TLDR: Singapore has introduced an updated Model AI Governance Framework specifically tailored for Generative AI, building upon its earlier frameworks. Launched by the IMDA and AI Verify Foundation on May 30, 2024, this framework aims to foster a trusted ecosystem for AI innovation by addressing the unique risks and opportunities presented by generative technologies through nine key dimensions, including accountability, data quality, and content provenance.
Singapore has taken a proactive step in shaping the future of artificial intelligence with the launch of its Model AI Governance Framework for Generative AI. This significant development, spearheaded by the Infocomm Media Development Authority (IMDA) and the AI Verify Foundation, was officially unveiled on May 30, 2024. The framework represents an expansion of Singapore’s existing AI governance approach, which previously covered traditional AI, with the first edition released in 2019 and a revised second edition in 2020.
The rapid advancements in generative AI, while offering immense transformative potential, have also brought to light new and reinforced existing risks. These include issues such as hallucinations, copyright infringement, and value alignment, in addition to known AI risks like bias, misuse, and lack of explainability. Recognizing these evolving challenges, the new Generative AI Framework seeks to establish a systematic and balanced approach to address these concerns while simultaneously fostering an environment conducive to innovation.
The framework is built on the principle of fostering a trusted ecosystem, which is deemed critical for public confidence in AI and for maximizing innovation. It emphasizes the collective effort required from all key stakeholders, including policymakers, industry leaders, the research community, and the broader public.
Central to the Generative AI Framework are nine key dimensions that are proposed to be evaluated comprehensively to build this trusted ecosystem:
1. Accountability: This is a crucial consideration to incentivize all players along the AI development chain to be responsible to end-users. The framework suggests allocating responsibility upfront in the development process (ex-ante) and providing guidance on redress mechanisms if issues arise later (ex-post). This aligns with the cloud industry’s shared responsibility models.
2. Data Quality: Ensuring the quality and integrity of data used to train and operate generative AI models.
3. Content Provenance: Addressing the origin and authenticity of content generated by AI, which is vital for combating misinformation and deepfakes.
4. Safety: Implementing best practices by model developers and application deployers across the AI development lifecycle, from development to disclosure and evaluation. This extends principles from the 2020 framework for traditional AI.
5. Security: Protecting AI systems from malicious attacks and ensuring their resilience.
6. Fairness: Mitigating biases in AI outputs and ensuring equitable treatment.
7. Transparency: Providing greater clarity on how AI systems work, especially for high-risk models with national security or societal implications, which may require greater transparency to the government.
8. Explainability: Making AI decisions understandable to humans.
9. Human Agency and Oversight: Ensuring that humans retain control and oversight over AI systems.
The framework also highlights the need for a more comprehensive and systematic approach to safety evaluations for generative AI, moving beyond traditional benchmarking and red teaming. It builds on policy ideas from Singapore’s earlier Discussion Paper on Generative AI: Implications for Trust and Governance, issued in June 2023, and incorporates insights from international discussions and leading AI organizations.
Also Read:
- Singapore Leads Regional Dialogue on Responsible AI Governance and Collaborative Innovation
- Artificial Intelligence Set to Drive Singapore’s Sustained 3% Annual GDP Growth
Singapore’s approach to AI governance is holistic, viewing it at the intersection of the economy and broader society. The IMDA, acting as an economic developer, regulator, and social leveller, aims to push boundaries not only within Singapore but also across Asia and beyond, enabling the safe and trusted use of this dynamic technology. The framework is designed to evolve as technology and policy discussions progress, welcoming feedback for further refinement.
