Community resourceWorksheet
1CP2-P-10.2 Accountability safety bias and legal liability in AI systems
Part 2 of 5 · 1CP2-P-10 · AI, personal data and ethical/legal issues
AI outputs can influence real decisions, so people and organisations remain responsible for design, deployment and oversight. Algorithmic bias is a systematic unfair tendency caused by data or design; it is not a computer having an opinion and not every isolated wrong result.
Students will:
- explain how data or design can produce systematic bias
- identify accountability and safety responsibilities
- connect an AI output to stakeholder impact
- develop a balanced discussion with practical safeguards
Inside: 6 explanation cells, 1 fill-in-the-blanks cell, 2 multiple-choice questions and 3 written answers. 18 marks, about 45 minutes.
Series: 1CP2-P-10 · AI, personal data and ethical/legal issues, part 2 of 5.
Shared by Coding PathwayVerified teacher
- 12 cells
- About 45 minutes
- CC BY-SA 4.0
- Shared 17 Aug 2026
Preview
The whole resource, exactly as a class sees it. Answers and marking are held back.
Accountability safety bias and legal liability in AI systems
AI outputs can influence real decisions, so people and organisations remain responsible for design, deployment and oversight. Algorithmic bias is a systematic unfair tendency caused by data or design; it is not a computer having an opinion and not every isolated wrong result.
1. Build an accurate model
Unrepresentative historical data, incorrect labels, design choices and feedback loops can produce different error rates or outcomes for groups. Controls include representative data, contextual testing, audit, transparency, human review and a safe fallback. Human review helps only when reviewers have information, authority and time to challenge the output.
- accountability
- bias
- liability
Which best describes algorithmic bias?
- AAny single incorrect output
- BA systematic unfair tendency arising from data or design and affecting decisions
- CA type of computer hardware
- DThe computer deliberately disliking someone
2. Worked application
A hiring model trained mostly on past successful applicants from one group may learn patterns that disadvantage other qualified applicants. The chain is training data → model score → interview decision → unequal opportunity. Testing group outcomes, improving data and meaningful human review can interrupt the chain.
Explain how unrepresentative training data could create unfair hiring decisions.
Build the complete cause-to-impact chain.
Students type their answer here.
Explain two safeguards for an AI system controlling medical priorities.
Link each safeguard to a named risk.
Students type their answer here.
3. Apply to the stated people and system
Name the input or data, the processing or decision, the affected stakeholder and the resulting effect. Avoid claims that could fit any technology.
Discuss whether a school should use an AI model to flag students for extra support.
Consider benefit, false results, bias, privacy, oversight and a justified conclusion.
Students type their answer here.
Which action alone cannot guarantee fairness?
- ARemoving one sensitive field from the input
- BTesting outcomes across relevant groups
- CAuditing data quality
- DProviding a route to challenge decisions
Examination method
For Explain, link cause and consequence. For Discuss, consider more than one aspect, apply each point to the scenario and reach a conclusion supported by the evidence.
Route forward
Next you will examine how organisations collect personal data and how consent, purpose and digital footprints affect people.