What is AI?
Artificial intelligence is the umbrella term for computer systems that perform tasks normally requiring human intelligence: recognising a face, translating a sentence, spotting a pattern in data, holding a conversation. It is not one technology or one product. It is a category, and the systems inside it range from a simple spam filter to a chatbot that can draft an essay.
An umbrella, not a thing
The label “AI” covers a wide spread of tools that share almost nothing at the engineering level. A spam filter deciding whether an email is junk is AI. A recommendation engine suggesting what to watch next is AI. A self-driving car reading the road ahead is AI. So is a chatbot answering a question in plain language. These systems solve completely different problems using different methods, yet all sit under the same term because each one automates a task that used to require a human judgement call.
The current wave of excitement is really about one branch of this family: generative AI, and within it, language models — systems trained to produce text, and increasingly images, audio and code, in response to a prompt. ChatGPT is the name most people recognise here, though it is one product among several performing a similar role; others include Gemini, Copilot and Claude. When people say “AI” in conversation today, they usually mean this branch. The broader term is much bigger than that.
Narrow, not general
Researchers generally describe the field along a narrow-versus-general axis, and every AI system in everyday use sits firmly on the narrow side. A narrow AI is very good at a specific task and nothing beyond it: a chess engine plays exceptional chess and cannot hold a conversation; a language model can write fluent prose but has no body, no senses and no persistent life outside the exchange it is having with you.
“Artificial general intelligence” — a system that matches human capability across the full range of tasks, not just one — does not exist. It is a long-standing goal in the field, and researchers at institutions such as Stanford HAI treat it as genuinely contested: some see it as a natural extension of current techniques, others as a different problem altogether that scaling up today’s approach will not solve. Whether it is achievable, and on what timeline, remains an open question rather than a settled prediction.
Why now
AI as an idea is old; the techniques behind today’s systems were proposed decades before they became useful. What changed was scale. Far larger amounts of data and far greater computing power became available and were applied to that older idea, and capabilities that had barely worked in the lab suddenly worked well enough to ship as products. This is a description of a trend, not a single breakthrough moment — the shift happened as data, computing power and technique improved together, each making the others more effective.
That combination is why AI moved from a specialist research topic to something showing up in search boxes, phones and office software within the span of a few years. The underlying mathematics did not change overnight; the resources behind it did.
Where this leaves you
AI is a tool, and like any tool it has sharp edges. It can produce fluent, confident-sounding output that is wrong. Researchers debate, and NIST frameworks for trustworthy AI explicitly flag, questions about reliability, bias and the limits of what these systems actually “understand” — a word used cautiously here, since whether a language model understands anything in the way a person does is itself disputed rather than settled. Treat claims about what AI “knows” or “thinks” as descriptions researchers argue over, not established fact.
The rest of this topic goes into how the current generation of tools actually works, and where their failures come from.