Community resourceWorksheet

OCR H446 1.5.2 Automated decisions, AI and bias

Part 2 of 9 · H446 1.5.2 · Moral and ethical issues

Bias in an automated decision needs a cause, an outcome and a response. Written for OCR H446 1.5.2, this worksheet follows data into a ranking decision and back through a feedback loop, using apprenticeship shortlisting, hospital prioritisation and AI-set credit limits as its contexts.

Students will:

  • locate the points at which bias enters an automated decision system
  • explain how an unequal outcome feeds back into future training data
  • describe a test that would show whether a ranking model produces unequal outcomes
  • propose mitigations that address the cause of a bias rather than its symptom
  • discuss the implications of an AI credit decision for customers, the lender and communities

Inside: 5 explanation cells, 1 multiple-choice question, 2 fill-in-the-blanks cells and 2 written answers. 26 marks, about 30 to 50 minutes.

Series: H446 1.5.2 · Moral and ethical issues, part 2 of 9.

Shared by Coding PathwayVerified teacher

  • 10 cells
  • About 45 minutes
  • CC BY-SA 4.0
  • Shared 31 Aug 2026
  • Updated 3 Sept 2026

Preview

The whole resource, exactly as a class sees it. Answers and marking are held back.

Automated decisions, AI and bias

This worksheet explores how bias can enter an automated or AI-assisted decision and affect people. When an examination question asks about AI bias, explain where the problem starts, how it changes a decision and how that decision affects a person.

Trace the whole decision

AI decision and feedback looptrace the outcome, not just the algorithmDATApast examplesMODELlearned patternDECISIONscore / rankingOUTCOMEreal effectcheck each link• missing or unrepresentative data?• unsuitable goal or cut-off score?• unequal harm, no explanation or appeal?bad outcomes can become new training data and reinforce the pattern

An automated system can follow fixed rules and does not have to use AI. Artificial intelligence (AI) is the field of designing computer systems to carry out tasks associated with human intelligence, such as recognising patterns, making predictions or supporting decisions. Machine learning is one approach within AI: the system learns patterns from example data instead of being given every decision rule. A robot is a physical machine; it may use AI, but it can also follow fixed instructions without AI. None of these systems necessarily understands the situation or will always be fair and correct.

Bias can enter a system in several ways: past data may not represent everyone, labels may be inaccurate, or the goal may be unsuitable. A proxy is one item of data used as a substitute for something the system cannot measure directly. A threshold is the cut-off score used to place cases into different groups or trigger a decision. The choice of threshold, or the way a system is used, can create unequal outcomes.

Worked example: apprenticeship ranking

Most successful applicants in the old data came from schools that offered one particular course. The model learns to treat that course as a sign of likely success. Equally capable applicants from other schools may then receive lower scores and fewer interviews. If future training data only contains people who were selected, the pattern may continue.

Useful checks include:

  • testing the model on representative data;
  • checking whether the course is an unfair substitute for school background;
  • comparing outcomes between groups; and
  • explaining decisions and allowing human review.
Multiple choice1 mark

Which action best checks whether a ranking model produces unequal outcomes?

  • ARemove every item of personal data before checking whether the model still makes useful decisions
  • BCompare error and selection rates for relevant groups using representative test data
  • CUse a larger historic data set without checking who is represented in it
  • DLet the model make every decision so that human judgement cannot affect the result
Fill in the blanks4 marks
An automated system may follow fixed gap 1. AI systems carry out tasks associated with human gap 2. Machine learning finds patterns in example gap 3. A robot is a physical gap 4 that may or may not use AI.
  • rules
  • intelligence
  • data
  • machine
  • appeal
Written answer8 marks

A hospital uses a model trained mainly on adult cases to decide which patients should receive an urgent clinic appointment. Explain two risks. For each risk, explain one action that could reduce it.

State how the training data could change the decision and affect a patient. Make each action address the problem you identified.

Students type their answer here.

Apply the ideas independently

The next task uses a new decision context and gives you less support. Explain each stage from the data to the effect on a person.

Written answer9 marks

A lender uses an AI model to set credit limits from customers’ purchase and location histories. Discuss the moral, social and ethical issues raised by this use of AI.

Use details from the scenario. Consider possible benefits as well as privacy, unreliable assumptions, unequal outcomes, explanations, appeals and responsibility. Finish with a supported conclusion.

Students type their answer here.

Fill in the blanks4 marks
Bias can begin with unrepresentative checkpoint gap 1 or an unreliable proxy. It may change a system's checkpoint gap 2 and create unequal outcomes. People need a route to request an explanation or checkpoint gap 3, while the organisation remains checkpoint gap 4.

Review your responses

Check that each answer follows the data into a decision and a real effect. Make every safeguard address the problem you identified.