LATENT LIBRARY FUNDAMENTALS AI Foundations How Machines Learn, ExplainedWithout the Jargon LATENT LIBRARY EDITORIAL

AI Foundations

How Machines Learn, Explained Without the Jargon

Latent Library Editorial

IntroductionWhy AI Feels Like Magic

A few years ago, most people met artificial intelligence only in films. Today it writes emails, recognises faces in photos, translates conversations, recommends what to watch next and answers questions in fluent paragraphs. Because it arrived so quickly, and because it often talks like a person, AI can feel like magic. Magic is a poor foundation for decisions. If you do not know how something works, you cannot tell when to trust it, when to double-check it, or when to leave it alone.

This book removes the magic without removing the wonder. It explains, in plain language, how modern AI systems are built, how they learn, why they succeed and why they fail. You will not need any mathematics beyond everyday arithmetic, and you will not need to write code. Where a technical word is unavoidable, it is explained the first time it appears and collected in the glossary at the end.

The book follows a simple path:

Practice questions with answers follow the last chapter, and a glossary closes the book.

Chapter 1What "Artificial Intelligence" Actually Means

A moving target

"Artificial intelligence" is an old phrase. It was coined in the 1950s for a research programme with a bold goal: to make machines that could do tasks that seem to require intelligence when humans do them. Over the decades the label has been attached to very different technologies: chess programs, expert systems full of hand-written rules, spam filters, route planners and, most recently, systems that generate text and images.

There is a well-known pattern here. As soon as a task is solved by a computer, people tend to stop calling it intelligence. Route finding on a map was once an AI research problem; now it is simply "the satnav". This is a useful reminder that AI is not one thing. It is a broad family of techniques for getting computers to handle tasks that are hard to describe with explicit step-by-step instructions.

A working definition

For this book we will use a practical definition:

An AI system is a computer system that produces outputs such as predictions, recommendations, decisions or content, by applying patterns it has learned from data, rather than by following rules a programmer wrote out by hand for each case.

Notice two things. First, the definition is about behaviour (producing useful outputs), not about consciousness, feelings or understanding in the human sense. Second, the heart of the definition is learning from data. That is what separates today's AI from ordinary software.

Narrow and general

Every AI system in use today is what researchers call narrow: it is built and trained for a particular kind of task. A model that detects tumours in X-rays cannot plan a holiday. Language models are broader than earlier systems, since one model can summarise, translate, draft and explain, but their abilities still come from, and are limited by, the data and training they received.

Artificial general intelligence (AGI) refers to a hypothetical system that could learn and perform any intellectual task a human can, across domains, with the flexibility of a person. Whether, when and how such a system might be built is debated among experts. For practical purposes you can treat any AI you encounter today as a powerful but specialised tool.

Three things AI is not

  1. AI is not a database of facts. Many systems, especially language models, do not look up answers in a table of verified facts. They generate outputs from learned patterns, which is why they can be fluent and wrong at the same time.
  2. AI is not a person. A system that writes "I think" is producing text that resembles how people write. It has no personal stake in being right and no memory of you unless it was specifically built to have one.
  3. AI is not neutral by default. It learns from data collected by people, in a particular time and place, for particular purposes. Patterns in that data, including unfair ones, can be learned and repeated.

The main branches

You will meet these terms often. Here is how they relate:

TermWhat it meansExample
Artificial intelligenceThe whole field of making machines perform tasks that seem to need intelligenceEverything below
Machine learning (ML)The part of AI where systems learn patterns from dataA spam filter trained on labelled emails
Deep learningMachine learning with large, many-layered neural networksImage recognition, speech-to-text
Generative AIModels that produce new content: text, images, audio, codeChatbots, image generators
Large language model (LLM)A very large deep-learning model trained on text to predict the next piece of textChat assistants

Each row is a subset of the one above it. All LLMs are generative AI; all generative AI here is deep learning; all deep learning is machine learning.

You have reached the end of the free sample.
The full book continues with 14 more chapters, plus everything else in its 81 pages.

Get the full book