Scope of explainable AI in contemporary AI/ML applications

Abstract

Artificial Intelligence has become an indispensable part of corporate decision making today. Integration of AI algorithms in business activities have led to reduction in inefficiencies and a net improvement in customer experience and higher revenues. However, AI has its own set of challenges, the biggest one being its black-box nature. While the scope and depth of AI applications have increased multi-fold with institutions like banks, governments, hospitals, etc. relying on it on a daily basis, it is often difficult to explain the exact functioning of AI models. The inner workings of AI models are perceived to be very complex by humans and this subsequently leads to limited application of AI. The understanding of AI’s inner functioning in ‘human terms’ is of paramount importance so as to weed out model demerits and inherent biases1. This is where 'Explainable AI' comes in. These methods help us understand the logic and reasoning behind each prediction made by the AI model. It provides specific information on how a certain decision can be attributed to a certain feature/set of features used in the AI model. The ‘explainable AI’ brings a sense of accountability and trust in business, allowing for more inclusive and ethical AI2. Further, explainability makes it easier to spot the model's shortcomings and fix them to ensure that the AI stays within the moral and ethical limits.