An Emancipatory Vision for Designing (Generative) AI for Learner Flourishing

Synthesis: Prieto and Dimitriadis argue that current generative AI educational design, oriented toward workforce productivity, pushes learners toward individualism, isolation, over-reliance, and dependence — driven by wider systemic factors and the human tendency to take the easiest path. They contend that human-centered design approaches like value-sensitive design are insufficient because they ignore these systemic drivers, and that Learner Agency is effectively a zero-sum game: the more agency bestowed on tools, the less learners retain. The paper proposes an emancipatory vision oriented toward learner flourishing that considers the wider complex systems learners inhabit, offering tentative design principles and an overall design methodology, while acknowledging many open questions remain.

The Critique of Current GenAI Design

The paper identifies several "problematic assumptions" underlying most current AIED proposals:

1. That we can build increasingly complex and agentic technologies without damaging learner Learner Agency and flourishing, when agency may be a zero-sum game.
2. That we can predict the outcomes of introducing complex, stochastic technology into education's hypercomplex system without unwanted side effects.
3. That education's goal is mainly preparing learners for workforce participation, rather than socializing them into the knowledge, values, and democratic citizenship of their social groups and wider culture.

The authors draw on evidence of "over-reliance", "metacognitive laziness", social isolation, and lower Well-Being linked to naturally-emerging uses of general-purpose GenAI, alongside the learning-performance paradox and environmental costs of adoption.

Toward Learner Flourishing

Rather than accepting the productivity framing, the paper advances an emancipatory vision oriented toward learner flourishing within the wider complex systems learners inhabit (classrooms nested in schools, nested in societies and environments). It offers tentative design principles and an overall design methodology, though it explicitly notes many open questions remain before the vision can be realized. This connects to critical perspectives on AI in education that question whether agentic tools serve learners or displace their development.

What this means for practice

• Designers. Build tools and services rather than autonomous agents, treating Learner Agency as zero-sum: automate only extraneous effort that interferes with learning, not decisions that belong to the learner.
• Designers. Under-design for appropriation — make systems user-adaptable or purposefully incomplete — and plan from the start to observe how learners and teachers repurpose them in practice.
• Designers. Specify fading Scaffolding and explicit hand-off points, including activities that move attention off the screen into collaboration or the physical environment, and treat learner independence from the tool as the success metric.
• Designers. Avoid design choices that let a stochastic system "pose as a person"; anthropomorphic framing can produce emotional attachment to an Large Language Models (LLMs), so support the emotional aspects of learning without inviting dependence.
• Researchers. Evaluate at multiple levels (learner, classroom, home, school) and across multiple outcomes — Metacognition and Well-Being, not only knowledge — at both immediate and delayed timeframes, rather than optimizing one metric.

Limitations

• This is a position paper presented at an EC-TEL 2026 workshop: it offers tentative design principles and a proposed design methodology, with no implemented system, no participants, and no empirical evaluation of the vision.
• The authors list the open questions themselves — concrete design methods that resolve value tensions, exemplars of AIED technology with appropriation-focused multi-level evaluations, measurement of social-group and environment-level constructs such as societal and environmental flourishing, and inherent design tensions in even open-weight LLMs.
• The critique of the productivity framing draws on evidence of over-reliance, metacognitive laziness, isolation, and lower well-being from naturally emerging uses of general-purpose GenAI, not from studies of AI built specifically for education.
• The dismissal of human-centered design approaches such as value-sensitive design is conceptual — they are judged insufficient to handle systemic drivers like economic incentives — and no empirical comparison of outcomes under the two design philosophies is provided.

Citation

Prieto, L. P., & Dimitriadis, Y. (2026). An Emancipatory Vision for Designing (Generative) AI for Learner Flourishing. International Workshop on Critical and More-than-human Perspectives on AI in Education (EC-TEL 2026). arXiv:2609.07715 [cs.CY].