AI principles should pave way for open probes on data usage

THERE is gnawing unease when it comes to the use of artificial intelligence (AI) for data processing by financial institutions. But as the AI principles launched by the Monetary Authority of Singapore (MAS) have begun to suggest, the more critical question is how consumers should be empowered to hold banks and insurers accountable for the ethical and accurate use of AI.

To unpack this, let's look at the common worry over AI today. This concern is that the use of AI would lead to greater discrimination in the financial system, as financial institutions look to further segment their clients by using data.

The ensuing worry is that individuals would be cut out from the financial system on the basis of their race, gender, age, and income, among other factors, thereby exacerbating levels of inequality in various forms.

It is important first to note the subtle but critical difference between discernment, which is acceptable, and discrimination, which is not.

The age-old practice of client segmentation by banks and insurers over the years has allowed customers and financial institutions to discern the benefits of a financial relationship shared between them. Different firms have different risk metrics, in the same way that different customers have different financing needs. AI comes in many shades, but at its core, it should help to predict more quickly and accurately if a financial institution should take on a potential customer, as has always been the case.

This discernment is not overtly problematic, so long as the variables in the assessment are fair and reasonable. If customers are rejected because they do not meet regulatory requirements or risk metrics in know-your-customer standards, for example, this should not cause overriding concern.

Indeed, as the AI principles from MAS showed, financial institutions should provide clear explanations on how data has driven final decisions on the pricing of a transaction.

For example, a firm selling car insurance should explain how a bad driving record may affect premiums - as is likely already the case now - so that a customer who is charged a higher premium can approach the insurer for a "meaningful explanation" on the reasons for the increase.

The principles also made clear that banks can continue to subscribe to customer segmentation when the product calls for it. A customer's age is an obviously relevant factor to discern on selling retirement-related financial services and products.

So discernment here should not be conflated with discrimination, which as a hypothetical example, is the damaging situation where silver-haired customers are automatically and sweepingly excluded from universal services such as e-payments just because they are older.

Once this is clarified then, what is most critical in the discussion of AI ethics, is in the accuracy behind its use. For example, to understand data is to know that correlation should not be confused with causality - which is to say, certain data points alone cannot fully determine behaviour.

The concern then, is how to prevent banks and insurers from using AI badly, and as an excuse to perpetuate financial exclusion.

The AI principles set out by MAS should eventually pave the way for consumers to probe the accuracy and suitability in the methodology of data assessment by banks and insurers.

Routine AI audit

The transparency may come to the point where in time financial institutions may have to open up their systems on an anonymised basis for a routine AI audit. The audit findings, which ideally should be public, would ensure that the data analysis is running as it should, based on kosher AI standards of relevant data input for meaningful data conclusions.

The temptation to discriminate, and the risk of inadvertent prejudice, exists when financial assessment of loans or insurance sits in a blackbox.

The rise of AI can shed new light, but only if financial institutions are held to higher standards of transparency, and as more power rests on consumers and an educated ability to question. Where necessary, consumers should also be able to opt out of transactions with poor data design. Regulations out of Europe are moving towards openness in this regard.

Greater consumer rights will also spur improvements today, AI or not. Customers will be able to take banks and insurers to task for financial exclusion if certain credit or insurance decisions today are indeed found to be based on discrimination and gut instinct, and not the judicious application of facts. This would make a strong basis for querying poor business ethics.

So given the uncertainty behind the use of AI, there ought to be moves to grant greater consumer protection rights in this area. In tandem with this, consumers will also need to educate themselves on better ways to scrutinise data design.

No doubt, such moves towards greater transparency are tricky and difficult to navigate. But it may also be the most resilient and responsible path that comes with opening up AI, the Pandora's Box of our time.