The most important technology of our time — decoded Millions use AI every day. Few understand why it works.

We are just at the start and I have not introduced any theory whatsoever. Before we embark on any systematic exploration, I want to give you a first idea of how things work behind the scenes. It will not be a deep technical explanation, simply an informal introduction to the workflow; the techniques come in the following chapters. We are going to apply this exercise to two examples: one involving analyzing a text and answering a question about it; the other showcasing one of the biggest successes of Modern-AI, automatic translation of human languages.

Equipped with a first informal glimpse at how AI-based systems work, we embark, in this chapter and the next, on general principles which underlie the entire technology. The present chapter singles out one principle which you need to grasp if you are to understand everything else. It is a kind of meta-principle, refined by some of the more concrete principles of the next chapter. If you are new to Modern-AI technology, it will only sink in as you read the rest of this book. It is the most offensive and misunderstood feature of Modern-AI. It would not have been applicable 20 years ago, and it may not hold in its exact present form 20 years from now. But in today’s AI technology it influences everything. …

Let not the words “Artificial Intelligence” fool you. Whether we judge it “intelligent” or not, today’s AI is the application of a specific set of technical mechanisms. This chapter explains the principles that underlie the whole approach. We already saw a “meta-principle”, Empiricism, in the previous chapter. As a preview, and roadmap for the present chapter, here are the other eight. …

“Artificial Intelligence” today covers a specific range of techniques, reviewed in the rest of this book. They are not the first to have used the term, which dates back to 1955; they are in fact radically different from the techniques that were tried in the first decades of AI. To understand Modern-AI, it is essential to form a good picture of what came before.

This book stays away from speculation and philosophy; it leaves it to others to expound on the deeper meaning of AI. Its goal is more pragmatic: to provide the essential knowledge enabling you to understand what Modern-AIreally is, under the hood. Only with this understanding can you see through the speeches (whether messianic or apocalyptic) and make up your own informed judgments. One abstract question, however, deserves our attention as our last preamble before we go into AI techniques: the notion of intelligence, the “I” in AI.

We start our study of Modern-AI systems by taking a look at today’s dominant architecture: neural networks. Specifically, “Feed-Forward Neural Networks”. It will be an opportunity to familiarize ourselves with the basic terminology. If you have been terrified of words such as “neural” and “neuron”, you will discover that they actually cover simple and clear concepts. …

“The math you need”. Do you need the math? Maybe you think “I can master the concepts without reading the formulas!”. Well, yes and no. True, lots of people pontificate about AI without having a clue. But, honestly — do you want to be like them? As the title of the book states, I want to you to understand AI. That requires understanding the basic math under it. …

I hope you took as much pleasure reading the preceding chapter about the necessary mathematical concepts as I had trying to make them as simple and clear as humanly possible, devoid of any unnecessary jargon or intimidation. Whether they were just a boring recap of things you knew since Kindergarten or before, a welcome refresher of stuff you vaguely remembered, or an obstacle course through which you had to struggle, the good news is that you now have what you need to understand the basics of Modern-AI systems. …

A neural network is a machine to produce answers. We want these answers to be the right ones; more precisely, in Modern- AI’s statistical approach, they should be right as often as possible. In the previous chapter we saw how a neural network functions, computing at each layer a function of the form NL (W * X + B). Modern-AI builds the network through machine learning, using a process described in the present chapter. …

The previous chapters have given us the overall architecture of neural networks — as simple in its basic ideas as it is impressive. That technology is the basis for most Modern-AI systems but other ideas complement it and, in particular, underlie the amazing performance of Large Language Models. They are particularly relevant for systems (such as LLMs) which deal with human-language texts as their inputs and outputs. They include in particular the topics of this chapter: how to encode tokens into “embeddings” (as previewed in our early example chapter) and a gradation of techniques for handling the full scope of a possible complex input: from convolution, recurrent and “long-short-term memory” networks all the way to the technology that made LLM success possible: attention and transformers. …

With Modern-AI, we have incredibly powerful tools at our command. Powerful — and imperfect. How do we use them right, to our benefit (and without undue detriment to others’ benefits)? This chapter collects practical guidance, essential if you want to use AI as part of an effective solution rather than letting it create new problems.

Passion around AI runs both ways, for and against. Our goal here, beyond passions, is to understand the negatives (some already encountered when the previous chapter examined which applications do not easily lend themselves to AI). Knowing the risks is the first step towards mitigating them — when at all possible, which as we shall see is not always the case. …

A part of our review of applications of AI deserves special attention: using AI tools, specifically Large-Language Models, to help learning. I have singled out this topic into a chapter of its own since its objective is to give a piece of advice — both simple and important — on how best to take advantage of LLMs for personal learning. The question of using AI in educational environments is separate, and the topic of the next chapter (15, page 219). Here we examine self-learning for adults and professionals, not education in schools and universities. There is no suggestion that such a process, or AI tools, remove the need for a teacher in standard education. …

No area of human activity may be more affected by Modern-AI than education. You hear enthusiastic predictions about how AI will revolutionize education, and doomsday assessments that it is killing education. In the general spirit of the rest of this book, this short chapter puts aside both the hype and the gloom and tries to derive some practical lessons. Often it seems that he enthusiasm comes from AI experts and others outside the educational system, and the doom from educators. Curious, is it not? …

AI is a product of software development; in turn, software development is one of the areas most spectacularly affected by the development of Modern-AI. “Affected” is a mild way of putting it: over just about a year (2023), AI tools revolutionized the practice of programming, leading to a form of development for which one of its promoters, Andrej Karpathy— in a 2025 post on X (Twitter) Karpathy 2025 — introduced a catchy name: vibe coding. Many people go further, stating that the world no longer needs programmers, the profession being doomed to follow the fate of horse-drawn carriage drivers and, more recently, draftsmen (experts at technical drawing), whose great skills were rendered obsolete by the appearance of the AutoCAD drafting software in the1980s. This chapter assesses the state of the art of applying AI to software development and the challenges ahead. …

The AI wave is not a passing fad. It may not be the messianic or apocalyptic event that its most extreme promoters and adversaries picture, but it is having a profound effect on the world, which will intensify as the technology develops in the coming years. This book has given you a solid handle on what it is about: its technical features, its benefits, its limits, its risks. We conclude our review with a few notes on how to adapt everyone’s personal career strategy to this radical innovation. …