Artificial Intelligence and the Impact of Computing: Unit Exam A
Machine learning and training data, spotting AI, bias and its causes, responsibility, and the effects of AI on jobs, privacy and the environment.
Instructions
- Time allowed
- 25 minutes
- Total marks
- 25
- Calculator
- No calculator is permitted.
- Answer all the questions.
- The number of marks for each question is shown in brackets. Use it as a guide to how much to write.
- The clock at the top counts down from the real time allowed. When it reaches zero the paper is locked and marked, as it would be collected in the exam hall.
- Read every question to the end before answering. Where a question says tick two boxes, tick exactly two: ticking more than asked scores nothing. Type one word and number answers carefully, because they are marked exactly.
Questions
Answer every question, then finish the paper to have it marked. Marking as you go teaches you far less.
Tick one box.
- LC A type of AI where a computer learns patterns from data instead of being given every rule
Machine learning finds the rules from examples.
Tick one box.
- LA The examples an AI system learns from
A model learns from its training data.
| Uses AI | Does not need AI | |
|---|---|---|
| A streaming service recommending films you might like | ||
| A calculator adding two numbers | ||
| A phone unlocking by recognising your face | ||
| A timer counting down from 60 |
Tick one box in each row.
- LA streaming service recommending films you might like: Uses AI
- LA calculator adding two numbers: Does not need AI
- LA phone unlocking by recognising your face: Uses AI
- LA timer counting down from 60: Does not need AI
Recommendations and face recognition learn from data. Adding numbers and counting down follow fixed rules.
- LIts training data is unrepresentative or contains past human bias
- LIt learns and repeats those patterns, treating some groups unfairly
An AI learns only from its training data. If that data leaves some groups out, or records unfair decisions people made in the past, the AI learns those patterns and repeats them, treating some groups unfairly without anyone intending it to.
Tick one box.
- LD Its training data had far fewer pictures of some groups
A system is only as good as the variety in its training data.
Tick one box.
- LB An AI can spot patterns without knowing why they happen, so it can make confident mistakes
Spotting that two things happen together is not knowing why.
- LIt could be the manufacturer or the programmers who built the system
- LIt could be the owner or the person meant to be supervising it
- LNo person made the decision, and the AI's reasoning is hard to explain, so the law is unclear
Blame could fall on the manufacturer, whose system made the decision, the programmers or the people who supplied its training data, or the owner who was supposed to be supervising. No human actually chose what happened, the AI cannot explain its reasoning, and the law has not caught up, so there is no clear answer.
| Employment | Privacy | Environment | |
|---|---|---|---|
| Self service checkouts replace some shop workers | |||
| Smart speakers record conversations in people's homes | |||
| Training a large AI model uses huge amounts of electricity |
Tick one box in each row.
- LSelf service checkouts replace some shop workers: Employment
- LSmart speakers record conversations in people's homes: Privacy
- LTraining a large AI model uses huge amounts of electricity: Environment
Jobs, personal information and energy use: employment, privacy and environment.
- LIt is fast and cheap, and can handle thousands of applications
- LIt applies the same rules to everyone
- LIt may learn bias from past hiring and treat some groups unfairly, and its decisions are hard to explain
- LReaches a judgement, such as using it only with a human checking its decisions
AI can sift thousands of applications in minutes, far more cheaply than people, and it applies the same rules to every applicant without getting tired. But if it learns from a company's past hiring it can copy past unfairness, for example against women or older applicants, and it may not be able to explain why someone was rejected. It is reasonable as a first sift only if a person checks its decisions and the system is regularly tested for bias.
Tick two boxes.
- LB Who benefits, and who bears the cost?
- LD Were the people affected consulted?
- LNo marks if more than 2 boxes are ticked
Asking who gains, who loses and who was asked gets to the heart of most technology decisions.
Tick one box.
- LC A spam filter sorting your email
Spam filters learn from millions of emails.
- Laccountability
Accountability.
Tick one box.
- LA More varied examples help them learn patterns that work on new cases
A model has to generalise from the examples it has seen.
This is an original Key Stage 3 assessment. Most questions are closed, so the marking is exact, and the written answers are marked against a mark scheme in the same style as a GCSE paper, so the format is familiar long before it counts.