Artificial Intelligence’s Role in Banking 3.0
By Richard Shearer, CEO of Tintra PLC
In the modern banking world new technologies play a widely reported role in anti-money laundering (AML) protocols preventing financial crime – however it is important that we do not overlook technology’s potential for establishing financial innocence.
To businesses and institutions operating in and between developed markets, whose international transactions are fast and painless, this sentiment may seem counter intuitive. AML compliance is necessary for regulatory reasons, and catching out bad actors is, of course, a primary goal of any business – but why should we view AML technology through the lens of establishing innocence?
This is a question which emerging market corporates will have no difficulty answering if they have ever attempted to interface with counterparts in developed markets.
Entities based in emerging markets are often tarred with the brush of AML risk due to their geography and unrelated to their specific business, and – consequently – such organisations find international transactions lengthy, arduous and expensive as they navigate an AML compliance process that operates from a base level that is an unfair assumption of their risk.
As such, in my view, embracing advances in technologies such as natural language processing (NLP) and machine learning (ML) is essential – not only for financial services firms looking to enhance their ability to properly mitigate, but to progress the much bigger, and indeed more noble, goal of removing the biases against emerging markets, nationalities or cultures that currently colour the AML landscape.
How then, can NLP and ML technologies help, not only in addressing financial crime, but in creating an environment where those in emerging markets with upstanding credentials are treated and serviced free from these baked in prejudices?
It’s worth taking a moment to define these terms.
“Natural Language Processing” pertains, in broad terms, to anything related to processing language. As such, NLP varies in terms of complexity – it may be employed for tasks like “term frequency”, calculating how often a given word appears in a text, but NLP can equally be used for the purposes of translation; classifying the sentiment of a piece of text; or even detecting sarcasm, irony, and fake news in a social media context.
In order to perform the more complex tasks in this spectrum however, machine learning may also be required.
“Machine Learning” describes a variety of artificial intelligence (“AI”) with an emphasis on allowing machines to learn in a similar manner to humans, through a mix of data and algorithmic methods.
ML differs from traditional programming. Traditional programs see a solution to a problem defined through hand-crafted rules that are implemented in computer code. In ML, by contrast, the algorithm itself learns those rules – and, by extension, how to solve the problem by analysing data.
This principle makes ML considerably more powerful than traditional programming, since it is capable of learning a complicated sets of rules that are impossible to define manually.
AML applications of these Technologies
In the context of AML practices, it’s not difficult to see the appeal of technology like this.
After all, manual investigations into potentially rogue activities are lengthy processes which involve employees investigating vast swathes of transaction histories and other information – and often only happen after the event.
This process is made all the more difficult to manage for financial institutions when a large number of “suspicious incident” alerts are often false alarms. But each potential issue must be investigated with the same vigour to ensure a robust AML framework.
By contrast NLP/ML allows financial institutions to automate these processes – the more sophisticated solutions, that my team and I are very focused on, are capable of interpreting the vast amounts of text-based data that a human would otherwise need to analyse.
These systems are able to recognise patterns and relevant information, consider appropriate context and cross reference faster and more accurately than a human, or indeed teams of humans, may overlook.
Crucially for me, NLP/ML – performed by intelligent machines capable of learning – can potentially undertake these tasks at the same time as neutralising human bias, which has promising implications for organisations and individuals in emerging markets who face these preconceptive biases frequently.
Less human, more humane
This application of NLP/ML has a range of benefits for all stakeholders, not least with reductions in the level of false positives representing savings in time and money for financial services companies.
There is, however, equal value to be found in NLP/ML tools which bring this power to bear on addressing the inequities that currently prevent frictionless transactions between these markets.
This piece began with reference to establishing cases of financial innocence as well as financial crime – and, while NLP/ML makes this possible, it would be wrong to assume that such tools will magically resolve the issue of AML bias.
As such, “establishing innocence” isn’t just a different perspective on the benefits of NLP/ML solutions – it’s an ethos that I believe should be actively pursued by financial services businesses as our global economy becomes more and more integrated.
Removing human prejudice from the decision-making process is vital, but a truly fair approach can only be achieved when the creators of these solutions acknowledge that the prejudice exists in the first place.
After all, NLP/ML is entirely subject to bias – or “algorithmic unfairness,”. A good example – taken from research published in ACM Computing Surveys – is a piece of software called COMPAS, used by US judges to assess offenders’ risks of reoffending, which was found to exhibit bias against African-American individuals – illustrating clearly that human prejudice can inflect algorithmic decision-making. To make the technology better we need to be better, is may be one way of thinking about it.
This kind of example gives food for thought. If NLP/ML tools are trained without thought being given to how to eradicate bias in an AML context, then we’ll be left with intelligent machines that simply replicate that bias – meaning that prejudice will be automated rather than eliminated! A terrifying concept and one fraught with complex ethics.
The next step
The financial services sector is in the midst of digital transformation and as such the time is ripe to seize the wheel and ensure, as we embrace more sophisticated tech solutions, that the journey ends at a fair and equitable destination – no matter where a given transaction takes place.