Experts Outline Framework for Trustworthy AI in Public Services
A panel of European GovTech leaders and ethicists explored the design and governance of trustworthy AI in digital public infrastructure in a video published February 18. Panelists stressed that trust must be a foundational design principle, requiring transparency, explainability for automated decisions, and robust human fallback options to ensure accessibility and fairness.
- The European Union's AI Act, which entered into force in August 2024, establishes a risk-based legal framework for AI systems to ensure they are safe, transparent, traceable, and non-discriminatory. It categorizes AI applications into unacceptable, high, limited, and minimal risk, with specific obligations for each tier. - AI systems are classified as high-risk in the public sector when used for applications such as biometric identification, critical infrastructure management, education, employment, and essential government services like social benefits and law enforcement. These systems are subject to strict requirements, including risk assessments, high-quality data usage to prevent discriminatory outcomes, and human oversight. - Prohibited AI applications under the EU AI Act include social scoring by governments, indiscriminate scraping of facial images for recognition databases, and emotion recognition in workplaces or schools. These rules came into effect on February 2, 2025. - To support the implementation of the AI Act, a new European AI Office was established in August 2024 to regulate general-purpose AI models at the EU level. Additionally, each member state must establish at least one AI regulatory sandbox by August 2, 2026, to help companies and public authorities develop compliant AI. - The Organisation for Economic Co-operation and Development (OECD) provides a complementary framework specifically for AI in the public sector, addressing challenges like procurement processes not designed for evolving AI systems and the need to maintain public trust. Its principles emphasize human-centered values, fairness, transparency, and accountability. - A 2024 report indicated that over a third of public administrations in the EU had adopted AI, with Northern and Western Europe leading at 45% adoption compared to 25% in parts of Southern and Eastern Europe. Common applications include healthcare for predicting patient risk, tax authorities identifying fraud, and chatbots for citizen services. - Public authorities are considered "deployers" of AI under the AI Act and have distinct obligations, such as conducting fundamental rights impact assessments before implementation and registering high-risk systems in a new EU database. - Case studies from cities like Amsterdam and Helsinki demonstrate the use of AI to improve urban infrastructure, traffic management, and energy efficiency. However, successful adoption often hinges on involving end-users early in the design process and overcoming challenges posed by legacy IT systems.
