MAS Sets Out Recommendations on Use of Artificial Intelligence

The paper highlights observations from the review of the policies and governance frameworks of institutions to meet the Fairness objectives under the Fairness, Ethics, Accountability, and Transparency (FEAT) Principles and their implementation effectiveness in actual artificial intelligence and machine learning use cases. The paper also outlines the recommendations, good practices, and illustrative examples observed from the review. As per the recommendations, financial institutions should:

• ensure that the frameworks to determine whether models are in scope for materiality and fairness assessments are sufficiently comprehensive, taking into consideration all relevant Artificial Intelligence and Data Analytics models.
• consider whether models under the regulatory models have Artificial Intelligence and Data Analytics features and are subject to materiality and fairness assessments.
• adopt frameworks for materiality that take into account a comprehensive range of factors and quantitative parameters to supplement qualitative, to facilitate more consistent and holistic assessments.
• apply a calibrated approach when performing FEAT and fairness assessments based on the materiality of their models and ensure that fairness assessments should be substantiated and supported by adequate justifications.
• consider specifying a list of protected attributes and their proxies, the use of which is subject to justifications and approvals, as a safeguard against discrimination.
• review models for fairness on a regular basis, especially those with high materiality.
• base their fairness objectives and measures by comparing across individuals and groups.
• incorporate FEAT considerations during the model development life cycle and develop a roadmap to put existing models through the FEAT assessments.
• set up appropriate structures and processes for an independent, or second line of defense function, to perform validations on the materiality and fairness assessments.
• ensure that the units involved in assessing and validating models against the FEAT Principles are adequately resourced and have the necessary skillsets as they scale up their use of Artificial Intelligence and Data Analytics.

The paper concludes that the use of Artificial Intelligence and Data Analytics by financial institutions must be accompanied by good governance and risk management as well as sustainable strategies. In line with a risk-based approach, financial institutions that use or have plans to use Artificial Intelligence and Data Analytics models extensively in decision-making should take into account the aforementioned recommendations to incorporate the FEAT Principles in their Artificial Intelligence and Data Analytics policies and workflows. Going forward, MAS will continue to work with the industry to promote the responsible use of Artificial Intelligence and Data Analytics through the effective implementation of the FEAT Principles.

Keywords: Asia Pacific, Singapore, Banking, Insurance, Securities, Fintech, Artificial Intelligence, Machine Learning, Predictive Analytics, Lending, Regtech, AIDA, Governance, FEAT Principles, MAS