Conceptualising Contestability: Perspectives on Contesting Algorithmic Decisions

Contestability is emerging as a critical safeguard: The paper identifies contestability—the ability to challenge algorithmic decisions—as an increasingly recognized protection for individuals affected by high-stakes algorithmic decision-making, suggesting it should be a core principle in AI governance frameworks.

There is no consensus on what contestability means: While organizations and individuals responding to Australia's AI Ethics Framework agreed contestability is important, they disagreed on its specific definition and requirements, indicating the field still lacks clear guidance on how to implement this principle in practice.

People expect algorithmic contestability to work like human decision contestability: Respondents drew parallels between contesting algorithmic decisions and existing processes for challenging human decision-making, suggesting that familiar contestation mechanisms could serve as a model for algorithmic systems.

Contestability protects vulnerable individuals: The paper emphasizes that contestability is fundamentally viewed as a mechanism to protect people from potentially harmful or unfair algorithmic decisions, particularly in high-stakes contexts where decisions significantly impact individuals' lives.

Implementation requires further development: The research reveals that while there is broad support for contestability as an ethical principle in AI frameworks, substantial work remains to translate this concept into concrete, actionable requirements that organizations can actually implement.