AI & Accountability

When a machine decides

AI, machine learning and robotics increasingly make decisions that people used to make: approving a loan, sorting job applications, even driving a car. When one of those decisions goes wrong, who is responsible? This module digs into the four ideas the exam cares about: accountability, safety, algorithmic bias and legal liability.

Four words that matter

These four ideas run through every AI ethics question:

Where does the bias come from?

A recruitment AI was trained on a company's past hiring, which had mostly hired men. It now rejects strong female applicants. What is the root cause?

• The training data was biased, so the model learned that bias
• The computer dislikes certain applicants on purpose
• The applicants filled in the form incorrectly
• AI is always random, so its results mean nothing

Match each issue to an example

Who could be to blame?

When an autonomous system causes harm, the law has to decide who is liable. It is rarely just one party, and the answer depends on the exact scenario.

Which raise a bias issue?

Select the TWO situations that mainly raise an algorithmic BIAS issue.

• A facial recognition system misidentifies people with darker skin far more often
• A CV-screening AI scores women lower than men with the same experience
• A surgical robot arm moves unexpectedly and could injure a patient
• A firm copies a rival's logo onto its own product

Be specific, or lose marks

In the exam, never just write "AI can be unfair". For the marks, name a specific issue tied to the scenario, say who is affected or accountable, and suggest a safeguard. For example: "A self-driving car brakes late for pedestrians it was under-trained to recognise. That is a safety and bias issue; the developer is accountable; the fix is wider testing data plus human oversight."

Bias, data and accountability

Machine learning learns its patterns from _____, so unrepresentative examples lead to algorithmic _____. Because the system then decides on its own, the law must decide who holds legal _____ for any harm. Good practice adds human _____ so a person can step in, and keeps the system _____ enough for its decisions to be explained.

Build it responsibly

Put the steps of developing an AI system responsibly into a sensible order.

• Choose training data that fairly represents everyone the system will affect
• Test the trained system for bias and unsafe behaviour before release
• Add human oversight so a person can review or override decisions
• Release it with clear limits on what it should be used for
• Monitor it in use and correct problems that appear

The crash decision

A self-driving car brakes too late for a pedestrian it was poorly trained to recognise, and injures them. Work through who is accountable.

• The software failed to recognise the pedestrian in time. Which issue is this mainly?
• Investigators find the training data barely included pedestrians at night. Who is most accountable?
• What is the best safeguard to stop this happening again?

No excuse

An online shop's AI wrongly cancels thousands of valid orders. A manager says "the algorithm did it, so no one is responsible". Why is that wrong?

• People chose to build, train and deploy the system, so responsibility stays with them
• The algorithm itself should be sent to prison
• Customers agreed to the mistake by shopping there
• AI decisions are never actually wrong

Spot the exam-ready answers

Tap the TWO answers that would earn marks: each names a specific issue, says who is accountable, and links to the scenario.

• AI can sometimes be a bit unfair to people.
• The recruitment AI is biased against women because its data favoured past male hires, so the developer is accountable.
• Technology is bad and should be banned completely.
• The car's late braking is a safety issue; the manufacturer and developer share liability, and wider testing is the fix.
• Computers just do whatever they want to do.

Make the case

A supermarket brings in an AI that predicts which staff to promote, using years of past performance data. Some staff complain it is unfair. Discuss the ethical and legal issues, and who should be accountable.

• Name a specific issue the AI could raise, such as algorithmic bias learned from the past data
• Explain who could be held accountable if it makes an unfair decision
• Suggest a safeguard, such as human oversight or checking the data is representative
• Explain why "the computer decided" is not a valid excuse in law