COMPSCI110 · Introduction to Computer Systems
Artificial Intelligence
Module 11 of University of Auckland COMPSCI110 closes the course with the question 'can machines think?' It covers the Turing test and the Winograd schema challenge, the kinds of tasks AI tackles, and an introduction to neural networks and the back-propagation learning algorithm. It is a conceptual module — there are no numeric worked calculations — and it carries 5 marks on the final exam, typically as multiple-choice and multiple-response questions, so it rewards precise definitions rather than computation.
What this chapter covers
- 01The Turing test (imitation game): a machine passes if a human interrogator cannot reliably tell it from a person by conversation alone
- 02The Winograd schema challenge: commonsense pronoun-resolution problems used as an alternative test of machine understanding
- 03Types of tasks AI addresses and what makes them hard for machines
- 04Neural networks: layers of simple artificial neurons connected by weighted links (conceptual — no numeric calculation examined)
- 05Back-propagation: the learning algorithm that adjusts the weights by propagating error backward through the network
- 06Robots and embodied AI as an application area
- 07Assessment style: conceptual MCQ / multiple-response questions — do not expect a numeric neural-network computation
Applying the Winograd schema (a conceptual, reasoning-marked item)
- +1Identify the ambiguity: the pronoun 'it' has two grammatically valid referents — the trophy or the suitcase. Sentence structure alone does not decide which.
- +1Resolve with world knowledge: 'too big' explains why something doesn't fit, and it is the trophy that fails to fit, so 'it' = the trophy. This uses commonsense physical reasoning, not grammar.
- +1Why it's hard for AI: a system needs real-world understanding (a larger object won't fit inside a smaller container) to pick the referent — pattern-matching on syntax is not enough, which is exactly what the Winograd schema is designed to probe.
- +1Link to the Turing test: swapping 'big' for 'small' flips the answer to the suitcase, showing the resolution depends on meaning, not wording. The Winograd schema is offered as a sharper, less-foolable alternative to the conversational Turing test for judging machine understanding.
Key terms
- Turing test
- Alan Turing's 'imitation game': a human interrogator holds text conversations with a hidden machine and a hidden person and must tell which is which. If the interrogator cannot reliably distinguish them, the machine is said to pass — a behavioural test of intelligence based on conversation rather than on how the machine works inside.
- Winograd schema
- A carefully constructed sentence with an ambiguous pronoun whose correct referent requires commonsense world knowledge to resolve, with a one-word change that flips the answer. Proposed as an alternative to the Turing test because it is hard to pass by trickery or canned conversation — it targets genuine understanding.
- Neural network
- A computing model loosely inspired by the brain: layers of simple artificial neurons connected by weighted links, where each neuron combines its weighted inputs to produce an output. Passing data through the layers lets the network learn to map inputs to outputs. In this course it is introduced conceptually, without numeric calculation.
- Back-propagation
- The standard learning algorithm for neural networks: after the network makes a prediction, the error at the output is propagated backward through the layers to adjust the connection weights so the next prediction is closer. Repeated over many examples, this is how the network 'learns'. Examined conceptually here, not as arithmetic.
- Weight (connection)
- A number on a link between two neurons that scales the signal passing along it, determining how strongly one neuron influences another. Learning in a neural network is essentially the process of finding good weights — which is what back-propagation adjusts.
- Task types in AI
- The categories of problem AI addresses — such as perception, reasoning, language and learning — and the recognition that tasks easy for humans (commonsense reasoning, understanding an ambiguous sentence) can be very hard for machines, while some tasks hard for humans are easy for machines. Knowing what makes a task hard for AI is the examinable idea.
Artificial Intelligence FAQ
What is the Turing test?
It's Alan Turing's proposal for deciding whether a machine can 'think', framed as an imitation game: a human judge converses by text with a hidden machine and a hidden human and tries to tell them apart. If the judge can't reliably identify the machine, it passes. The key feature is that it judges behaviour (conversation) rather than internal mechanism — a machine passes by acting indistinguishably from a person, regardless of how it does so.
What is a Winograd schema and why does it matter?
It's a sentence with an ambiguous pronoun that can only be resolved using commonsense knowledge, plus a small wording change that reverses the correct answer (e.g. 'too big' vs 'too small' flips which object 'it' means). It matters as an alternative to the Turing test: it's much harder to pass by conversational tricks because getting the pronoun right genuinely requires understanding the world, so it's a sharper probe of machine comprehension.
Can AI help me with the AI module in COMPSCI110?
Yes, as a study aid — with a nice symmetry. Sia can explain the Turing test and Winograd schema, give you fresh ambiguous sentences to reason about, and describe how a neural network and back-propagation work at a conceptual level. Use it to sharpen the definitions and rehearse the reasoning — it explains and checks your understanding but does not do graded assignments or the exam for you, and the University of Auckland's academic-integrity rules apply to controlled assessments.
Do I need to do neural-network maths for the exam?
No. Module 11 is introduced conceptually, and it's examined as multiple-choice and multiple-response questions about ideas — what a neural network is, what back-propagation does, the Turing test and Winograd schema. There's no numeric neural-network calculation to perform, so don't spend time inventing one; spend it on being able to state each concept precisely and explain why certain tasks are hard for machines. Confirm the exam style on Canvas.
Is Module 11 on the final exam?
Yes — Artificial Intelligence carries 5 marks on the final exam, typically as multiple-choice and multiple-response items. Because it's the conceptual close of the course, the marks go to clear definitions and reasoning rather than computation, which makes it efficient to revise: tight flashcards on the Turing test, the Winograd schema, neural networks and back-propagation cover most of it. Confirm the current per-module breakdown on Canvas.
Exam move
Module 11 is 5 conceptual marks, so revise it with precise definitions rather than practice calculations — and don't waste time inventing neural-network arithmetic that the exam doesn't ask for. Make tight flashcards for the four pillars: the Turing test (a behavioural, conversational test of intelligence), the Winograd schema (commonsense pronoun resolution, a harder-to-fool alternative), what a neural network is (layers of neurons joined by weighted links), and what back-propagation does (adjusts weights by sending error backward to learn from examples). Be ready to explain, in a sentence each, why a task can be easy for a human but hard for a machine, and why the Winograd schema resists the tricks that can fool a Turing-test conversation. Because the questions are multiple-choice and multiple-response, practise discriminating close statements — the distractors turn on subtle wording — and confirm any offering-specific emphasis on Canvas. This is a good module to lock in early since it needs understanding, not drilling.
Working through Artificial Intelligence in COMPSCI 110? Sia is AskSia’s AI Computer Science tutor — ask any COMPSCI 110 Artificial Intelligence question and get a clear, step-by-step explanation grounded in how COMPSCI 110 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
