What ethical and societal issues does artificial intelligence and automation raise, and how should they be addressed?
Discuss the ethical and societal impacts of AI and automation, including bias, accountability, privacy and the effect on work
A focused answer to the H2 Computing outcome on AI ethics. Algorithmic bias, accountability and transparency, privacy and data use, the effect of automation on employment, and responsible approaches.
Reviewed by: AI editorial process; not yet individually human-reviewed
Have a quick question? Jump to the Q&A page
Jump to a section
What this dot point is asking
SEAB wants you to discuss the ethical and societal impacts of AI and automation - including bias, accountability, privacy and the effect on work - and how they might be addressed. The central idea is that powerful automated systems bring real benefits but also real harms, and that because they learn from data and make consequential decisions, they raise questions of fairness, responsibility and the future of employment.
The answer
Algorithmic bias
Algorithmic bias is when a system produces systematically unfair outcomes for certain groups. It most often arises because a model learns from biased data: historical data reflecting past prejudice, or data that under-represents some groups, teaches the model to reproduce that unfairness. Biased feature or label choices, and biased deployment, also contribute.
Mitigations: use representative, balanced training data; audit outcomes across groups to detect disparities; keep human oversight of important decisions; and involve diverse perspectives in development.
Accountability and transparency
When an AI system makes a consequential decision - a loan refusal, a medical recommendation, an autonomous vehicle's action - it can be unclear who is responsible for harm: the developer, the organisation deploying it, the data provider, or the user. Many models are opaque ("black boxes"), making decisions hard to explain or contest. This matters for fairness, legal liability and public trust, which is why transparency and human oversight are stressed.
Privacy and data use
AI relies on large amounts of data, often personal. This raises privacy concerns: how data is collected, consented to, stored and used, the risk of surveillance, and the danger of re-identifying people from supposedly anonymous data. Responsible use means collecting only what is needed, securing it, and respecting consent and data-protection law.
Impact on work
Automation reshapes employment, with both costs and benefits:
- Negative - it can displace jobs, especially routine or repetitive roles, causing unemployment and a need to retrain.
- Positive - it can create new roles (building, maintaining and overseeing AI), raise productivity, and remove dangerous or tedious tasks, freeing people for higher-value work.
The net effect depends on how society manages reskilling and the transition.
Responsible approaches
Across these issues, responsible AI emphasises fairness (unbiased data and tested outcomes), transparency (explainable, auditable decisions), accountability (clear responsibility and human oversight), and privacy (lawful, consented, secured data).
Examples in context
Example 1. Facial recognition accuracy gaps. Some facial-recognition systems have been shown to misidentify certain demographic groups far more often, because their training data under-represented those groups. This real pattern of algorithmic bias shows why representative data and per-group accuracy testing are essential before deployment in sensitive uses.
Example 2. Automation in logistics. Warehouses increasingly automate picking and sorting, removing strenuous repetitive jobs (a benefit) while reducing demand for some manual roles (a cost), and creating new roles maintaining and supervising the robots. The same change is simultaneously positive and negative, which is why managing reskilling and transition is the real policy question.
Try this
Q1. What is algorithmic bias, and what is its most common cause? [2 marks]
- Cue. Systematically unfair outcomes for certain groups, most commonly caused by learning from biased or unrepresentative training data.
Q2. Give one positive and one negative effect of automation on employment. [2 marks]
- Cue. Positive: new roles, higher productivity, safer/less tedious work. Negative: displacement of routine jobs, requiring retraining.
Q3. Why is accountability a concern when AI systems make decisions? [1 mark]
- Cue. It can be unclear who is responsible for harm (developer, deployer, data provider, user), worsened by opaque models that are hard to explain or contest.
Exam-style practice questions
Practice questions written in the style of SEAB exam questions on this dot point, with worked answer explainers. The year tag is the paper they imitate, not the source.
Original6 marks(a) Explain what is meant by algorithmic bias and how it can arise in a machine learning system. (b) Suggest two measures that could reduce bias in such a system.Show worked answer →
(a) Algorithmic bias is when an automated system produces systematically unfair outcomes for certain groups. It commonly arises because the system learns from biased training data: if the historical data reflects past prejudice or is unrepresentative (under-representing some groups), the model learns and reproduces that bias. It can also arise from biased choices of features or labels, or from how the system is deployed.
(b) Two measures (any two):
- Use representative, balanced training data - ensure the data fairly covers all relevant groups, and audit it for skew before training.
- Test and audit outcomes across groups - evaluate the model's accuracy and error rates separately for different groups to detect unfair disparities, and correct or retrain where found.
Other valid measures: human oversight of important decisions, transparency about how decisions are made, and including diverse perspectives in development.
Markers reward bias as systematically unfair outcomes arising from biased/unrepresentative data, and two genuine mitigations such as representative data and testing outcomes across groups.
Original5 marksDiscuss the impact of automation and AI on employment, giving one negative and one positive effect, and explain why accountability is a concern when AI systems make decisions.Show worked answer →
Impact on employment. A negative effect: automation can displace jobs, particularly routine or repetitive roles (in manufacturing, data entry, some clerical work), causing unemployment or the need for workers to retrain. A positive effect: it can create new kinds of jobs (developing, maintaining and overseeing AI systems), increase productivity, and remove dangerous or tedious tasks, freeing people for higher-value work.
Accountability. When an AI system makes a decision (a loan refusal, a medical recommendation, a self-driving action), it can be unclear who is responsible if it causes harm - the developer, the deployer, the data provider, or the user. Complex models can also be hard to explain ("black box"), making it difficult to justify or contest a decision. This matters for fairness, legal liability and trust, which is why human oversight and transparency are emphasised.
Markers reward one valid negative (job displacement) and one positive (new jobs/productivity/safer work) employment effect, and accountability as the difficulty of assigning responsibility for AI decisions, worsened by opacity.
Related dot points
- Explain what machine learning is, how a model learns from training data, and how it differs from rule-based programming
A focused answer to the H2 Computing outcome on machine learning fundamentals. Learning a model from data rather than coding explicit rules, training and inference, features and labels, and the role of data quality.
- Distinguish supervised, unsupervised and reinforcement learning, and identify classification, regression and clustering tasks
A focused answer to the H2 Computing outcome on types of machine learning. Supervised learning with labelled data (classification and regression), unsupervised learning with unlabelled data (clustering), reinforcement learning, and matching a task to a type.
- Describe an artificial neural network in terms of neurons, weights and layers, and explain how training adjusts weights
A focused answer to the H2 Computing outcome on neural networks. Artificial neurons with weighted inputs and an activation function, input, hidden and output layers, how training adjusts weights to reduce error, and what deep learning adds.
- Explain symmetric and public-key (asymmetric) encryption, digital signatures and common network threats and defences
A focused answer to the H2 Computing outcome on network security. Symmetric and public-key encryption, the key-distribution problem, digital signatures for authenticity, hashing, and common threats and defences.
- Describe the stages of the software development lifecycle and compare the waterfall and agile (iterative) approaches
A focused answer to the H2 Computing outcome on the software development lifecycle. The analysis, design, implementation, testing, deployment and maintenance stages, and the contrast between the waterfall and agile iterative models.