This is the first entry in AI Fluency Corner — a 16-part weekly series intended to build a connected mental model of artificial intelligence (AI), one concept at a time, in plain language, for anyone who needs to understand what is actually happening.You open YouTube. You have not typed a word. Within seconds the homepage is full of things you actually want to watch — not because someone curated them, but because a system studied what you watched, how long you stayed, and what you replayed at midnight. It found the pattern. It made a prediction: this one.That is not coincidence. It is not magic. It is AI — not the cinematic kind with glowing red eyes and a grudge, but the quiet, pattern-spotting kind already woven into ordinary daily life.So what is it, precisely?AI is the use of computational systems to perform tasks requiring the kind of intelligence we traditionally associated with human beings — recognising speech, identifying images, predicting outcomes, generating language — by learning from data rather than following rigid rules.That last phrase is where it gets interesting. Traditional software is rule-based. A developer writes instructions for every foreseeable situation: if the invoice is overdue by 30 days, flag it; if the PIN is wrong three times, block the card. That is reliable, but completely helpless the moment something unanticipated happens. Conditions change. Rules break.AI systems work differently. Instead of rules they are given examples — thousands, sometimes billions. From those they learn to figure out the rules themselves.Think of it as four ingredients working in sequence.The first is data — the raw material. Location pings, clicks, purchases, search terms, photos, transaction records, voice commands. The cookie permissions you accept with the same attention you give an aeroplane safety demonstration. On its own, data is a warehouse; connected to a specific question, it becomes powerful.The second is an algorithm — Traditional software starts with instructions. A human first writes the rules, logic or code, and the computer follows them to produce an output. In simple terms: instruction/code → data input → execution → result.AI works almost in reverse. It usually starts with the desired output or task — the prompt — then draws on trained data through a model, performs inference, and only then produces a result. In simple terms: prompt → trained data → model → inference → result.That is why the quality of the result depends on the whole chain: the prompt, the data available, the strength of the model, and whether the right tool is being used for the task. Weak prompts, poor data, missing examples, or the wrong tool (Claude is better for reasoning, while ChatGPT is good with general knowledge, for example) can lead to hallucinated references, flawed reasoning or broken code.Nobody wrote those rules. The system inferred them. This is why AI can solve problems no programmer anticipated. It is also why it can confidently produce a wrong answer: the inferred rules are statistical, not logical. The system is completing a pattern, not following a proof.The third is a model — the trained artefact. Your navigation app has never driven a kilometre, yet it has absorbed enough data on routes, speeds and incidents to behave as if it understands congestion. It does not. It has learned to predict it. The difference matters. Because AI is built from human-shaped data, labelled examples, algorithms and trained models, it can simulate understanding, put things together and produce useful outputs — but it has no intent, independent judgment, true originality or agency. Its power is utility, not authorship.The fourth is inference — where the model earns its keep. It encounters something new and decides based on patterns from training. Training is school; inference is the exam. AI sits that exam millions of times a day without complaint.Every time you click ‘report spam’, you provide a live training signal. You are, without realising it, a part-time AI trainer.This learn-from-examples approach has a name: machine learning. Not a synonym for AI, but one of its most consequential methods. A spam filter is not given a dictionary of suspicious phrases — it is shown millions of labelled emails and finds the fingerprints itself.Every click of “report spam”’ is a live signal; you are a part-time AI trainer without realising it. YouTube runs the same way: every video you finish or abandon is feedback sharpening the next prediction.Pattern recognition at this scale is genuinely powerful. AI can identify a phrasing that tends to precede a complaint and flag the ticket before the customer gets to composing in capitals. It can sift transaction datasets for anomalies faster than a human analyst can find the right spreadsheet.In business, AI works where tasks are high-volume, data-rich and pattern-dependent: fraud detection, document classification, lead scoring, demand forecasting, call routing.And then there are the wallsAI systems do not think. They do not reason from first principles or hold values. When a language model produces a fluent paragraph it is completing a pattern — the most probable sequence of words for that prompt, not an opinion.Fluency is not truth. This is why the same system that drafts a polished memo can minutes later invent a nonexistent regulation or quote someone who never said it. Probable and correct are not the same thing.Fluency means knowing enough to ask the right questions: what was this trained on, what pattern is it applying, and what does it get wrong? Common-sense questions available to any reader, regardless of background.Tasks for the weekThe next time you open YouTube, pause before you scroll and ask — what does this system know about me, what data trained it, and what is it predicting?The next time you encounter an AI product — in a demo, a pitch, or your own phone — ask: what was this trained on? That single question separates an informed user from someone who is just along for the ride.• Mafinyani is senior partner in financial engineering & artificial intelligence at specialised finance, risk and applied technology firm Intellica Analytics. Next week’s article will cover data — the raw material everything runs on and the single biggest determinant of whether any AI project succeeds or fails.