Artificial intelligence is a label that covers several different technologies. The underlying idea connecting them is that a computer system can produce outputs that would, if a human produced them, require intelligence of some kind. Recognising a face in a photograph, translating between languages, predicting which customers are likely to cancel a subscription, identifying the likely outcome of a medical image, and drafting a legal memo are all tasks that humans accomplish through intelligence, and AI systems now accomplish through different mechanisms that arrive at useful outputs.
Those mechanisms vary across different types of AI. The earliest AI systems worked by applying explicit rules that human experts wrote down. If a patient's temperature is above a threshold, and they report certain symptoms, the system would recommend a particular test. Later AI systems worked by learning patterns from data rather than following written rules. Given thousands of example photographs labelled as containing a cat or not containing a cat, a machine learning system would identify the statistical patterns that distinguish the two and apply those patterns to new photographs. The most recent AI systems, which power the tools most professionals encounter today, work by learning patterns from enormous quantities of text and using those patterns to generate responses to new questions.
What unifies these different types of AI is that they all produce useful outputs without the system itself understanding what it is doing in the way a human understands a task. An AI system does not know that a photograph shows a cat. It identifies that the photograph has statistical properties that, during training, were associated with the label cat. An AI system drafting a memo does not know what the memo is for. It produces the sequence of words that, given everything it has seen before, is most likely to follow the request the professional made. The outputs are useful. The mechanism that produces them is different from the mechanism of human thought, and understanding this distinction prevents the most common misunderstandings about what AI can and cannot do.
The current wave of AI is dominated by a specific type of system called a large language model. The tools the professional is most likely to encounter in daily work, whether they are named ChatGPT, Claude, Gemini, or something else, are large language models or systems built on top of them. Everything in this module builds toward a working understanding of what a large language model is, why it works, and why its emergence over the past several years has changed what AI can do for professional work.