Artificial Intelligence papers at this level usually run for 2 to 3 hours and carry 60–70 marks. A very common layout is one or two compulsory questions (often 15–20 marks each) plus any two of three or four optional questions; some schools instead ask you to answer any four of six. Each question is normally split into parts — (a), (b), sometimes (c) — worth 5 or 10 marks apiece. Work out your minutes per mark as soon as you sit down: on a 70-mark, 2-hour paper that's a little under 2 minutes a mark, so a 10-mark part deserves about 15 minutes and no more.

The questions have a clear personality. About half are "describe and justify": define weak and strong AI, write a PEAS for an agent, classify an environment, choose and defend an agent type, explain an expert system. These are usually dressed in a Nigerian scenario — a traffic system for a busy town, a farm drone, a clinic assistant, a bank's fraud detector — and the marks go to answers that use the scenario. The other half are "trace and translate": run Uniform Cost Search, Breadth-First Search or A* on a small graph, turn English sentences into propositional or first-order logic, or fire rules with forward chaining. Tracing questions are the kindest marks on the paper — they have one right answer, and if you show the frontier at every step you collect them all.

A calm strategy: answer the compulsory questions first while you're fresh, then choose the optional questions with the most tracing and translating in them, because those reward method rather than memory. In every describe-and-justify answer, tie each point to the scenario given — a PEAS that mentions okadas, potholes and harmattan dust beats a generic one every time. And leave five minutes at the end to check your arithmetic on any path cost.