Abstract

Algorithmic transparency debates have largely focused on technical strategies for unveiling how AI systems generate outcomes, often treating opacity as a computational problem detached from the institutional and social settings in which algorithmic decisions are enacted. Addressing this gap, this study examines how citizens perceive, trust, and demand transparency from algorithmic systems embedded in Chilean public services, particularly those that mediate access to education and health benefits. Drawing on the Transparency by Design framework, we conceptualize transparency as a sociotechnical, three-layered construct shaped by micro-level informational conditions, meso-level relational dynamics, and macro-level institutional structures. We conducted six participatory design focus groups (n = 48) to elicit citizens' perceptions, expectations, and proposals for transparency. Findings show that transparency expectations are tightly linked to institutional trust: participants tended to assume AI reliability when discussing it in abstract terms, but in concrete, high-impact algorithmic systems, they demanded explanations that clarify institutional processes, strengthen situational awareness, and preserve human control. Citizens articulated distinct ex-ante and ex-post transparency needs, favored progressive disclosure, and called for accessible communication about system integrity together with clear avenues to challenge decisions. We argue that algorithmic transparency in the public sector must extend beyond technical explanations to include relational and institutional aspects, and that participatory approaches are essential for designing transparency that fosters citizen agency.