30 Association of Research Libraries Research Library Issues 299 — 2019 The system-centric definition also has three key concepts: rationale, strengths and weaknesses, and future behavior. The rationale could pertain to the purpose of the AI, the logic of its model, a justification for its actions, or its application in specific situations. The disclosure of strengths and weaknesses indicates a level of openness and transparency that would make obvious system limitations and key assumptions. It also seems likely to conflict with trade secrecy, intellectual property issues, and data privacy. The emphasis on future behavior recognizes that AI will be an ongoing part of everyday life, hence the need for predictability and consistency. It also implies that AI will be subject to longitudinal evaluations to ensure levels of performance. European Union General Data Protection Regulation It is difficult to overestimate the impact of the European Union (EU)’s 2018 General Data Protection Regulation (GDPR) on XAI. The GDPR’s “right to explanation” regarding algorithmic decisions is having a global reach (the so-called “Brussels effect”), causing debate and regulatory review well beyond the EU. While interpretability has always been a concern in computer science, the GDPR has refocused this issue as an explainability problem and made it a public policy question. The explanatory requirements in the GDPR are actually quite narrow, but their impact has been much broader, with jurisdictions as diverse as Canada and the City of New York developing impact assessment protocols with respect to algorithmic decisions that include explainability requirements. As seen with the “right to be forgotten,” international legislation or regulation can have a profound effect on national affairs. The global nature of digital technologies is a reminder that monitoring the policy agendas of other jurisdictions is important.