How Singapore is developing trustworthy AI

Artificial intelligence (AI) is becoming ubiquitous. It underpins all kinds of functions, including critical ones. It impacts our work, life and play. It is used, for example, in medical imagery to detect severe illnesses and, through facial recognition, to unlock a smartphone.

There are instances, however, where the AI models output does not perform as intended. When the AI model is not trained and tested against representative datasets, for example, there can be bias against certain populations.

With greater maturity and more pervasive adoption of AI, society needs to be assured that AI systems are fair, explainable and safe and that the companies deploying them are accountable and transparent about their use of it. So, how can companies demonstrate their implementation of responsible AI in an objective and verifiable way to their stakeholders?

In Singapore, we believe that voluntary self-assessment is a start. The Info-Communications Media Development Authority (IMDA) and Personal Data Protection Commission (PDPC) have taken the first step by developing A.I. Verify. This is an AI governance testing framework and toolkit that enables industries to be more transparent about their deployment of AI through technical tests and process checks.

A.I. Verify has two components. Firstly, the governance testing framework specifies the testable criteria and the corresponding processes required to carry out the test. Secondly, the software toolkit conducts technical tests and records the results of process checks.

Available as a Minimum Viable Product (MVP), companies can use A.I. Verify to demonstrate what their AI systems can do and what measures have been taken to mitigate the risks of their AI systems. The testing reports from A.I. Verify keep company stakeholders better informed of the capabilities of the AI systems they are interacting with. This process brings transparency and helps build trust in AI.

A.I. Verify does not set ethical standards. It validates companies’ claims about their AI systems vis-à-vis a set of internationally accepted AI governance principles that countries coalesce around and on which Singapore’s AI governance initiatives also stand. These principles, organized according to people's concerns about AI systems, are shown in the diagram below. For the MVP, we have selected an initial set of eight principles (shown in green). Security and Data Governance (including privacy) testing regimes are well-developed internationally and hence not included in the current pilot.

With this MVP, Singapore invites companies to participate in the international pilot and hopes to achieve the following objectives:

As more AI system developers and owners use A.I. Verify, we can crowd-source and co-develop benchmarks for acceptable levels of adherence to AI governance principles for different sectors and use cases. We can also work with technology solution providers to identify gaps and development opportunities in current methods and technologies of testing AI models and devise better ways to verify trustworthy AI.

A key element for building trust is to communicate effectively to all stakeholders about the behaviour of AI applications. Stakeholders include regulators, board members, senior management, business partners, auditors, customers and consumers. They all need different information at varying levels of detail and insights to facilitate their decision-making. Hence, customised report templates addressing specific stakeholder information needs must be developed.

These goals cannot be achieved without the collective wisdom and efforts of the broader community. As part of community building, IMDA/PDPC will organize roundtables where industry can engage regulators for early policy thinking, as well as industry-specific workshops to develop consensus on industry benchmarks for trustworthy AI. These benchmarks can be shared with international standards bodies as the community’s contribution to the global discourse on standards building.

AI ethics principles may be universal, but often their interpretation and implementation are influenced by cultural and geographic variations. These variations can lead to fragmented AI governance frameworks, which, in turn, can raise barriers to implementing trustworthy AI for companies and hinder their capacity to provide AI-enabled products and services across borders.

A.I. Verify is Singapore’s first step to identifying and defining an objective and verifiable way to validate the performance of AI systems. The international pilot will facilitate the collation and development of industry benchmarks for AI ethics principles. We look forward to working with like-minded players in this learning journey and to piloting A.I Verify.