Singapore - Ethical AI Advisory Council

Overview

The Advisory Council on the Ethical Use of AI and Data was announced as part of Singapore’s 2018 AI governance and ethics package and formally convened in late 2018. Its primary role is advisory: to provide high‑level guidance to government agencies and to shape practical guidance for businesses on ethical and governance issues arising from private‑sector use of data‑driven technologies. The Council’s creation was announced by the Infocomm Media Development Authority (IMDA) and is closely linked to the Personal Data Protection Commission (PDPC)’s June 2018 Discussion Paper on AI and Personal Data and wider national initiatives to build a trusted AI ecosystem.

Definitions

For the Council’s work, key terms are defined pragmatically: "AI" refers to data‑driven algorithmic systems used for decisioning, prediction or automation; "ethical use" denotes practices that respect human rights, fairness, transparency and accountability; "data" includes personal and non‑personal datasets used by AI systems; and "organisations" denotes private‑sector entities deploying AI systems. The PDPC Discussion Paper and the subsequent Model AI Governance Framework provide working definitions that informed the Council’s deliberations.

Governance and Institutional Framework

The Council was appointed by the Minister for Communications and Information and coordinated through IMDA, with cross‑agency engagement (notably PDPC and AI Singapore). It comprised private sector, academic and civil society representatives and was chaired by senior counsel V.K. Rajah. The Council functioned as a consultative advisory body rather than a regulator; its outputs feed into policy guidance and voluntary frameworks. For official background and IMDA’s AI coordination role, see IMDA: Artificial Intelligence and the IMDA announcements accompanying the Council’s launch.

Key Focus Areas

The Council concentrated on translating ethical principles into operational practices for organisations. Focus areas included: (1) governance and oversight practices for AI lifecycle management; (2) data protection and privacy safeguards where AI uses personal data; (3) explainability and transparency measures for automated decisions; (4) risk‑based assessment and mitigation for high‑impact uses (e.g., finance, healthcare); (5) human‑in‑the‑loop and human‑on‑the‑loop controls; (6) standards for safety testing, validation and documentation; and (7) sectoral engagement so that guidance could be adapted to industry specifics. These priorities directly informed the PDPC/IMDA model documents and practitioner toolkits that followed.

Implementation Framework

The Council advocated a voluntary, risk‑based, and sector‑sensitive implementation approach. Key mechanisms included publication of model frameworks (e.g., the PDPC Model AI Governance Framework), stakeholder consultations to refine drafts, a compendium of use cases, and later development of testing and self‑assessment tools (such as AI Verify). Implementation emphasised organisational governance (roles and responsibilities), data governance (quality and provenance), documentation (model cards, audit trails) and testing/monitoring regimes to detect biases, performance drift and safety issues.

Monitoring and Evaluation

Monitoring is largely achieved through voluntary reporting, sectoral regulator engagement, case studies and the use of self‑assessment tools. The Council recommended periodic reviews of frameworks, publication of compendia to track best practice adoption, and promoted independent evaluation tools. Sector regulators may apply monitoring or oversight where statutory obligations (e.g. PDPA, sectoral rules by MAS) intersect with AI deployments.

Penalties, Liability, and Appeals

The Advisory Council itself does not impose penalties; it produces guidance. Legal liability and enforcement continue to rest with existing legislative and regulatory regimes (for example, PDPA enforcement for personal data breaches, MAS rules for regulated financial institutions, and the Cybersecurity Act for critical systems). The Council’s guidance helps organisations reduce legal and compliance risk but does not replace statutory obligations or judicial remedies.

Relationship to Other Instruments

The Council’s outputs were inputs to the PDPC’s Discussion Paper and the Model AI Governance Framework (and later its editions and companion guides). It complements sectoral initiatives such as the MAS FEAT/Veritas work for finance and national testing/verification efforts like AI Verify. The Council’s role is coherent with Singapore’s National AI Strategy and inter‑agency coordination led by IMDA and the Smart Nation office.

International Alignment

The proposed governance framework aims to establish baseline principles for responsible artificial intelligence deployment that can serve businesses and society at large. In developing these principles, the Personal Data Protection Commission (PDPC) designed a technology-neutral and sector-agnostic framework that focuses on governance, risk management, and consumer trust without imposing prescriptive rules prematurely. The framework draws upon broad human-centric design concepts, such as beneficence and harm avoidance, referencing established international guidance including UNICEF's Humanitarian Principles to inform its approach to protecting individuals from physical, psychological, emotional, and economic harm.

By prioritizing explainability, transparency, fairness, and human-centric decision-making, the framework aligns Singapore's domestic AI governance discussion with broader international consensus on trustworthy AI. The framework is structured to be adaptable across different sectors and organizational needs, allowing private sector entities to develop voluntary codes of practice that remain interoperable with international best practices. This approach supports regulatory clarity and market competition while facilitating cross-border commercial deployment of artificial intelligence solutions.

Implementation Timeline

Sources and References