“The greatest thing by far is to be a master of metaphor; it is the one thing that cannot be learnt from others; and it is also a sign of genius, since a good metaphor implies an intuitive perception of the similarity in the dissimilar.”

– Aristotle

“Ethics is in origin the art of recommending to others the sacrifices required for cooperation with oneself.”

– Bertrand Russell

“I am in the camp that is concerned about super intelligence.”

– Bill Gates

“The upheavals [of artificial intelligence] can escalate quickly and become scarier and even cataclysmic. Imagine how a medical robot, originally programmed to rid cancer, could conclude that the best way to obliterate cancer is to exterminate humans who are genetically prone to the disease.”

– Nick Bilton, tech columnist for The New York Times

This introductory chapter presents a detailed overview of the key terms related to explainable and interpretable AI that paves the way for further reading.

In this chapter, you will get familiar with safe, ethical, explainable, robust, transparent, auditable, and interpretable machine learning terminologies. This should provide both a solid overview for novices and serve as a reference to experienced machine learning practitioners.

This chapter covers the following topics:

• Building the case for AI governance
• Key terminologies – explainability, interpretability, fairness, explicability, safety, trustworthiness, and ethics
• Automating bias – the network effect
• The case for explainability and black-box apologetics

Artificial intelligence (AI) and machine learning have significantly changed the course of our lives. The technological advancements aided by their capabilities have a deep impact on our society, economy, politics, and virtually every spectrum of our lives. COVID-19, being the de facto chief agent of transformation, has dramatically increased the pace of how automation shapes our modern enterprises. It would be both an understatement and a cliché to say that we live in unprecedented times.

The increased speed of transformation, however, doesn’t come without its perils. Handing things out to machines has its inherent cost and challenges; some of these are quite obvious, while other issues become apparent as the given AI system is used, and some, possibly many, have yet to be discovered. The evolving future of the workplace is not only based on automating mundane, repetitive, and dangerous jobs but also on taking away the power of human decision-making. Automation is rapidly becoming a proxy for human decision-making in a variety of ways. From providing movies, news, books, and product recommendations to deciding who can get paroled or get admitted to college, machines are slowly taking away things that used to be considered uniquely human. Ignoring the typical doomsday elephants in the room (insert your favorite dystopian cyborg movie plot here), the biggest threat of these technological black boxes is the amplification and perpetuation of systemic biases through AI models.

Typically, when a human bias gets introduced, perpetuated, or reinforced among individuals, for the most part, there are opposing factors and corrective actions within society to bring some sort of balance and also limit the widescale spread of such unfairness or prejudice. While carefully avoiding the tempting traps of social sciences, politics, or ethical dilemmas, purely from a technical standpoint, it is safe to say that we have not seen experimentation at this scale in human history. The narrative can be subtle, nudged by models optimizing their cost functions, and then perpetuated by either reinforcing ideas or the sheer reason of utility. We have repeatedly seen that humans will trade privacy for convenience – anyone accepting End User Licensing Agreements (EULAs) without ever reading them, feel free to put your hands down.

While some have called for a pause in the advancement of cutting-edge AI while governments, industry, and other relevant stakeholders globally seek to ensure AI is fully understood and accordingly controlled, this does not help those in an enterprise who wish to benefit from less contentious AI systems. As enterprises mature in the data and AI space, it is entirely possible for them to ensure that the AI they develop and deploy is safe, fair, and ethical. We believe that, as policymakers, executives, managers, developers, ethicists, auditors, technologists, designers, engineers, and scientists, it is crucial for us to internalize the opportunities and threats presented by modern-day digital transformation aided by AI and machine learning. Let’s dive in!