Singapore proposes framework to foster trusted Generative AI development

1. The AI Verify Foundation (AIVF) and Infocomm Media Development Authority (IMDA) have developed a draft Model AI Governance Framework for Generative AI. This framework expands on the existing Model Governance Framework that covers Traditional AI, last updated in 2020¹.

2. Generative AI has significant transformative potential - above and beyond what Traditional AI has been able to achieve. This also comes with risks. While it remains a dynamically developing space, there is growing global consensus that consistent principles are needed to create a trusted environment - one that enables end-users to use AI confidently and safely, while allowing space for cutting-edge innovation. The use and impact of AI is not limited to individual countries. Hence this proposed framework aims to facilitate international conversations among policymakers, industry, and the research community, to enable trusted development globally.

3. With Generative AI, there is a need to update the earlier model governance framework to holistically address new issues that have emerged. The proposed framework integrates integrates ideas from our earlier discussion paper on Generative AI², which put forward a conceptual foundation. It also draws on earlier technical work to provide an initial catalogue and guidance on suggested practices for safety evaluation of Generative AI models³. On top of this, it draws on practical insights from ongoing evaluations tests, conducted within our Generative AI Evaluation Sandbox⁴.

4. The framework looks at nine proposed dimensions to support a comprehensive and trusted AI ecosystem. The core elements are based on the principles that decisions made by AI should be explainable, transparent, and fair. Beyond principles, it offers practical suggestions that model developers and policymakers can apply as initial steps (see Annex (157.58KB) for more information).

5. AI governance remains a nascent space. Building international consensus is key, as demonstrated by the successful mapping and interoperability of national AI governance frameworks between Singapore and the US, through the IMDA and the US National Institute of Science and Technology (NIST) crosswalk⁵. The proposed Model Governance Framework for Generative AI takes this one step further by covering the latest developments in Generative AI. In turn, this will inform Singapore’s next steps, as we adopt a practical approach to maximise both trust and innovation. This framework will evolve as techniques and technologies develop.

6. For more details, please refer to the proposed draft framework. We welcome views from the international community, which can be sent to info@aiverify.sg. This will support finalisation of the Model AI Governance Framework in mid-2024.

Resource:

• Annex: Nine dimensions of the proposed Model AI Governance Framework (157.58KB)