On November 12, 2018, the Monetary Authority of Singapore (“MAS”) released guidelines for financial services firms to consider when they make decisions related to artificial intelligence and data analytics (“AIDA”). The guidelines, entitled “Principles to Promote Fairness, Ethics, Accountability and Transparency (FEAT) in the Use of Artificial Intelligence and Data Analytics in Singapore’s Financial Sector” (the “Principles”), are believed to be the first AIDA guidelines issued by a central bank or financial regulator, and emphasize the importance of fairness, ethics, accountability and transparency in how firms utilize AIDA with respect to their customers.
A core message of the guidelines is that firms should not simply rely on AIDA solutions without understanding how they are designed to operate, what types of data serve as analytics inputs, and whether the application of technology solutions results in fair, equitable, accountable and transparent outcomes for customers, counterparties and the market more generally. The MAS’s guidance is thus also a useful reminder that while technology can be extremely useful in removing bias and ensuring fair treatment, it can also be applied – either intentionally or not – in ways that are unfair or result in undesirable outcomes.
The FEAT principles were “co-created” by the MAS and a committee of senior industry partners drawn from among the largest banks and financial services firms in Singapore as well as technology firms like Microsoft and Amazon Web Services. While the guidelines do not themselves give rise to enforceable standards, they are intended to provide a “foundational framework” for the use of AIDA, and in that sense may also be useful in helping firms apply AIDA in a manner that does not violate other enforceable standards of conduct.
The FEAT principles outline four categories of conduct, and provide illustrative examples with respect to each:
- Firms can use AIDA to customize their services available to customers, but they should not use AIDA to disadvantage customers without some justification. For example, while a firm could use AIDA to evaluate whether to increase a customer’s credit line, the firm must be able justify the selection and weight of the variables it supplies to the AIDA system.
- Moreover, firms should ensure that AIDA systems function as intended and do not unintentionally bias different customers. To do so, the Principles recommend having an internal governance framework that reviews any AIDA system’s performance for accuracy.
- Firms should consider whether using AIDA aligns with their ethical values and whether their decision, from an ethical standpoint, would be different had a human been assigned the work of the relevant AIDA system. For example, firms could create and utilize ethical review boards when making AIDA-based decisions.
- Firms should develop internal structures that ensure AIDA-related decisions are made by the correct authority and are explained to theirs internal stakeholders. Specifically, the Principles recommend that firms delineate who in its corporate structure shall make AIDA-related decisions and hold information sessions for internal stakeholders as to how it uses AIDA.
- Externally, firms should provide their customers and stakeholders sufficient information about the AIDA it uses so those groups can hold firms accountable. For example, firms could give customers a way to learn more about its use of AIDA and to review and update information that firms’ AIDA systems use.
- Firms should offer its customers—and the general public—clear explanations of how they utilize AIDA and the impact doing so may have. As a result, firms would promote public confidence in both their own decision making and AIDA in general.
- The Principles note business risks of transparency, such as exposing a firm’s competitive advantages. For example, a firm using an AIDA system to detect fraud may not want to share information about how this system operates because a bad actor could use that information to manipulate the system. Nonetheless, the firm can still be transparent with its customers about other uses of AIDA, like explaining how it may use AIDA to determine a customer’s insurance premium.
The FEAT principles reflect thoughtful consideration about how technology is applied in the financial services sector. They are important guidance for firms doing business in Singapore, but also useful for firms outside Singapore to consider in evaluating their own use of technology. Among other considerations:
- It is important to have a robust and well-documented governance structure in place with respect to the use of AIDA. That means ensuring that responsible staff members have the resources to understand how technology solutions function and what inputs they consider, and to evaluate and manage the implementation of those solutions. While this responsibility is sometimes left to IT or technology functions, it may be important for AIDA governance to include relevant business people who understand the relevant markets, products or customers.
- Firms should consider transparency of the AIDA solution itself when choosing and implementing the solution. Firms need to ensure that they understand and can adjust both the factors and the weight of the factors being used to make classifications on data. This means that firms should carefully consider what types of algorithms and processes are driving the classification in the AIDA solution prior to implementation of the solution.
- Evaluation of the use of AIDA solutions does not end when the technology is implemented. Firms should continuously evaluate how such solutions are functioning and develop metrics for measuring how technology solutions are impacting outcomes.
- The quality of AIDA is only as good as the inputs, and those inputs should be well understood. In evaluating the effectiveness of a technology solution, firms should consider whether their systems are relying on robust and accurate data, and whether the data includes factors (for example, race or ethnicity) that may be inappropriate to consider as part of a decision-making process.
- AIDA solutions focused on services provided to the general public demand even greater scrutiny. This point bears particular consideration given the increasing prevalence of so-called Robo-Advisor financial services advisors, which provide financial advice based on algorithmic models. But is also relevant to other public-facing systems, such as those used to screen employment resumes or evaluate the approval of credit applications.
The FEAT principles represent a continuing effort by the MAS to maintain itself at the cutting edge of financial services regulation in an industry that continues to evolve in light of new and innovative technology solutions. These principles provide sound, reasonable guidance for financial services firms relying on or implementing AIDA solutions and, particularly given the wide industry consultation that informed them, are worthy of consideration.