This study examines how university stakeholders speculatively imagine futures for artificial intelligence in education (AIED) and the challenges these futures entail. Sixty-nine teachers, students, researchers and PhD candidates from Swedish higher education submitted short speculative scenarios via a bespoke web platform. Using Causal Layered Analysis (CLA), each text was coded across litany, systems, worldview and myth/metaphor layers to surface the assumptions and values underpinning imagined futures. Further analysis yielded four recurring configurations: Enhancement (AI as assistant/partner that personalises learning while keeping human judgement central), Transformation (human-AI fusion and continuous learning ecosystems that reconfigure institutions), Displacement (market logics and automation that deskill educators and render universities credential factories), and Resistance (protective constraints and AI-free spaces to preserve autonomy, authenticity and empathy). Across configurations, three cross-cutting tensions persisted: the human remainder (what stays uniquely human), the assessment paradox (how to evaluate learning amid AI-assisted outputs), and the efficiency-depth trade-off. It is argued that these tensions reflect AIED's character as a wicked problem: they are not resolvable by technical fixes alone, but demand negotiated, value-explicit choices. The findings suggest that debates about AI in education reflect fundamental educational philosophies rather than merely technological capabilities. By documenting diverse stakeholder voices through education fiction, this study provides empirical grounding for understanding AIED as a site of contested imaginaries requiring negotiation between multiple futures. The study contributes methodologically by demonstrating education fiction's value for exploring complex sociotechnical futures and practically by revealing tensions that educators and policymakers must navigate in designing AI-integrated educational systems.