A Methodology for Addressing Ethical Challenges in AI-Enabled Online Learning
V. Zalizko
International Innovation Centr for Artificial Intelligence, ETH
https://orcid.org/0000-0001-5362-8270
Abstract
The rapid adoption of artificial intelligence (AI) and large-scale digital platforms in online educational settings has exacerbated long-standing ethical issues concerning learner autonomy, privacy, equity and well-being. Drawing on two key areas of research — Zawacki-Richter et al.’s (2019) systematic review of AI in higher education, and Akgun and Greenhow’s (2021) analysis of ethical issues in primary and secondary education — this article presents a practical approach to mitigating ethical risks in innovative international online schools. Particular emphasis is placed on preventing bullying, ensuring inclusive access for learners with physical impairments, making algorithmic decision-making transparent, and implementing educator-centred governance. The proposed framework integrates pedagogical oversight, technological safeguards, institutional policy and continuous ethical auditing, ensuring that AI-enhanced online education promotes human-centred learning and does not reproduce social or structural harm.
Based on foundational research in artificial intelligence in education (AIEd), the article builds on existing ethical frameworks by introducing a distributed human–AI assistance model tailored to international innovative online schools. In this model, each learner is supported by a personalised AI learning assistant, while each teacher is supported by two specialised AI agents and a human assistant tutor. We argue that this configuration mitigates ethical risks such as surveillance, bias and learner isolation, while strengthening inclusion, anti-bullying mechanisms, pedagogical quality and educator agency. The proposed methodology reframes AI as an ethical infrastructure that redistributes cognitive and emotional labour in online learning environments, rather than as a replacement for human roles.
Keywords: artificial intelligence in education, ethical issues, prospects, MRIA application
1. Introduction
Online schooling has evolved from an emergency substitute to a stable educational approach in many countries, with increasing support from AI-driven systems for tutoring, assessment, analytics and student support. While these tools offer personalisation and scalability, concerns have been raised about surveillance, bias, reduced teacher autonomy, and psychosocial risks. In their large-scale systematic review of 146 studies, Zawacki-Richter et al. (2019) highlighted that much AI-driven education (AIEd) research has historically focused on technical innovation rather than pedagogical or ethical implications, and that educators remain underrepresented in design processes. In contrast, Akgun and Greenhow (2021) focused on K-12 schooling, explicitly foregrounding ethical tensions such as data protection, fairness, and the developmental vulnerability of minors, and calling for normative frameworks adapted to school contexts. This article synthesises insights from these two traditions and situates them within the operational reality of an innovative international online school, where cross-border regulation, cultural diversity and student well-being are central concerns. Early AIEd systems were largely tool-oriented, comprising intelligent tutoring systems, automated grading and learning analytics dashboards. As documented by Zawacki-Richter et al. (2019), these systems prioritised technical feasibility over pedagogical and ethical integration, often marginalising educators in decision-making processes.
Recent scholarship emphasizes that ethical governance must be embedded directly into the architecture of AI-supported online schooling rather than treated as an external regulatory layer. Systematic reviews in AIEd reveal that early deployments privileged technical optimization over pedagogical and ethical reflection, often marginalizing educators in design processes (Zawacki-Richter et al., 2019; Williamson & Eynon, 2020). In K–12 contexts, concerns surrounding privacy, fairness, learner vulnerability, and well-being have been foregrounded, prompting calls for transparent, human-in-the-loop systems (Akgun & Greenhow, 2021; UNESCO, 2021). Complementary work has highlighted the need to mitigate algorithmic bias and to ensure explainability in automated assessment and tutoring systems (Holmes et al., 2019; Yadav & Chakraborty, 2020).
Ukrainian scholars similarly stress that large-scale digitalization in schooling requires principled ethical oversight, particularly regarding surveillance, equity, and data protection (Богатирьова, 2021; Шевчук, 2020). Studies on algorithmic grading and digital assistants further underline the necessity of transparency, proportional data use, and teacher mediation in distance learning environments (Коваленко & Панченко, 2022; Кириченко, 2023). Together with international conceptual frameworks advocating educator-centered AI ecosystems (Luckin et al., 2016; Zawacki-Richter et al., 2020), this body of literature supports the development of distributed human–AI support models in which personalized student agents, teacher-facing analytic systems, and human tutors jointly uphold psychological safety, inclusion, and pedagogical authority.
However, recent advances in large language models (LLMs), multimodal AI, and agent-based architectures are enabling a transition towards more personalised, dialogical and role-specific AI agents. This shift is particularly relevant in K–12 online education, where ethical concerns such as developmental vulnerability, well-being, and fairness are amplified (Akgun & Greenhow, 2021). We argue that ethical challenges in online learning should be addressed by structuring AI roles, rather than by limiting AI capabilities. We propose that ethical challenges in online learning are best addressed not by limiting AI capabilities, but by structuring AI roles explicitly within a transparent human–AI division of labor.
2. Theoretical Foundations for Ethical AI in Online Schooling
Artificial Intelligence in Education and the Centrality of Pedagogy
Research in artificial intelligence in education (AIEd) has historically focused on the technical capabilities of intelligent tutoring systems, learning analytics, and adaptive platforms. However, large-scale reviews demonstrate that pedagogical theory and educator participation have frequently been secondary considerations. Zawacki-Richter et al. (2019) analyzed 146 studies spanning two decades and concluded that teachers were rarely positioned as central actors in the design and governance of AI systems. This imbalance has raised concerns that automation could erode professional judgment and reconfigure schooling around efficiency rather than educational values. Subsequent work has sought to redress this orientation by situating AI within sociotechnical systems shaped by institutional norms, cultural expectations, and policy regimes (Williamson & Eynon, 2020). From this perspective, educational AI is not a neutral instrument but an infrastructural force that redistributes authority, visibility, and responsibility within digital classrooms. Ethical evaluation therefore requires attention not only to algorithmic performance but also to how AI restructures teacher–student relationships, assessment regimes, and learning trajectories. Luckin et al. (2016) and Holmes et al. (2019) further articulate a human-centered paradigm in which AI functions as a cognitive amplifier for learners and teachers rather than an autonomous decision-maker. Their frameworks emphasize transparency, teacher mediation, and learner agency as core design principles, anticipating later regulatory and ethical guidelines.
Ethical Challenges in K–12 Online Education
While early AIEd research was dominated by higher-education contexts, recent scholarship has increasingly addressed the distinctive ethical sensitivities of school-aged learners. Akgun and Greenhow (2021) identify privacy, data governance, algorithmic bias, and the psychological vulnerability of minors as central ethical fault lines in K–12 environments. They argue that educational AI must be governed by developmental considerations, proportional data use, and mechanisms that allow children and parents to contest automated inferences. UNESCO’s (2021) global recommendation on AI ethics reinforces these concerns, advocating human oversight, accountability, inclusiveness, and the protection of children’s rights as non-negotiable normative standards. In online schools that operate across jurisdictions, these principles are particularly salient, as data flows and regulatory obligations extend beyond national boundaries. Algorithmic bias constitutes a recurring theme in this literature. Educational systems trained on narrow datasets may reproduce linguistic, cultural, or socioeconomic inequalities unless continuous auditing and representational diversity are ensured (Yadav & Chakraborty, 2020). Ethical deployment therefore demands dynamic monitoring processes rather than one-time compliance checks.
Ukrainian Perspectives on Digitalization and Ethical Governance
Parallel debates have emerged within Ukrainian educational research communities, particularly in response to rapid digital transformation and the expansion of distance learning. Богатирьова (2021) emphasizes that digital schooling environments intensify ethical dilemmas surrounding surveillance, consent, and equity, calling for institutional ethics frameworks and teacher training in digital responsibility. Шевчук (2020) similarly frames AI in schooling as a dual-use technology whose pedagogical benefits are inseparable from risks of exclusion and depersonalization if governance structures are weak. Studies focusing on automated assessment highlight transparency and contestability as foundational ethical requirements. Коваленко and Панченко (2022) argue that algorithmic grading systems must remain subordinate to human evaluative authority and that students should be informed about how automated scores are generated. Кириченко (2023) extends this argument to conversational agents and digital assistants, warning that persistent individualized tutors may create asymmetries of information and dependency unless students retain control over data retention and interaction modalities. These national contributions resonate strongly with international calls for educator-centered AIEd ecosystems, underscoring that ethical tensions are not geographically bounded but structurally embedded in digital schooling.
2.4 Toward Distributed Human–AI Governance Models
Across these strands of literature, a convergent theoretical position emerges: ethical AI in education requires institutionalized human oversight, transparency of algorithmic operations, and explicit allocation of responsibility. Zawacki-Richter et al. (2020) and Holmes et al. (2019) advocate moving from ad hoc adoption toward systemic governance frameworks that incorporate ethics committees, professional development, and continuous auditing. Synthesizing these perspectives, this article adopts a distributed human–AI governance approach in which multiple specialized AI agents operate within clearly delimited roles and under human supervision. Such architectures respond directly to the sociotechnical critique advanced by Williamson and Eynon (2020), ensuring that AI systems are embedded in normative structures rather than functioning as opaque infrastructural backbones. In theoretical terms, the proposed model draws on:
– human-in-the-loop decision theory, preserving educator authority;
– preventive ethics, using AI for early detection of harm rather than punitive control;
– inclusive design theory, ensuring accessibility and dignity for learners with physical or sensory impairments;
institutional accountability frameworks, aligning technological practice with regulatory and moral obligations.
Together, these theoretical foundations justify the move toward agent-based educational ecosystems as a principled evolution of online schooling rather than a purely technological escalation.
3. Results
Each learner is assigned a personal AI learning assistant that operates as a private, student-centered agent with the following ethically grounded functions: adaptive explanation of learning materials aligned with the teacher’s curriculum; formative feedback without public ranking or comparative labeling; emotional and motivational support signals (e.g., detecting disengagement, stress indicators); mediation between the learner and the digital environment (e.g., accessibility adaptation).
Crucially, this assistant does not evaluate or discipline the student autonomously. All high-stakes decisions remain under human oversight, ensuring compliance with the principle of human-in-the-loop governance. From an ethical standpoint, this design: reduces public exposure and peer comparison, a known trigger for online bullying; supports learners with physiological or neurological challenges by enabling discreet accommodation; strengthens learner autonomy through explainable and controllable AI behavior.
To counteract the risk of teacher marginalization identified in the literature, the proposed model reinforces educator agency through role specialization: Supports the teacher by: analyzing class-level learning patterns; suggesting differentiated instructional strategies; flagging conceptual bottlenecks without prescribing actions. Operates under strict policy constraints to: detect potential bullying, exclusion, or harmful interaction patterns; monitor workload balance and cognitive overload risks; generate alerts, never sanctions. A trained educational professional who: provides socio-emotional support; mediates between AI insights and human judgment; ensures cultural and contextual sensitivity in international settings. This triadic support structure ensures that AI augments, rather than replaces, professional teaching practice, while preserving accountability and ethical responsibility.
Ethical Justification of the Multi-Agent Model
The proposed architecture addresses core ethical challenges identified in AIEd research: Transparency: clearly defined AI roles reduce opacity and algorithmic overreach. Fairness: individualized support minimizes bias amplification inherent in cohort-based analytics. Psychological Safety: proactive, private interventions replace punitive or public corrective mechanisms. Inclusion: adaptive interfaces enable equitable participation for students with physical, sensory, or health-related constraints. Educator Centrality: teachers remain epistemic authorities, supported by AI rather than subordinated to it.
In this sense, AI becomes an ethical scaffold—a structural condition for humane online education.
AI as a Preventive Mechanism Against Bullying
Unlike traditional moderation systems that react after harm occurs, agent-based AI enables preventive ethics: early detection of exclusionary discourse; private redirection of harmful behavior; escalation exclusively to human staff.
Empirical insights from K–12 ethics research suggest that such preventive, non-stigmatizing approaches reduce conflict while preserving student dignity (Akgun & Greenhow, 2021). For students who have experienced bullying in physical schools, ethically designed online environments may thus represent a safer educational alternative. Comparative Advantages of the Proposed Model
Table 1. Ethical and Pedagogical Advantages of the Distributed Human–AI Support Model
| Dimension | Traditional Online Schooling | Proposed AI-Agent Model |
|---|---|---|
| Student Support | Generic tools, limited personalization | One personal AI assistant per learner |
| Teacher Role | High workload, limited analytics | Two specialized AI agents + human tutor |
| Bullying Prevention | Reactive reporting | Preventive, AI-supported detection |
| Inclusion | Manual accommodations | Adaptive, discreet personalization |
| Ethical Oversight | Policy-based, static | Continuous, agent-supported auditing |
| Transparency | Opaque algorithms | Explicit role separation and explainability |
| Learner Autonomy | Platform-driven | Student-controlled AI interaction |
Methodological Framework and Future Strategies for Ethical AI-Enabled Online Schooling: The MRIIA+ Ecosystem
Among the many educational ecosystems, AI MRIA+ is worth highlighting. The MRIIA+ application is conceptualized as a flagship AI-driven educational assistant embedded within the digital infrastructure of international innovative schools operating through platforms such as mriia-school.com and https://mriia.school-top.com. Rather than functioning as a generic tutoring tool, MRIIA+ is designed according to an ethics-by-design methodology in which normative principles—privacy, fairness, transparency, human oversight, and child protection—are integrated throughout the technological lifecycle.
This orientation operationalizes recommendations advanced in the AIEd literature (Luckin et al., 2016; Holmes et al., 2019; UNESCO, 2021) and responds directly to concerns that educational AI systems frequently prioritize technical efficiency over pedagogical governance (Zawacki-Richter et al., 2019). Within MRIIA+, ethical reflection is institutionalized through participatory system specification involving teachers, psychologists, parents, and legal experts affiliated with the international school network hosted at mriia-school.com. Methodologically, MRIIA+ adopts continuous socio-technical risk assessment cycles in which learning analytics pipelines, conversational agents, and monitoring modules are periodically evaluated for bias, pedagogical alignment, and unintended behavioral consequences.
Distributed Human–AI Support as a Platform Principle
At the core of MRIIA+ lies a distributed agent architecture that reconfigures ethical responsibility through explicit role separation: each student receives a personal MRIIA+ learning assistant responsible for adaptive explanation, formative feedback, motivational scaffolding, and accessibility mediation; each teacher is supported by two MRIIA+ professional agents—a pedagogical analytics agent and a well-being monitoring agent; a human assistant tutor mediates between algorithmic insights and contextual educational judgment.
This triadic configuration is embedded directly into the learning environments deployed through https://mriia.school-top.com, ensuring that automation does not collapse into centralized algorithmic authority. Instead, MRIIA+ functions as a methodological instrument for ethical governance: tutoring, assessment support, and safeguarding analytics are intentionally separated to prevent conflicts of interest and excessive surveillance. From a theoretical perspective, such decomposition aligns with sociotechnical systems theory and human-in-the-loop governance models (Williamson & Eynon, 2020; Akgun & Greenhow, 2021). Teachers remain epistemic authorities, while MRIIA+ augments professional practice through interpretive analytics rather than prescriptive control.
Data Governance and Accountability in the MRIIA+ Platform
Operating across jurisdictions through infrastructures associated with mriia-school.com, MRIIA+ must comply with heterogeneous regulatory environments, including European data-protection regimes and child-rights frameworks. Accordingly, the methodological core of the platform incorporates advanced privacy engineering and algorithmic accountability mechanisms. These include: federated learning architectures that minimize central data aggregation; granular consent dashboards for parents and learners; cryptographically secured audit logs for all high-impact AI interventions; pre-deployment algorithmic impact assessments; bias-testing protocols embedded in routine system updates.
Such technical–institutional hybrids translate the ethical imperatives articulated in K–12 scholarship (Akgun & Greenhow, 2021) and UNESCO’s global recommendations (2021) into enforceable operational standards within MRIIA+ deployments hosted on https://mriia.school-top.com.
Preventive Ethics, Well-Being Analytics, and Cyberbullying Mitigation A defining feature of MRIIA+ is its orientation toward preventive digital ethics. Rather than reacting only after harm occurs, the platform integrates early-warning analytics for disengagement, social isolation, and emerging harassment patterns.
Within the moderated digital classrooms of mriia-school.com, MRIIA+ performs longitudinal sentiment and interaction-pattern analysis while adhering to strict proportionality principles: alerts are visible exclusively to designated human staff; no automated sanctions are issued; escalation pathways are governed by child-protection protocols; false-positive and disparate-impact metrics are routinely audited. This preventive architecture reflects current ethical scholarship emphasizing dignity-preserving intervention strategies for minors in AI-mediated learning environments (Akgun & Greenhow, 2021; UNESCO, 2021) and positions MRIIA+ as an infrastructural safeguard rather than a disciplinary mechanism.
4. Strategic Trajectories for MRIIA+ and International AI-Driven Schooling
Longitudinal Evidence Generation and Platform Scaling
A central future strategy for MRIIA+ involves systematic empirical validation across the international school ecosystem represented by mriia-school.com and https://mriia.school-top.com. Mixed-method longitudinal studies will examine academic attainment, psychosocial safety, accessibility uptake, and teacher workload redistribution. Core evaluation indicators include: well-being and belonging indices; cyberbullying incidence rates; accessibility utilization patterns; explainability comprehension among students and parents; teacher trust in AI-mediated recommendations.
Such evidence-based scaling aligns MRIIA+ with contemporary demands for accountability in educational innovation. Given its cross-border deployment, MRIIA+ is envisioned as a participant in regulatory sandbox initiatives co-developed with ministries of education and data-protection authorities. Permanent ethics councils, public transparency reports, and algorithmic governance charters will be institutionalized across the MRIIA+ school network. Strategically, this positions the platform as not merely compliant but as a co-architect of emerging educational AI norms.
Teacher Professionalization and AI Literacy in the MRIIA+ Ecosystem
Teacher empowerment remains a strategic pillar. MRIIA+ will support structured professional-development programs in algorithmic literacy, bias interpretation, and ethical decision-making, integrated directly into educator dashboards. Future curricular modules include: data-interpretation workshops; participatory co-design laboratories; scenario-based ethics simulations; cross-school knowledge exchanges within the MRIIA+ network. The long-term horizon of MRIIA+ is the cultivation of collective intelligence ecosystems linking learners, educators, tutors, psychologists, and AI agents into collaborative epistemic communities. Open benchmarking repositories, fairness dashboards, and participatory governance portals will allow stakeholders to continuously shape the platform’s evolution.
Concluding Perspective
Reframed through the MRIIA+ application, the methodological and strategic framework advances beyond abstract governance toward a deployable institutional architecture for ethical AI in online schooling. Embedded within the international educational infrastructures of mriia-school.com and https://mriia.school-top.com, MRIIA+ exemplifies how AI assistants can function as protective, inclusive, and pedagogically aligned companions rather than opaque managerial systems. By uniting distributed agent design, preventive ethics, accessibility engineering, and regulatory co-evolution, the platform positions itself as a research-informed model for the next generation of trustworthy digital education systems.
Nevertheless, the deployment of pervasive AI assistants in school contexts inevitably raises substantive ethical challenges. These include the risk of excessive surveillance through continuous interaction monitoring, potential over-reliance of learners on automated guidance, algorithmic bias affecting students from linguistic or cultural minorities, and the opacity of large-scale decision-support models. MRIIA+ addresses these risks through strict role separation between tutoring, safeguarding, and analytics agents; data-minimization and federated-learning architectures; mandatory human-in-the-loop approval for all high-stakes educational decisions; and explainability interfaces for students, parents, and teachers. Periodic algorithmic audits, external ethics committees, and regulatory sandbox participation further institutionalize accountability, ensuring that technological innovation remains subordinate to educational values and child-protection principles. At the same time, the MRIIA+ framework introduces structural advantages that extend beyond risk mitigation. Personalized AI companions enable sustained formative feedback and discreet accessibility accommodations, thereby strengthening inclusion for learners with physiological limitations or chronic health conditions. Teacher-facing analytic agents reduce administrative burden and surface pedagogically meaningful patterns without displacing professional judgment, while preventive well-being monitoring contributes to safer digital communities and reduced exposure to cyberbullying. Collectively, these features transform AI from a managerial instrument into an enabling infrastructure that supports equity, psychological safety, and instructional quality—offering international online schools a scalable yet ethically grounded pathway toward digitally mediated education systems worthy of public trust.
References
Akgun, S., & Greenhow, C. (2021). Artificial intelligence in education: addressing ethical challenges in K-12 settings. AI and Ethics, 1, 431–440. https://doi.org/10.1007/s43681-021-00096-7
Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education: Where are the educators? International Journal of Educational Technology in Higher Education, 16(39). https://doi.org/10.1186/s41239-019-0171-0
Holmes, W., Bialik, M., & Fadel, C. (2019). Artificial Intelligence in Education: Promises and Implications for Teaching and Learning. Boston: Center for Curriculum Redesign.
Luckin, R., Holmes, W., Griffiths, M., & Forcier, L. B. (2016). Intelligence Unleashed: An Argument for AI in Education. Pearson.
Williamson, B., & Eynon, R. (2020). Historical and social contexts of AI in education. AI & Society, 35, 611–620.
Zawacki-Richter, O., et al. (2020). Ethics and AI in Higher Education: A Theoretical Framework. Educational Technology Research and Development.
Богатирьова, О. М. (2021). Етичні виклики впровадження цифрових технологій у середній освіті. Педагогічні студії, № 45, с. 87–95.
Шевчук, І. В. (2020). Штучний інтелект у шкільному навчанні: ризики та педагогічні можливості. Освітні технології та суспільство, 22(3), 102–110.
Коваленко, Т. О., & Панченко, С. В. (2022). Етичні аспекти застосування алгоритмів оцінювання у дистанційному навчанні. Journal of Distance Education, 10(2), 54–62.
Кириченко, Л. П. (2023). Цифрові асистенти у навчальному процесі: виклики етики та захисту даних. Science and Education, 4(12), 25–33.
UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence.
Yadav, A., & Chakraborty, S. (2020). Algorithmic Bias in Educational AI Systems: Causes and Mitigations. International Journal of Emerging Technologies in Learning (iJET), 15(21), 154–170.
International scientific journal “Artificial Intelligence in Education: Ukraine and the World”
