Impact of AI exam questions

4 Edexcel GCSE Computer Science (1CP2) exam-style questions on Impact of AI (5.2 Ethical and legal), 8 marks in all. Answer them on paper or print them, then open the mark schemes and example answers to see how each is marked.

1Describe2 marks

A hospital uses an AI system to recommend treatments to doctors.

Describe one accountability issue that can arise from using this AI system.

[2 marks]

Mark scheme

Award up to two marks for a linked description such as:

It may be unclear who is responsible (1) if the AI gives a wrong recommendation that harms a patient (1)
The doctor may be blamed (1) even though they followed the AI's recommendation (1)
The developer/programmer may be held responsible (1) because they designed the AI's algorithm (1)
The hospital may be held responsible (1) because it chose to use the AI system (1)

Do not accept responses about bias or safety unless they are linked to who is responsible.

Do not award a mark for naming a group/person without saying why responsibility is unclear or who may be accountable.

Example answers and how they were marked

Full marks2/2

It may not be clear whether the doctor, the hospital or the programmer is responsible (1) if the AI recommends the wrong treatment and the patient is harmed (1).

Why: Gives the accountability point that responsibility is unclear (1) and develops it with the harm caused by a wrong recommendation (1).

Some marks1/2

It might be unclear who is to blame if the AI makes a mistake.

Why: Identifies that responsibility may be unclear (1) but does not develop how or when this becomes an issue.

No marks0/2

The AI could be biased against some patients.

Why: Describes bias rather than accountability; no creditworthy point under this mark scheme (0).

2Describe2 marks

A warehouse uses robots to move shelves of products to the workers who pick the orders.

Describe one safety issue that can arise from using robots to move the shelves in this warehouse.

[2 marks]

Mark scheme

Award up to two marks for a linked description such as:

A robot's sensors may fail to detect a worker (1) so the robot collides with them and causes injury (1)
A robot may move unexpectedly/at speed (1) and injure a worker standing nearby (1)
A robot may trap/crush a worker against a shelf (1) causing serious injury (1)
A robot may drop a heavy load (1) which falls on a worker and injures them (1)

Allow one mark for a safety issue identified on its own with no development, e.g. a robot could hit a worker (1).

Do not accept descriptions of a robot malfunctioning/breaking down unless the response says how a worker could be harmed.

Do not accept generic statements that robots are dangerous without saying how a worker could be harmed.

Do not accept responses about job loss or other ethical issues; the question asks about safety.

Example answers and how they were marked

Full marks2/2

A robot's sensors might fail to detect a worker (1) so the robot collides with them and injures them (1).

Why: Identifies the safety issue of a collision (1) and develops it by saying how the worker is harmed (1).

Some marks1/2

A robot could hit a worker.

Why: A safety issue identified on its own, which the mark scheme allows for one mark, but there is no development of how the worker is harmed.

No marks0/2

Robots are dangerous.

Why: Generic statement with no description of how a worker could be harmed; no creditworthy point (0).

3Describe2 marks

Machine learning systems are trained using data sets.

Describe one cause of algorithmic bias in a machine learning system.

[2 marks]

Mark scheme

Award up to two marks for a linked description such as:

Training data (sets) is unrepresentative/incomplete (1) so the system learns patterns that are unfair/inaccurate for some groups (1)
Training data (sets) contains human bias/prejudice (1) which the system learns and repeats (1)
Features/weightings are chosen that favour one group (1) so the decisions made are unfair to other groups (1)
Training data (sets) is out of date (1) so the system's decisions do not reflect current situations (1)

Do not accept answers that only state that the data/system is biased (e.g. 'the data is biased', 'the AI is biased') without saying how the data is unrepresentative, incomplete or contains human bias.

Do not accept the size of the data set alone as a cause (e.g. 'it learns from a large data set') unless it is linked to the data not representing all groups.

Do not accept outcomes of bias (e.g. unfair decisions) as the cause unless they are linked to what caused the bias.

Do not accept responses about data privacy, cyber security or errors in general that are not linked to bias.

Example answers and how they were marked

Full marks2/2

The training data may not include enough photos of older people (1) so the system is less accurate for them and treats them unfairly (1).

Why: Identifies unrepresentative training data as a cause (1) and develops it into the unfair/inaccurate patterns the system learns (1).

Some marks1/2

The training data may only include photos of one age group.

Why: Identifies a specific cause, training data that is unrepresentative/incomplete (1), but does not develop what this leads to.

No marks0/2

The training data is biased.

Why: Only states that the data is biased without saying how it is unrepresentative, incomplete or contains human bias; the mark scheme's Do not accept line rules this out, so no marks (0).

4Describe2 marks

A delivery robot travels on pavements to deliver parcels.

Describe one legal liability issue that can arise when this robot injures a pedestrian.

[2 marks]

Mark scheme

Award up to two marks for a linked description such as:

It may not be clear who is legally liable (1) because the robot acted autonomously without a human controlling it (1)
The manufacturer may be legally liable (1) because the robot's design/software caused the injury (1)
The owner/operator may be legally liable (1) because they deployed the robot on the pavement (1)
The programmer may be legally liable (1) because the robot followed their code (1)

Do not accept responses about safety or job loss without reference to who is legally responsible.

Do not accept answers that name a law without describing who may be legally liable.

Example answers and how they were marked

Full marks2/2

It may not be clear whether the manufacturer or the owner is legally liable (1) because the robot acted without a human directly controlling it (1).

Why: Identifies that legal responsibility is unclear (1) and develops it with the robot acting autonomously (1).

Some marks1/2

It is not clear who is legally responsible.

Why: Identifies the legal liability issue (1) but gives no linked development of why responsibility is unclear.

No marks0/2

The robot could hurt someone.

Why: Describes harm rather than legal liability; no creditworthy point (0).

Revise Impact of AI first

Keep practising Impact of AI

More exam questions on 5.2 Ethical and legal