How Transparency and Contestability Shape Trustworthy AI in Public Regulation

On April 25, 2025, a study titled Two Means to an End Goal delves into how explainability and contestability intersect in regulating public AI systems through expert insights.

The abstract explores the intersection of explainability and contestability in AI systems, emphasizing their role in fostering trustworthiness. It addresses challenges in implementing these principles across technical, legal, and organizational dimensions through a study of 14 regulation experts. Key findings include distinctions between descriptive and normative explainability, judicial vs. non-judicial contestation channels, and individual vs. collective action. The research identifies three translation processes: aligning top-down and bottom-up regulation, assigning responsibility for interpretation, and fostering interdisciplinary collaboration. Contributions include empirically grounded conceptualizations and practical recommendations for integrating these principles into public institutions, aiming to enhance equitable AI design and deployment.

In an era where artificial intelligence (AI) increasingly influences public sector decision-making, ensuring accountability and trust has become paramount. The integration of explainable AI and contestability mechanisms is crucial for fostering transparency and allowing individuals to challenge decisions effectively.

Traditionally, frameworks have focused on either explaining how AI decisions are made or enabling people to contest them post-decision. However, this article highlights that these two aspects should be considered together. An integrated approach ensures that citizens can both understand and influence algorithmic decisions, enhancing trust and fairness in public services.

A study involving participants sorting 40 mechanisms into categories revealed four key areas: descriptive vs. normative explanations and individual vs. collective action. This categorization underscores the different ways people engage with AI systems. For instance, explaining an individual decision is distinct from ensuring broader societal alignment with AI values.

The study found that explainability is not merely about transparency but providing actionable information that empowers individuals to challenge decisions effectively. Additionally, contestability extends beyond post-decision appeals; it involves proactive influence over how the AI operates.

In conclusion, combining explainable AI and contestability creates a robust governance framework. This dual approach ensures citizens can understand and influence algorithmic decisions, essential for maintaining trust and fairness in public sector AI applications. By integrating these elements, governments can build systems where transparency and challenge mechanisms work hand-in-hand, ensuring equitable and trustworthy outcomes.

This approach not only enhances accountability but also strengthens the social contract between citizens and institutions, fostering a more inclusive and just society.