Explaining Artificial Intelligence with Python

Algorithm explainability began with the first complex machines in the 1940s, the first being the Turing machine. Alan Turing himself struggled to explain how the intelligence of his machine solved encryption problems. Ever since machines have made calculations and decisions, explainability has been part of any implementation process through user interfaces, charts, business intelligence, and other tools.

However, the exponential progress of artificial intelligence (AI), including rule-based expert systems, machine learning algorithms, and deep learning, has led to the most complex algorithms in history. The difficulty of explaining AI has grown proportionally to the progress made.

As AI spreads out to all fields, it has become critical to provide explanations when the results prove inaccurate. Accurate results also require an explanation for a user to trust a machine learning algorithm. In some cases, AI faces life and death situations that require clear and rapid explanations. In this chapter, for example, we will study a case that requires an early diagnosis for a patient unknowingly infected by the West Nile virus.

The term explainable artificial intelligence or artificial intelligence explainability describes the explanatory process. We will often refer to explainable AI as XAI.

The goal of this chapter is to understand the key features of XAI and apply them to a case study in Python.

We will first define explainable AI (XAI) and the challenges we face when implementing explanations. Each profile requires different AI explanations. An AI expert will expect different perspectives than end users, for instance.

We will then explore a life and death case study in which AI and XAI constitute critical tools for a medical diagnosis of the propagation of the West Nile virus in Chicago, Illinois, USA.

The goal of XAI in this chapter is not to explain AI to developers but to explain the predictions of an ML system to a general practitioner enough for the ML decisions to be trusted.

We'll build the components of an XAI solution in Python from scratch using a k-nearest neighbors (KNN) algorithm and Google Maps Location History data, among other functions.

This chapter covers the following topics:

• Defining XAI
• The key features of XAI
• Describing a case study from the perspective of various profiles
• An executive function XAI feature chart
• The different forms and methods of XAI
• The XAI timeline from conception to production
• AI accountability
• The XAI timeline from a user's perspective
• k-nearest neighbors in Python
• Reading Google Maps Location History stored in JSON in Python

The first step will be to explore the key features of XAI before building an XAI prototype from scratch.