The AI Book

CHAPTER 53Transforming Black Box AI in the Finance Industry: Explainable AI that Is Intuitive and Prescriptive

By Kathryn Rungrueng¹ and Ken So²

¹COO, Flowcast

²Founder and CEO, Flowcast

While artificial intelligence (AI) technology has become commonplace across industries, it has grown particularly integral in the evolution of the finance industry. Predictive technology and automation have streamlined nearly every facet of the space, whether looking at innovations in fraud detection, risk assessment, credit underwriting or identity verification. According to financial research firm Autonomous, the increased adoption of AI will support financial institutions by cutting 22% in costs over the next decade. Unfortunately, several obstacles remain when applying AI and machine learning models within finance. This is largely due to the complex regulations financial institutions must remain compliant with – these policies require that banks present a clear explanation for each and every decision, prediction or risk assessment.

As machine learning applications become increasingly specialized, though, the models become further opaque and thus more difficult to interpret. Logistic regression and other simple algorithms are part of a family of linear models that are globally explainable, with coefficients representing the marginal impact of each input. The marginal impact of the inputs on every prediction is the same globally for linear models. In contrast, with the recent developments ...