Software Engineering › Key terms › Y12 Software Automation › Bias (human and dataset)
Bias (human and dataset)
Term Key terms bank Y12 Software Automation
Also listed as: Human bias; Dataset source bias; Bias (human, dataset)
Definition
Human bias Y12 Software Automation Term
Developers' own assumptions or prejudices influence design, data selection or labelling and so the outcomes.
Example: Labellers' opinions shape the labels.
Found in: 2025 HSC marking guidelines, sample exam and familiarisation guidelines, school notes.
Dataset source bias Y12 Software Automation Term
Unfair results because the training data is unrepresentative or reflects past prejudice.
Example: A face detector trained mostly on one group fails on others.
Found in: syllabus, school notes.
Unfair skew in data, labels or decisions that favours some outcomes or groups.
Where it is used in the test
Syllabus area: Y12 Software Automation.
Question types it can appear in
- Short answer: Human bias: Define, outline or explain it and tie it to the scenario. Dataset source bias: Explain bias in AI (2025 Q27, sample Q23a)
- Extended response: Use it as a supporting point linked to the stimulus.
- Multiple choice: Pick the option that matches the definition or example.
Where it has appeared in the paper
Software Automation (Part A, B). Part letters follow each paper’s own layout.
Related past questions
- 2025 HSC Q27 (6 marks): Bias in AI (human bias and data set source bias) Software Automation (Part A) · 6 marks
- Sample exam Q23a (3 marks): Bias in machine learning Software Automation (Part B) · 3 marks
These are the questions the flashcard set links to this term. Open each one and compare your answer with its marking criteria.
Study links
- Flashcard: Human bias in Software Automation (Year 12) · study that set
- Flashcard: Dataset source bias in Software Automation (Year 12) · study that set
- Memorise: Impact of automation on individuals, society and the environment
- Key terms bank entry: Bias (human, dataset)
- Master answer: hsc2025-q27 (6 marks, extended response (explain, video stimulus)) (Master model answers: Software Automation)
- Notes: Investigate the effect of human and dataset source bias in the development of ML and AI… (Software Automation (Year 12))
- Notes: Human bias and dataset source bias (Software Automation (Year 12))
- Master answer: sample-q23a (3 marks, short answer (max 75 words)) (Master model answers: Software Automation)
How to use it in an answer
A model sentence
“Developers' own assumptions or prejudices influence design, data selection or labelling and so the outcomes.” Then apply it: Labellers' opinions shape the labels.
In a short answer, give the definition in your own words first, then tie it to the situation in the question. Do not copy the example above unless it fits the scenario.
Command words used with it
- Explain: Relate cause and effect; make the relationship between things evident; provide why and/or how. (2025 HSC Q27 (6 marks): Bias in AI (human bias and data set source bias))
Technique for the question types
- Short answer: One clear point per mark, use the stimulus and course terms, and answer the command word.
- Extended response: Plan with the command word, give a judgement or explanation supported by named details from the stimulus, use course terms and cover both sides when asked.
- Multiple choice: Read all four options, rule out wrong ones, pick one; no penalty for guessing so never leave blank.
What the marking guidelines and feedback say
Marker feedback (better responses): use specific examples from the video AND the broader effect beyond it; identify different categories of bias and how they affected the training data, the AI, its outputs and fitness for purpose; contrast failure types; discuss bias in a wider context (AI as a trusted source of information); do not just restate the video/transcript; describe the failure rather than only naming 'the shoe game' or 'the physicist example'; show the importance of checks to prevent learned bias being amplified.
Worked answers from the marking guidelines
2025 HSC Q27 (6 marks): Bias in AI (human bias and data set source bias) (6 marks): criteria and sample answer
Criteria
6: Explains how human bias could affect the usefulness of AI solutions AND supports with reference to the video. 5: Describes how, and refers to a scenario(s) in the video. 4: Outlines how human bias could affect AI solutions and refers to the video. 2-3: Identifies some sources/examples of human bias AND/OR effects on AI AND/OR outlines how AI learns. 1: Some relevant information.
Sample answer
AI develops its ability by analysing vast amounts of data. The quality of these data can be influenced by human bias, and in turn, affect the usefulness of AI solutions.
Human bias may be explicit or implicit. Explicit biases are intentional. For example, a researcher may favour male physicists and only collect pictures of male physicists for training an AI solution, as shown in the video. Implicit biases operate unconsciously and may be shaped by social and cultural factors. Whether the bias was explicit or implicit, the choice of male physicists would suggest to the AI solution that physicists are male.
Human bias can occur in various stages of the development of AI solutions. As shown in the photo library example, training data needs to be diverse and representative, otherwise AI solutions may produce inaccurate or unhelpful responses. The process of labelling training data may also introduce biases as people may have different interpretations of the same data. The video shows that if most people draw shoes in particular ways, certain types of shoes will not be recognised as a shoe.
Sample exam Q23a (3 marks): Bias in machine learning (3 marks): criteria and sample answer
Criteria
3: Shows thorough understanding of how the effects of human bias on training can be minimised. 2: Outlines some effects of human bias on ML OR features of training. 1: Some relevant information.
Sample answer
Training data is initially selected by humans, who may unintentionally introduce their own biases. To minimise this, ensure that training data is representative of the problem the ML model will address. Reviewing data by individuals from diverse backgrounds could identify and reduce potential biases, ensuring fairer and accurate model performance.
Common mistakes
- Use the term exactly as the syllabus words it, then link it to the scenario in the question; the 2025 feedback repeatedly rewards using the stimulus.
- Answer the command word: outline gives main points only, explain gives a reason or cause, discuss gives both sides.
Related terms
Same group in the syllabus: Impact of automation · Safety of workers · People with disability · Nature and skills of employment · Production efficiency, waste and the environment · Economy and distribution of wealth · Patterns in human behaviour