Towards Human-Centered Generative AI for Collaborative Educational Ecosystems: Design Principles for Strengthening School-Community Partnerships

Synthesis: This is a conceptual design paper, not a study. Lymperi synthesizes work on human-AI collaboration, generative AI in education, educational ecosystems, school-community partnerships, and responsible AI governance, then proposes six design dimensions for Human-Centered Generative AI in school-community ecosystems: Human Agency, Collaborative Intelligence, Contextual Intelligence, Community Engagement, Ethical Governance, and Sustainable Innovation. There are no participants, no site, and no outcome data; the author states that the framework, its illustrative scenario, and its preliminary indicators have not been empirically validated. The distinction worth keeping is architectural. Human Agency and Ethical Governance are conditions under which the other dimensions can operate, not features to add later; Contextual Intelligence is separated from personalization because it concerns the social, cultural, geographical, and institutional setting of collective decisions; and Community Engagement is separated from communication because it includes influence over problem definition and planning. The contribution is the relational ordering of concerns usually handled as separate checklists, offered as infrastructure for later participatory design and case research.

Key Findings

1. The framework names six design dimensions. Human Agency, Collaborative Intelligence, Contextual Intelligence, Community Engagement, Ethical Governance, and Sustainable Innovation are proposed as an integrated set rather than six separate constructs.
2. Six design propositions carry the dimensions into requirements. P1 to P6 restate each dimension as a design obligation, from preserving human decision authority to supporting continuous reflection and adaptation.
3. The dimensions are ordered relationally, not listed as a checklist. Human Agency and Ethical Governance supply foundational conditions, three middle dimensions do the interactional and adaptive work, and Sustainable Innovation supplies a long-term orientation.
4. The architecture is read at two levels. At the infrastructure level GenAI provides generation, synthesis, exploration, and communication; at the ecosystem level stakeholders contribute knowledge, values, and decisions, with the six dimensions mediating.
5. The synthesis ran in three phases across six literature domains. Following Jaakkola's guidance, the author identified concepts, grouped them functionally, and integrated them into propositions spanning HCAI, educational GenAI, collaboration, ecosystems, partnerships, and governance.
6. The paper offers six sets of preliminary qualitative indicators. Each dimension is translated into observable questions, and the author states that these are analytical criteria, not validated measurement instruments.
7. Four implementation trade-offs are stated rather than resolved. Human oversight costs time and expertise, contextual adaptation reduces standardization, broad participation complicates coordination and consent, and rapid innovation can outrun professional learning.

The gap: classroom AI versus ecosystem AI

The paper starts from where the literature looks. Work on generative AI in education concentrates on instructional effectiveness, learning outcomes, and routine teacher tasks, framing the technology as a tool for teacher-AI or learner-AI interaction inside a classroom. Lymperi argues this leaves out the relationships educational problems depend on: families, local authorities, universities, civil society organizations, cultural institutions. The concerns she names, sustainability, inclusion, digital transformation, and community resilience, cross institutional boundaries and require shared responsibility, not one teacher's judgment. GenAI is therefore repositioned as potential collaborative infrastructure for communication, knowledge sharing, participatory planning, and co-creation. The unit of collaboration changes accordingly: not one user with one model, but a network of participants holding different knowledge and authority. Rural, remote, and geographically dispersed schools, often community hubs, are named as where this matters most.

The six dimensions and how they connect

Each dimension carries a design requirement. Human Agency requires that stakeholders can question, modify, reject, or reinterpret AI output, and that responsibility is not transferred to a system because its prose sounds plausible. Collaborative Intelligence pairs human expertise with machine synthesis while keeping interpretation with people, the collaborative learning unit rather than the individual user. Contextual Intelligence requires local knowledge to shape problem definition, the criteria for comparing options, and the reading of proposed solutions, contrasted with personalization that only adapts output to an individual. Community Engagement extends participation past consultation into deliberation and influence over decisions, closer to sustained joint inquiry than a broadcast channel. Ethical AI Governance covers transparency, accountability, fairness, privacy, data practice, and oversight, and must address unequal authority, expertise, digital literacy, access, and confidence. Sustainable Innovation adds periodic review and the option to revise or stop. The relational claim is that these are not interchangeable modules: governance constrains and enables the rest, and no dimension can be optimized alone.

Tensions, power, and provisional indicators

The discussion does not pretend the choices are free. Human oversight can improve accountability but demands time and expertise, so approval cannot be a nominal signature. Local adaptation improves relevance but makes systems less comparable across schools, forcing an explicit decision about what stays common and what is local. Wider participation enriches knowledge but adds coordination, consent, and privacy work. Fast adoption can undermine sustainability of a practice never supported by professional learning or institutional capacity. A separate tension is fluency versus epistemic authority: a model can write persuasively while being incomplete or contextually wrong, risking elevation of generated text over locally situated knowledge. The framework keeps decision authority with stakeholders and treats local knowledge as an input to problem framing, and it flags that AI-mediated participation is not automatically equitable, since unequal access and confidence can amplify dominant voices. The indicators are questions for interviews, focus groups, workshops, or case observation, not a measurement scale.

What this means for practice

• Instructors. Use the six dimensions as a critique checklist before a collaborative task: who can override an AI suggestion, whose local knowledge shaped the problem, and what participants can change.
• Administrators. Treat GenAI adoption as organizational change rather than procurement, weighing professional autonomy, community communication, and institutional capacity together, and scheduling review with the option to stop a practice.
• Policymakers. Write governance that names responsibilities, transparency expectations, and data conditions for multi-stakeholder use, not only classroom rules that leave community participation unprotected.
• Instructional designers. Design for contestability: the framework makes rejecting or reinterpreting output the core of human agency and the main defense against fluent but wrong text.

Limitations

• The study is conceptual: the relationships among the six dimensions are theoretically informed design assumptions, not empirically established causal relationships.
• No empirical data, participants, or site are reported. The sustainability initiative is hypothetical, and the framework is a conceptual artifact, not a validated model.
• The preliminary indicators are not validated instruments and carry no claim of measurement validity.
• The named tensions (oversight time, reduced standardization, coordination and consent burden, capacity for professional learning) are left for future implementation studies.
• The paper is an EdArXiv preprint (v1) by a single author and proposes no new model, platform, or intervention that could be evaluated as shipped.

Citation

Lymperi, T. (2026). Towards Human-Centered Generative AI for Collaborative Educational Ecosystems: Design Principles for Strengthening School-Community Partnerships. EdArXiv preprint (v1).