Read a sample: 1. Why AI Has Its Own Language
Imagine that you are planning a small community garden. You have a patch of ground, a few volunteers, and a question: “What vegetables could we plant in a sunny garden if we want to harvest something within two months?”
You type the question into an AI assistant. In a few seconds, it suggests radishes, leafy greens, and several other possibilities. It organizes the suggestions into a tidy list and adds a few planting tips. The answer looks useful. It may even help you decide what to discuss at your next garden meeting.
But a second set of questions begins as soon as you look beyond the list. How did the assistant produce it? Did it check current growing advice, or did it respond from patterns learned earlier? Does it know where your garden is? If it names a vegetable that is unsuitable for your season, is that a hallucination, an ordinary mistake, or a consequence of missing information?
These are not questions only for programmers. They are the kinds of questions anyone can ask when using a product, reading a headline, or deciding whether to trust a recommendation. The difficulty is that the answers often arrive wrapped in unfamiliar terms: model, training data, tokens, parameters, context window, inference. The vocabulary can make a simple encounter with a garden question feel like a technical examination.
This book is about learning enough of that language to see what is being claimed—and what is not. You do not need to write code or memorize equations. You do need a few distinctions that make the technology less mysterious and the claims around it easier to evaluate.
Our garden question will return throughout the book. It gives us one ordinary request to follow as we examine how an AI system might receive information, produce an answer, make a mistake, or use an outside source. The example is deliberately modest. If we can understand the language in a familiar setting, we can apply the same habits to less familiar ones.
Artificial intelligence, usually shortened to AI, is a broad term for computer systems designed to perform tasks associated with abilities such as recognizing patterns, interpreting language, making predictions, or generating content.
That definition covers a great deal. A system that identifies likely spam in an email inbox can be described as AI. So can a program that recommends a film, software that helps analyze an image, or an assistant that drafts a planting plan. These systems do not all work in the same way, and success at one task does not imply skill at another.
Think of AI as the name of a large neighborhood, not the address of a particular house. Saying that a product “uses AI” tells you that its maker places it somewhere in that neighborhood. It does not tell you how the product was built, what information it uses, how often it makes mistakes, or whether it is suitable for your garden question.
The term also carries baggage because intelligence means different things in different conversations. In everyday speech, calling a person intelligent might suggest understanding, judgment, memory, and the ability to learn from experience. When a company calls software intelligent, it may mean that the software performs a useful task that would be difficult to accomplish with a simple, fixed set of instructions. The shared word should not lead us to assume the abilities are identical.
That does not make the technology uninteresting. A system can be genuinely useful without thinking like a person. It can also produce an impressive answer without having every ability that the answer seems to imply. Keeping those possibilities separate is one of the most useful habits you can develop as a reader of AI claims.
One way to make a computer perform a task is to give it explicit rules. Imagine a very simple garden helper built with instructions such as: if the user asks for a quick crop, display a prepared list of quick-growing vegetables. The result might be useful within narrow boundaries, but a person has written the rule and chosen the list in advance.
Machine learning is a related but more specific term than AI. It refers to approaches in which a system learns patterns from data rather than relying only on rules written out for every situation. The data might consist of examples, measurements, text, images, or other material, depending on the task.
For a simple illustration, imagine a system developed to sort photographs into two groups: those likely to show a garden and those unlikely to show one. During development, it might be shown many examples so that it can learn patterns associated with each group. A person does not need to write a separate rule for every possible arrangement of soil, leaves, paths, and fences.
The result is still imperfect. A photograph of plants in a shop might resemble a garden, while a winter garden with bare beds might be hard to recognize. Learning from examples can make a system flexible, but it does not guarantee that the patterns it learns are the ones a person intended.
Machine learning is commonly treated as a branch of AI. Not every AI system is best described as machine learning, and not every machine-learning system produces conversation or creative material. Some make predictions or classifications, such as estimating whether a message is likely to be unwanted. The label tells you something about…