AI in banking: Where artificial intelligence and machine learning may take us

With an aim to separate hype from reality in Day 4 at Sibos, I was on a mission to understand what the existing and near-term applications of Artificial Intelligence (AI) were in banking.  With machine learning described as “table stakes” now, Richard Harris (Feedzai) during The Ethical Side of AI panel, suggested that the closest we have to understanding the impact AI will have is by looking at the internet – knowing the internet would change everything but twenty years ago, we didn’t know how – describes the state of AI today.

Risk mitigation appears to be an active area for current AI application. For example, with a worldwide impact of money laundering estimates between 2% to 5% of global GDP (upwards of $2 trillion USD), Heike Riel, IBM (Sensemaker: The interconnectedness of everything and advanced AI) cited a case where they found a reduction in false positives of 95% to 50%, along with a reduction of 27% in manual effort by using AI/ML to help discover the undefined unknowns in the data. Using AI to help triage fraud for human interpretation and action is considered ‘narrow’ AI – the application of AI to one particular task.  

Broadening the scope of AI beyond a single task may be on the horizon. In the future I can see a time when an AI would become a new hire to the bank, employed to derive new, company-wide insights to improve processes, identify efficiencies or ways to improve customer experience.

As Ayesha Khanna (ADDO AI) mentioned in her breakfast keynote, we will need to be able to accept the insights from AI for this to be successful, and not dismiss them simply because we never thought of them before.

For now AI use is openly described for risk mitigation and advisory applications with a general expectation that this is only the beginning. And although AI begins with a use case – with a defined goal and data to learn from – ultimately the application of AI needs to create value. Currently value is focused on generating efficiencies, improving operations and cutting costs. But in the broader applications of ‘true AI’ we will likely need to reconsider how to measure value.

As Genevieve Bell put it during the closing plenary we will need question the metrics upon which we assess value, especially when considering autonomous applications of AI. Harkening back on previous industrial revolutions that created entirely new disciplines (like computer science during the 3rd industrial revolution) to this 4th industrial revolution powered data, AI, sensors and other advances she pointed out the likelihood of entirely new disciplines to form.  

Perhaps by then we’ll also have new metrics to ascribe value of AI  – like measuring the transparency, or trustworthiness of AI. The human doesn’t leave the equation in AI, for labeling data for example, but we may need to redefine how we treat it – possibly more, as Bell termed it during her session, a colleague than an algorithm.   To learn more about some of the people we are working with in AI, and their stories, don’t miss “The People behind OpenAI” from our Open Source Stories series.