WHAT ARTIFICIAL INTELLIGENCE (AI) IS AND IS NOT

A SHORT, FRIENDLY HISTORY The story of AI is a story of three big ideas, each building on the one before. The first idea was rule-based systems, also called symbolic AI. From the 1950s onward, researchers tried to capture human expertise as written rules. For example, a medical program might hold a rule such as: if the patient has a fever and a stiff neck, then suspect meningitis. Thousands of such rules were typed in by experts. These systems were useful in narrow areas, but they were brittle. Brittle means that the system breaks easily when it meets something its rules did not anticipate. Rules also have to be written and updated by hand, which is slow and expensive. The second idea was machine learning. Machine learning is a family of techniques in which a computer improves at a task by studying examples instead of being given explicit rules. Suppose you want a program that decides whether an email is junk. Instead of writing rules, you collect thousands of emails that people have labeled as junk or not junk. A machine learning method studies those examples and finds the patterns that tend to separate the two groups, such as certain phrases, odd sender addresses, or suspicious links. When a new email arrives, the program uses those learned patterns to make a guess. The third idea was deep learning. Deep learning is a kind of machine learning that uses neural networks with many layers. A neural network is a computing structure loosely inspired by the way brain cells connect, made of many small numerical units linked together, where the strengths of the links are adjusted during learning. Do not picture a real brain. Picture a very large set of adjustable dials. During learning, the dials are turned bit by bit until the network gives good answers on the examples. Deep learning became dramatically more effective in the 2010s, helped by two things: very large collections of data and powerful chips called Graphics Processing Units (GPUs), which you will meet again in Chapter 3. It produced big advances in image recognition, speech recognition, and game playing. Then, around 2017, researchers introduced a new neural network design called the transformer, which handled language especially well. Transformers made it practical to train LLMs, which are language-focused deep learning systems trained on huge amounts of text. Chapter 2 explains how LLMs work. Their arrival is what made the modern wave of AI assistants possible. Notice an important pattern in this history. Each new idea did not make the older ideas disappear. Plain rules are still the best choice for many jobs, and machine learning is still widely used. A good engineer picks the simplest tool that does the job, and later in this chapter you will practice making that choice.