SINGAPORE (Nov 18): At last year’s Singapore FinTech Festival (SFF), the Monetary Authority of Singapore (MAS) released a paper outlining a set of principles in the use of artificial intelligence (AI) and data analytics in Singapore’s financial sector. These principles are known collectively as FEAT (or fairness, ethics, accountability and transparency).
While FEAT was conceptually defined and developed by MAS with various industry players, how these principles were applied in practice was not immediately clear. Moreover, what constituted the minimum threshold to achieve FEAT as well as the evaluation and verification mechanisms were similarly not defined.
This is where Veritas comes into the picture. It is a framework that enables financial institutions to come together to figure out ways to implement and evaluate their AI and data analytics solutions against FEAT in a verifiable manner. It is also a platform to share open source tools, particularly coding.
The Veritas consortium comprises 17 members, including MAS, start-up ecosystem enabler SGInnovate and audit firm EY. The other members are international insurers, as well as local and foreign banks. Veritas was highlighted at this year’s SFF in a speech by Deputy Prime Minister Heng Swee Keat, who outlined the country’s National AI Strategy.
“What we are trying to do is ambitious. We are bringing together multiple financial institutions to see how FEAT can be implemented in their specific context and use. However, the approach is not a case of each doing for its own. This is a journey and an attempt to do it together,” says David Hardoon, AI special adviser at MAS, in an interview with The Edge Singapore on the sidelines of SFF.
So, how exactly does Veritas aim to achieve this? Hardoon says the first step is to explore the implementation of FEAT on a case-by-case basis. Using the example of the principle of fairness, he says the consortium members will identify and prevent individuals or groups of individuals from being systematically disadvantaged by an AI that is used in real-life business situations. These could include lending to small and medium-sized enterprises, establishing credit lines or calculating insurance premiums. “Think of it as a checklist of components to achieve FEAT,” he says.
Then, a minimum threshold can be set to evaluate whether an AI aligns with FEAT, Hardoon adds. The minimum threshold can vary according to suitability and complexities of the financial institution, industry and country, but not the “components”. “The objective is to approach this systematically,” he says.
Following that, a monitoring system can be embedded into the algorithm of an AI or externally developed to evaluate the results generated. This is a decision to be made collectively by the consortium members. “MAS doesn’t want to dictate. We want to provide the framework — a set of standards that we agree on together,” he explains.
Veritas will publish its first report — outlining the findings and considerations gathered from the consortium — by mid-2020. The source code will also be made available to the public for review, critic and improvement.
Asked whether Veritas was conceived in response to a case of AI discrimination, Hardoon denies that, saying that the framework is a “precautionary” step. “We don’t want to wait for something to happen and then say: We need it now. Governance does not prevent innovation. On the contrary, it allows for safe innovation. Similarly, we want innovation in AI, but we all have responsibility to think about where we draw the boundaries. We need to do it now before something unfortunate occurs,” he says.
So, will the Veritas consortium eventually include non-financial institutions in the framework? Hardoon will not rule that out but, with MAS as a financial regulator, he says, it was only natural to start Veritas with the financial institutions.
Interestingly, OCBC Bank is the only local bank that is not part of the consortium. Hardoon did not directly address why OCBC was excluded, but notes that the bank was involved in the formulation of FEAT. DBS Bank, on the other hand, was the only local bank that was not involved in FEAT’s formulation. Only United Overseas Bank was involved in both initiatives. “There is no right or wrong here. We are hoping to invite other participants as we progress and expand the use cases. Whether they are official or unofficial, many are involved,” he notes.
Hardoon says if Veritas proves to be a success, the next step is to replicate the framework across other industries and possibly other countries. After all, “AI isn’t a local agenda, but an international one”, he says. For now, there are many unknowns to be uncovered by Veritas. If these are resolved, MAS will be able to collaborate with foreign regulators. “We would like to be able to at least provide some degree of insight when we have these conversations with them,” he says.
Ultimately, Veritas — which is non-prescriptive — aims to build confidence in AI usage. “It really is about how to increase trust in AI,” says Hardoon.
