Qualification: Conceptualising Contestability: Perspectives on Contesting Algorithmic Decisions
Explanation begins a challenge. It does not complete one. Lyons, Velloso, and Miller examine contestability in algorithmic decision making through a thematic analysis of 65 submissions to Australia's AI Ethics Framework. They distinguish information from institutional process. An explanation supplies grounds to question a decision. Contestability also requires notice, a clear path into review, a reviewer with authority to act, and a new decision or outcome when review is complete. The paper identifies a basic failure condition: users may receive a nominal right to appeal and still receive no response or resolution. The remedial element is concrete. The reviewed submissions describe redress as a new decision, compensation for harm, or reversal of a decision's effects. They connect the availability of those remedies to enforceability. Internal review has value because the responsible institution may be positioned to reverse its decision or provide redress. External review remains necessary where independence or escalation controls. This evidence qualifies the institutional requirement for sovereign local agent runtimes. A runtime may record an explanation and expose a complaint interface without creating a functioning institution. The challenge path must identify what can be contested, who can decide the challenge, what remedial powers are available, and how a resulting decision becomes effective. Explanation supports that process. It cannot substitute for remedial capacity. The study does not test a deployed agent runtime. Its evidence consists of stakeholder submissions and the authors' qualitative synthesis, not observed remedy rates. It does not establish which reviewer, remedy, timing rule, or enforcement mechanism is sufficient for every context. It also warns that contestation places burdens on affected persons and cannot replace lawful design, oversight, audit, or prevention. The narrower implication remains. A challenge mechanism without authority to change consequences is an information channel, not an institutional remedy.
institutional-designgovernance-designai-and-agentsaccountabilitycontestabilityreviewremedyalgorithmic-decision-making