Responsible AI

“Code Inspector” is a card-based serious game developed as part of the PhD project Responsible AI in Risk-Based Inspectorates at TU Delft. The game is grounded in empirical research conducted within Dutch public-sector regulatory inspectorates, examining how transparency practices emerge and how dilemmas around data, accountability, and explainability are addressed in practice.

The game is played in two teams: a Domain Team of inspectors and domain experts, and a Tech Team of data scientists. Both teams share a common goal, building an effective inspection model that is also legitimate, but each team brings its own priorities: the Domain Team focuses on making the model understandable and usable in the field, while the Tech Team focuses on building a high-performing model. Along the way, players make decisions about data, risks, and how to mitigate those risks through documentation, justification, and other actions. These decisions involve real trade-offs, and the consequences follow.

Ana Gagua, TU Delft, agagua@tudelft.nl