Artificial intelligence (AI) refers to computer systems designed to perform tasks that typically require human intelligence, like recognising images, understanding language, or making predictions. Machine learning, a common approach to AI, involves training a model on large amounts of data so it can identify patterns and make predictions on new data — rather than being explicitly programmed with fixed rules for every situation.
Example
A spam email filter using machine learning is trained on thousands of examples of spam and non-spam emails, learning patterns (like certain phrases or sender behaviours) that typically indicate spam — rather than a programmer writing out every possible spam rule by hand.
Key terms
Artificial intelligence (AI):
Computer systems designed to perform tasks typically requiring human intelligence.
Machine learning:
An approach where a model learns patterns from data, rather than following fixed rules.
Training data:
The data used to teach a machine learning model to recognise patterns.
Questions
1. Artificial intelligence refers to:
Systems performing tasks typically requiring human intelligence
Only physical robots
A type of hardware component only
A password manager
2. Machine learning involves:
Training a model on data to find patterns
Writing every rule manually with no data used
A type of hardware only
Ignoring all data
3. Training data is:
The data used to teach a model to recognise patterns
A type of computer virus
Always irrelevant to AI
A type of password
4. A spam filter using machine learning is trained using:
Examples of spam and non-spam emails
No data at all
Only spam emails, never non-spam
A single hardcoded rule only
5. AI systems can be used for tasks like:
Recognising images or understanding language
Only performing basic arithmetic
Nothing useful
Only playing music
6. Machine learning models identify:
Patterns in data
Nothing meaningful
Only random noise
Hardware specifications only
7. Unlike traditional fixed-rule programming, machine learning:
Learns patterns from data rather than following pre-written rules for every case
Never uses any data at all
Is identical to fixed-rule programming
Cannot make any predictions
8. Why might a machine learning approach be better suited than fixed rules for a task like recognising handwriting?
Handwriting varies so much that writing an exact rule for every variation would be impractical
Fixed rules are always simpler and more effective for this task
Handwriting recognition requires no pattern recognition at all
Machine learning is never used for tasks like this
9. The quality of a machine learning model's predictions is heavily influenced by:
The quality and amount of training data it learned from
Only the colour of the computer used
Nothing at all; predictions are always random
Only the model's file size
10. A voice assistant recognising spoken commands is an example of AI applied to:
Understanding language
Only physical movement
Only mathematical calculation
Nothing related to AI
11. If a spam filter is trained mostly on old spam examples, it might struggle with:
New, different types of spam it hasn't seen before
Old spam emails, which it would always miss
Nothing; it will always be 100% accurate forever
Non-spam emails only
12. AI recommendation systems (like those suggesting videos or products) work by:
Identifying patterns in your past behaviour and similar users' behaviour
Randomly selecting content with no pattern
Ignoring all user data completely
Using only a single fixed rule for every user
13. Why is having a large, diverse training data set generally beneficial for a machine learning model?
It helps the model learn to handle a wider range of realistic situations accurately
A small, narrow data set always produces better results
Diversity in training data always confuses a model
Training data size has no effect on model performance
14. Why might an AI model trained primarily on data from one group of people perform less accurately for a different group?
If the training data isn't representative of all groups, the model may not generalise well to underrepresented ones
Models always perform identically regardless of what data they were trained on
Training data representation has no connection to model accuracy
This scenario is impossible with machine learning
15. Why is "bias in training data" considered an important ethical consideration in AI development?
A model can learn and reproduce unfair patterns present in its training data, leading to unfair outcomes
Training data can never contain any bias
Bias in data has no effect on a model's behaviour
This is a purely technical issue with no ethical dimension
16. Why might it be misleading to think of AI as always being completely "objective" or "neutral"?
AI systems reflect patterns from the data and choices made by the humans who built them
AI is always completely free of any human influence or bias
Objectivity is guaranteed automatically in every AI system
AI systems never reflect any patterns from their training data
17. Why do AI developers often test models on data separate from what was used to train them?
It checks whether the model generalises well to new, unseen situations rather than just memorising training examples
Testing on separate data is never a meaningful practice
Models should only ever be tested on their exact training data
This kind of testing has no purpose in AI development
18. Why might explaining how an AI model reached a decision (explainability) be especially important in fields like medicine or law?
High-stakes decisions benefit from being understandable and justifiable, not just accurate
Explainability is never relevant in any AI application
Accuracy is the only thing that ever matters in high-stakes AI decisions
AI decisions in medicine and law require no scrutiny at all
19. Why is ongoing human oversight often recommended for AI systems making significant real-world decisions?
AI systems can make mistakes or reflect flawed data, so human review helps catch and correct errors
AI systems are always perfectly accurate and need no oversight
Human oversight has no meaningful role in AI systems
AI decisions should never be reviewed once they are made
20. Why might an AI chatbot sometimes generate a confident-sounding but factually incorrect answer (a "hallucination")?
Models predict plausible-sounding patterns from training data, which does not guarantee factual accuracy
AI models are always 100% factually accurate with no exceptions
Chatbots never generate any incorrect information
Hallucination in AI has no connection to how these models work
21. Why is distinguishing between narrow AI (designed for one specific task) and general AI (human-like broad intelligence) an important distinction?
Most AI in use today is narrow, excelling at specific tasks but lacking broad human-like understanding
All current AI systems already have complete general intelligence
This distinction has no relevance to understanding modern AI
Narrow and general AI are exactly the same thing
Answer key (parent copy)
1. Systems performing tasks typically requiring human intelligence
2. Training a model on data to find patterns
3. The data used to teach a model to recognise patterns
4. Examples of spam and non-spam emails
5. Recognising images or understanding language
6. Patterns in data
7. Learns patterns from data rather than following pre-written rules for every case
8. Handwriting varies so much that writing an exact rule for every variation would be impractical
9. The quality and amount of training data it learned from
10. Understanding language
11. New, different types of spam it hasn't seen before
12. Identifying patterns in your past behaviour and similar users' behaviour
13. It helps the model learn to handle a wider range of realistic situations accurately
14. If the training data isn't representative of all groups, the model may not generalise well to underrepresented ones
15. A model can learn and reproduce unfair patterns present in its training data, leading to unfair outcomes
16. AI systems reflect patterns from the data and choices made by the humans who built them
17. It checks whether the model generalises well to new, unseen situations rather than just memorising training examples
18. High-stakes decisions benefit from being understandable and justifiable, not just accurate
19. AI systems can make mistakes or reflect flawed data, so human review helps catch and correct errors
20. Models predict plausible-sounding patterns from training data, which does not guarantee factual accuracy
21. Most AI in use today is narrow, excelling at specific tasks but lacking broad human-like understanding