from dominating the outcome. The work was carried out as part of EDUsummIT 2025, an international summit held at the DCU Institute of Education in Ireland, where thematic working groups of researchers, policymakers and practitioners convened to develop strategies and action plans for technology integration in education. The group whose findings now appear in the journal focused specifically on how AI agents might be woven into educational ecosystems without eroding the human relationships at their core.

Methodologically, the study relied on a Delphi approach, a structured consensus technique long used in education and policy research to surface areas of agreement among experts who might otherwise never deliberate together. In a Delphi study, participants respond to successive rounds of questionnaires, and their feedback is aggregated and fed back to the panel so that positions can converge on the issues that matter most. This iterative process is particularly valuable for emerging technologies, where the evidence base is thin and expert judgment carries disproportionate weight. By applying the method to AI in education, the researchers were able to move beyond the polarized rhetoric that often surrounds the topic, in which AI is framed either as an existential threat to teachers or as a cure-all for educational stagnation, and instead identify concrete areas where coordinated action is both necessary and feasible.

The consensus process yielded five priorities, and their ordering is itself informative. Ranked highest was the development of AI-related competencies for learners and educators. The experts emphasized that competence in this context means far more than operational fluency with a chatbot or an automated grading tool. Genuine AI literacy requires the ability to critically evaluate what these systems produce, to recognize their limitations, and to reason about the ethical implications of delegating cognitive and administrative tasks to algorithms. In other words, the panel placed human understanding, not technical deployment, at the top of the agenda, a finding that directly challenges the tool-first mentality that has characterized much of the recent wave of educational technology adoption.

The second and third priorities address the two sides of the classroom relationship: how learning is assessed and what teachers do. Automated formative assessment and feedback was rated as a key priority because AI systems can, in principle, provide students with rapid, personalized responses to their work at a scale no human teacher could match. Formative assessment, the ongoing diagnostic feedback that helps learners understand their own progress, has always been constrained by teacher workload, and AI promises to relieve some of that constraint. Yet the study is careful to frame this as an augmentation rather than a substitution. The third priority, the evolving role of teachers working alongside AI agents, reflects the panel’s conviction that educators will remain central to learning. Rather than simply adopting whatever tools arrive, teachers may increasingly act as orchestrators of learning, combining the analytical capabilities of AI with professional judgment about what each student needs, when to intervene, and how to interpret machine-generated feedback in light of classroom context.

The remaining two priorities confront the structural risks that technological optimism tends to obscure. Explainability and trustworthiness emerged as a priority because AI systems in education must be transparent, reliable and fair if they are to be legitimately entrusted with decisions that affect learners’ futures. When an algorithm recommends a learning pathway, flags a student as at risk, or scores an essay, the reasoning behind that output matters, both pedagogically and ethically. Opaque systems undermine the trust that educational relationships depend on. The fifth priority, accessibility and equal access, addresses the distributional question head-on: students from different socioeconomic, geographic and linguistic backgrounds must have equitable opportunities to benefit from AI-supported learning. The experts’ warning is explicit in its implication, namely that technological progress alone does not guarantee better education, and that without deliberate attention to equity, AI could widen precisely the achievement gaps it is often marketed as closing.

One of the study’s most consequential arguments is that AI adoption should not be treated as a purely technological issue. Education, the researchers contend, is an interconnected ecosystem involving learners, educators, institutions, data, infrastructure, policies and society. Each of these elements shapes how an AI system performs in practice. A sophisticated tutoring agent is of little value in a school without reliable connectivity; a well-designed feedback tool can do harm if institutional policies give its outputs unwarranted authority; a national AI literacy curriculum will falter if teachers receive no professional development to support it. Framing education as an ecosystem shifts the unit of analysis from the individual tool to the network of relationships and resources in which that tool is embedded, and it implies that evaluation of AI in education must be similarly holistic.

From this ecosystem perspective follows the study’s central prescription: sustainable AI integration requires coordinated action across research, educational practice and policymaking. No single actor can deliver the five priorities alone. Researchers must generate evidence about what works, for whom and under what conditions, including rigorous studies of how automated feedback affects learning outcomes and how explainability requirements can be met without rendering systems unusable. Practitioners must translate that evidence into classroom routines that preserve teacher agency and student wellbeing. Policymakers must build the infrastructure, regulatory frameworks and funding mechanisms that make equitable access possible, from broadband provision to data protection rules to standards for algorithmic transparency in educational settings. The Delphi consensus suggests that experts across eighteen countries see this coordination, not any particular technology, as the real bottleneck.

The study’s conclusions point toward a future in which AI in education is built around human capability, pedagogical innovation, ethical governance and equity. That formulation is notable for what it excludes: nowhere does the panel position AI as a replacement for educators. Instead, the researchers describe AI as a tool to support teachers, empower learners and strengthen educational ecosystems. This human-centered framing carries practical weight. If the highest priority is competencies, then investment should flow to professional development and curriculum design before procurement. If teachers are orchestrators, then AI systems should be designed to make their judgment more informed, not to bypass it. If equity is a co-equal priority with innovation, then deployment decisions must be evaluated by their distributional effects, not only their average performance.

For education systems now drafting their first AI strategies, the study offers a checklist that is as much political as technical. The five priorities, AI competencies, automated formative assessment and feedback, evolving teacher roles, explainability and trustworthiness, and accessibility and equal access, function as a set of interlocking commitments: progress on any one of them depends on progress on the others. A school that deploys automated feedback without building AI literacy among its staff and students invites uncritical reliance on opaque outputs. A government that mandates AI tools without addressing unequal access risks entrenching disadvantage. The Kumamoto University-led team, drawing on the structured judgment of seventy experts across eighteen countries, has made the case that the transformation of education by AI will be judged not by the sophistication of its algorithms but by whether learners and educators, wherever they are, actually benefit from them. The roadmap is now on the table; the coordinated action it demands is the harder task ahead.

Subject of Research: Responsible integration of artificial intelligence into education based on an international expert consensus study

Article Title: How can AI transform education without leaving anyone behind?

Article References: How can AI transform education without leaving anyone behind?. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, education, Delphi study, AI literacy, formative assessment, teacher roles, explainable AI, digital equity, educational technology, Kumamoto University, EDUsummIT, education policy