Pluralistic-Alignment Urbanism: Operationalizing a Right to AI for Inclusive Public Space

Municipal agencies increasingly use machine learning to inventory sidewalks, score streetscapes, and generate visualizations of public-space interventions. These systems produce scores, maps, and synthetic imagery that enter budgeting, design iteration, and public justification. Because judgments about inclusion, safety, and belonging remain contested in public space, municipal AI governance cannot rely on a single evaluation target as if it were neutral or universally shared. This paper proposes Pluralistic-Alignment Urbanism (PAU), a procedural governance framework that treats public-space AI systems as civic infrastructure and formulates a procedural Right to AI for municipal uses of such systems. We use two participatory case studies to examine what disagreement, subgroup variation, bounded predictive scaling, and neutral preference judgments can realistically support in municipal practice. PAU organizes governance around entitlements, municipal duties, documentation artifacts, and recourse triggers for systems that represent public space or generate planning imagery. The cases are grounded in participatory collaborations with community organizations in Montréal, Canada. Street Review elicits resident criteria for streetscape evaluation and trains a subgroup-aware scaling model that maps co-produced judgments; the model attains R2 = 0.89 on a held-out test set. Here, train/test evaluation is used to assess the bounded feasibility of scaling co-produced judgments into auditable planning artifacts, rather than to assert a single correct target for inclusive streets. LIVS (a Local Intersectional Visual Spaces dataset) constructs pluralistic preference data for aligning text-to-image models and treats neutral selections as evidence of indeterminacy. In an evaluation collected after preference tuning (2,100 comparisons), neutral selections are 52.4% (bootstrap 95% CI [50.2, 54.5]), and Direct Preference Optimization (DPO)-tuned outputs are preferred in 33.3% ([31.3, 35.3]). Across the cases, disagreement appears structured, deliberation changes what counts as evidence, scaling is feasible but limited by modality and data coverage, and neutrality in generative evaluation constrains what preference tuning can justify. We translate these constraints into a municipal governance architecture with disaggregated reporting, a versioned value register that distinguishes benchmarkable from contested dimensions, standing deliberative cells, procurement clauses, and defined pause and rollback authority.