Five Chapters In!
Last week I sent the first five chapters of my book Apress. The working title: The AI Handbook for Product Design; I don't love it yet... but we shall see – I may change my mind, or come up with a better title. My contract asked for three chapters by July 24, but I wrote five. While exploring the depths of AI theory and history, I concluded that Part 1 only holds together as a complete arc, and stopping in the middle of it would have handed my editor a story without an ending. Also... I was just too excited to stop, and to afraid that without the complete narrative it's hard to see where I am going. So here we are. 5 chapters are done.
What is my book about? It is a guide for designers, product managers, and technologists who want to build the next generation of AI-powered software. Part 1 is about how intelligence became computable in the first place. I wanted the foundation to feel complete, interesting, and intellectually challenging.
Chapter 1 asks what intelligence actually is, and takes the question seriously enough to go back to Aristotle, then to the long argument between Descartes and Locke about where knowledge comes from, then to Kant, Piaget, and Friston. It took me back to my roots in Philosophy and Psychology – I happen to have degrees in both. The chapter ends on a working definition that carries the rest of the book: intelligence is the ability to compress experience into models that predict well enough to support effective action in a changing world.
The next four chapters follow that idea as it turns into machinery. Chapter 2 covers classical machine learning, from Rosenblatt's perceptron through the first AI winter to the algorithms that quietly powered the first generation of intelligent products. Chapter 3 is about deep learning and the moment machines stopped needing humans to tell them what to look at. Chapter 4 covers transformers and language models, including the property I consider the most important and most dangerous thing a designer can know about them, which is that a model is exactly as articulate when it is wrong as when it is right. Chapter 5 is about diffusion models and the shift from recognition to creation.
The content in each chapter is written from a very particular angle – I want designers, researchers and product managers to really understand the underlying concepts and being able to apply them in practice. This led me to two writing rules.
The first is that every technical concept has to earn a design implication. If I explain attention or latent space or the bias-variance tradeoff and cannot tell you what it changes about the interface you are building on Monday, the section does not belong in the book. Each chapter now mints a small number of named design principles, and they accumulate across Part 1 into a set you can carry into Part 2.
The second is that I refused to skip the math. Yep, it may be the first UX and product book that is filled with math. I'm taking some risks here. You do not necessarily have to have a deep understanding of linear algebra, and calculus to read the book, but you'll see the mathematical foundation of AI, and hopefully, grasp how all the concepts fit together.
Part 2 is next, and it moves from how AI works to how you design with it: retrieval, reasoning, multimodal interfaces, and agents. I will keep writing about it here as I go.