Here, we describe how the terms AI and ML relate to each other. It should be noted that these terms are often used interchangeably, as well as the abbreviated term, AI/ML, which serves as an umbrella term to encapsulate both AI and ML. We also describe how the terms DL and GenAI fit in under the umbrella of AI/ML.

We’ll begin by briefly including officially-accepted definitions of the terms AI and ML. We have chosen to include definitions from the Collins English Dictionary, in which AI is defined as “a type of computer technology concerned with making machines work in an intelligent way, similar to the way that the human mind works” and ML is defined as “a branch of artificial intelligence in which a computer generates rules underlying or based on raw data that has been fed into it.” The term DL has not yet been officially included as a dictionary term, but the Collins English Dictionary lists it as a...

If we traveled back in time by only a few years — to the year 2015 — and compared the state of the AI/ML industry to what it is today, we would see that relatively few companies had commercially implemented large-scale AI/ML use cases at that point. Although we would find academic research being performed in this space, we wouldn’t regularly hear AI/ML being discussed in mainstream media, and successful commercial or industrial implementations had mainly been achieved only by some of the world’s largest, industry-leading technology or niche companies. Jumping forward by just 2 years, we find that by the end of 2017, the tech industry is abuzz with discussions of AI/ML, and it seems to be the main topic—or at least one of the main topics—on everybody’s mind.

Based on our time-traveling adventure, one would not be faulted for believing that AI/ML is a brand-new term that suddenly emerged only in the past few...

AI/ML applications are usually intended to make some kind of prediction based on input data, with perhaps the exception of Generative AI, because Generative AI is intended to generate content rather than simply making predictions. In order to make predictions, ML models first need to be trained, and how they are trained depends on the approach being used. While ML is a broad concept that encompasses many different fields of research, with endless new use cases being created almost every day, the industry generally groups ML approaches into three high-level categories:

• Supervised learning (SL)
• Unsupervised learning (UL)
• Reinforcement learning (RL)

SL

SL is the most commonly used type of ML in the industry and perhaps the easiest to describe. The term supervised indicates that we are informing the ML model of the correct answers during the training process. For example, let’s imagine that we want to train a model to be able...

Mathematics is the hidden magic behind ML, and pretty much all ML algorithms function by using mathematics to find relationships and patterns in data. This book focuses on practical implementations of AI/ML on Google Cloud; it is not a theoretical academic course, so we will not go into a lot of detail on the mathematical equations upon which ML models operate, but we will include mathematical formulae for reference where relevant throughout the book, and here we present some basic concepts that are widely used in AI/ML algorithms. There are plenty of academic materials available for learning each of these concepts in more detail. As an architect, understanding the mathematical concepts could be considered an extracurricular credit rather than a requirement; you usually would not need to dive into the mathematical details of ML algorithms in your day-to-day work, but if you want to have a better understanding of how some of the algorithms work...

Companies typically run into common kinds of challenges when they embark on an AI/ML development journey, and it is often a key requirement of an architect’s role to understand common challenges in a given problem space. As an architect, if you are not aware of challenges and how to address them, it’s unlikely that you will design an appropriate solution. In this section, we introduce the most frequently encountered challenges and pitfalls at a high level, and in later sections of this book, we discuss ways to address or alleviate some of these hurdles of AI/ML development.

Gathering, processing, and labeling data

Data is the key ingredient in ML because, in general, ML models cannot function without data. There’s an often-quoted adage that data scientists spend up to 80% of their time working on finding, cleaning, and processing data before they can begin to make use of it for analytical or data science purposes...

In this chapter, we introduced basic terminology related to AI/ML and some background information on how AI/ML has developed over time. We also explored different AI/ML approaches that exist today and some of their applications in the real world. Finally, and perhaps most importantly, we summarized common challenges and pitfalls that companies typically run into when they begin to implement AI/ML workloads.

In the coming chapters, we will dive deeper into the model development process.