Singapore’s Info-Communications Media Development Authority (IMDA) has published A Proposed Model Artificial Intelligence Governance Framework (Model Framework). The Personal Data Protection Commission (PDPC), an IMDA agency, is accepting feedback on the Model Framework until 30 June 2019.

The PDPC has prepared the Model Framework in a context of rapidly evolving AI capabilities and uses and a corresponding rise of concerns from consumers, regulators and experts about privacy, fairness and ethics.  The Model Framework seeks to provide guidance to organisations that incorporate AI solutions into their products and services on ways to approach those issues that will inspire consumer confidence and enhance regulatory compliance.  The PDPC consulted a broad range of academics, technologists and industry representatives as it developed the Model Framework.

The Model Framework is based on two guiding principles:

1.  Decisions made or assisted by AI should be explainable, transparent and fair.

2.  AI solutions should be human-centric, meaning the interests of human beings should be paramount.

The Model Framework encourages organisations to develop and operate their AI solutions in a business environment that has been consciously defined in four key dimensions:

• Internal Governance.  An organisation’s existing governance structures should be adapted or augmented to address AI issues.  An organisation’s board and top management should be aware of and involved in AI issues.  Clear accountability for the responsible use of AI within an organisation should be assigned.  The risks related to specific AI solutions should be assessed and appropriate risk management policies and controls adopted.

• AI Solution Design.  There are many different ways to implement AI.  When designing its AI solutions, an organisation should consider its corporate values, the societal norms of the countries in which it operates and the potential risks to individuals resulting from its AI solutions, among others.

• Operations Management.  Two essential elements of AI are data and algorithms.  An organisation should consider carefully the sources of its data and take care to ensure quality and avoid bias.  It must also develop policies to review and update its data periodically.  Protecting personal data is essential.[1]

Similarly, an organisation should carefully evaluate its AI algorithms to ensure they are explainable, repeatable and traceable. To what degree can the operation of the algorithm be explained to consumers at a level they can understand? Does the algorithm produce repeatable results? Does it handle exceptions properly and consistently? Does it enable active monitoring and an audit trail? Those factors and more should inform the design of the AI solutions and the larger product support environment in which they will operate.

• Managing Customer Relationships.  An organisation should communicate clearly and openly with customers about its AI solutions, including where and how AI solutions are used and how they may affect customers.  An organisation should consider whether to offer customers a choice to opt-out of an AI solution and should offer ways for customers to provide feedback and to challenge adverse decisions by AI solutions.

The Model Framework is voluntary guidance intended to promote thoughtful and responsible implementation of AI solutions.  AI is seen as an enabler of the digital economy in Singapore and one of four ‘frontier technologies’, the others being cybersecurity, immersive media and the internet of things.  It is hoped the Model Framework will enhance the public’s understanding of how AI solutions affect them, which will in turn lead to greater trust in and acceptance of those solutions.

IMDA announced the Model Framework at the World Economic Forum (WEF) in Davos, Switzerland, on 23 January 2019.  IMDA will be collaborating with the WEF to facilitate discussion and adoption of the Model Framework outside Singapore.

[1]IMDA noted that Singapore’s Personal Data Protection Act is under review to strengthen consumer trust and provide greater clarity for organisations that use data.

‍