An artificial neural network is layers of very simple units, joined by connections that each carry a weight. Numbers go in, pass through the layers, and a number comes out. Training adjusts the weights.
A small network, what training changes, and why the middle cannot be read.
Interactive diagram
Explore the labelled diagram, charts and maps for this topic in full study mode.
Free preview
This is the free notes preview
You're reading the free notes. Aimnova Pro unlocks the full study experience — and you can try it with your first topic free to keep:
- FlashcardsLock in vocabulary and key terms with spaced repetition.
- Practice questionsAnswer exam-style questions and get instant AI marking.
- Mock exams & past-paper vaultSit full mocks and see exactly how examiners award marks.
- Personalised study planA daily plan built around your exam date and weak areas.
The guide's own phrase is that they learn and model complex, non-linear relationships and generalize from initial inputs. In plain words: they find patterns too tangled to write down as a rule, and then apply them to things they have not seen.
What that buys
- Recognising a face at an angle nobody photographed during training.
- Reading handwriting that looks nothing like the training examples.
- Predicting a protein shape from a sequence never measured before.
- Translating a sentence nobody has ever written.
Generalising is the whole point: A system that only handled cases it had already seen would be a lookup table. The value — and the danger — is that it produces an answer for inputs it has never met.
Practice with real exam questions
Answer exam-style questions and get AI feedback that shows you exactly what examiners want to see in a full-marks response.
This is the part that matters for digital society. A trained network contains no rules and no words — only millions of numbers. There is nothing in there to read out.
A written rule
- Can be read by anyone shown it
- Can be argued with, line by line
- Can be changed deliberately
- Its reasoning IS the rule
A trained network
- Millions of weights, no sentences
- Can be tested, but not read
- Changing it means retraining, not editing
- There may be no reason to give
Real-world examples you can name
COMPAS recidivism scoring — ProPublica investigation May 2016
A commercial risk-scoring tool used in some US courts to estimate how likely a defendant was to reoffend. Journalists found that among people who did not go on to reoffend, Black defendants were roughly twice as likely as white defendants to have been labelled high risk. The company disputed the measure of fairness used, and the argument that followed showed that competing definitions of a fair algorithm cannot all be satisfied at once.
Who it affected: Defendants whose bail and sentencing decisions were informed by the score.
The EU Artificial Intelligence Act — agreed December 2023; entered into force August 2024, phased to 2027
The first broad law to regulate AI by risk. Some uses are banned outright, such as untargeted scraping of facial images and social scoring by public authorities; high-risk uses in areas like hiring and education carry duties for data quality, documentation and human oversight.
Who it affected: Anyone deploying AI systems in the EU, and users subject to those systems.
How this is tested — you have to describe what a network does and explain why explaining its decisions is hard. It comes up two ways:
Paper 1 — structured question
- Part a: identify characteristics of a neural network
- Part b: explain why a decision cannot be explained
Paper 2 — source-based question
- Q2: explain a claim a source makes about a model
- Q4: synthesise sources on algorithmic accountability
The trap: “it thinks like a brain”: The name is an analogy and a loose one. Saying a network “thinks” invites an examiner to doubt everything else in the answer.
Identify two characteristics of an artificial neural network.
Model answer plan
See the mark-by-mark plan — for / against / judgement, with marking guidance — in study mode.
Explain one reason why the decision of a neural network can be difficult to explain.
Model answer plan
See the mark-by-mark plan — for / against / judgement, with marking guidance — in study mode.