OCR A-Level Computer Science Computer systems (01), June 2025: Question 7
13 marks · Hard difficulty · Extended Response
Discuss the legal and moral implications of training an AI image generator on scraped web photos, and decompress a run-length encoded image onto a grid.
Practise this questionQuestion
Question text
(a) OCR Photo is a free website designed to use Artificial Intelligence (AI) to generate unique
photographs from a text description given by the user.
The AI learns by searching the internet randomly for millions of photographs and then matching
these to a textual description of the photograph. The photographs include people, property,
landscapes and objects.
Discuss the legal and moral implications of this approach and what OCR Photo should consider.
You should include the following in your answer:
• what is meant by AI
• the legal and moral implications of the learning approach described in the question
• a conclusion justifying if OCR Photo should use this approach or not. [9]
(b) OCR Photo uses run length encoding to compress the generated photographs.
A black and white photograph is encoded using B for the colour black and W for the colour white.
The encoded sequence is:
W4B1W4
W3B3W3
W2B5W2
W1B7W1
Use the grid to show the result of using run length encoding to decompress this sequence.
[4]
Mark scheme
Show the mark scheme
7 (a) Mark Band 3 – High level (7-9 marks) 9 Answers may include, but are not
limited to, some of the points below:
The candidate demonstrates a thorough knowledge and understanding of the moral
and legal implications of AI. The material is generally accurate and detailed. AO1 Knowledge
AI behaves and perf orms tasks that
The candidate is able to apply their knowledge and understanding directly and normally require human intelligence.
AI needs to learn by understanding
consistently to the context provided. Evidence/examples will be explicitly relevant to languages, recognising patterns,
the explanation. solving problems and learning f rom
experience. They can be trained to
The candidate provides a thorough discussion which is well balanced. Evaluative complete specific tasks or can be used
comments are consistently relevant and well-considered. to complete a wide range of dif f erent
tasks.
There is a well-developed line of reasoning which is clear and logically structured.
The information presented is relevant and substantiated. AO2 Application
Moral
Mark Band 2 – Mid level (4-6 marks) • Individuals may not know that their
The candidate demonstrates reasonable knowledge and understanding of the moral images are being used to train the
and legal implications of AI. The material is generally accurate but at times AI.
underdeveloped. • Individuals are entitled to privacy
and they may not want their
photograph to be used by the AI.
The candidate is able to apply their knowledge and understanding directly to the They may think that this is an
context provided although one or two opportunities are missed. Evidence/examples invasion of privacy.
are for the most part implicitly relevant to the explanation. • The AI is searching the internet
randomly and therefore, it needs to
ensure that it uses a complete
range of diverse photographs such
The candidate provides a sound discussion, the majority of which is focused. as people f rom dif f erent cultures
Evaluative comments are for the most part appropriate, although one or two and diversity of scenes to
opportunities for development are missed. represent reality and ensure
f airness.
There is a line of reasoning presented with some structure. The information
presented is in the most part relevant and supported by some evidence. Legal
• It needs to ensure that the
Mark Band 1 – Low Level (1-3 marks) photographs are accessed legally
without criminal activity to ensure
this does not breach the
The candidate demonstrates a basic knowledge of some aspects of the moral or Computer Misuse Act.
legal implications of AI. The material is basic and contains some inaccuracies. • Organisations must ensure that
they keep personal data saf e
The candidate makes a limited attempt to apply acquired knowledge and under the Data Protection Act.
understanding to the context provided. This includes photos and therefore
they need to ensure these cannot
The candidate provides a limited discussion which is narrow in focus. Judgments if be accessed by the AI.
made are weak and unsubstantiated. The information is basic and communicated in • Some images will be protected by
an unstructured way. the Copyright, Design and
Patents Act which includes
The information is supported by limited evidence and the relationship to the evidence people’s intellectual property.
may not be clear. These theref ore cannot be
accessed without permission f rom
0 marks the author.
No attempt to answer the question or response is not worthy of credit.
A03 Evaluation
• AI provides many benef its,
however, f or it to be ef f ective it
needs to learn and be trained.
• Learning f rom paired photos and
descriptions is a good approach to
use to allow the AI to learn,
however they should consider:
o ensuring that all relevant
computer laws and other non-
computer laws are f ollowed.
o only using royalty free images
to avoid copyright
inf ringements.
o ensuring individuals have
given their consent before their
photographs are used.
o ensuring that a wide range of
photographs are used to
ensure the AI is not biassed.
7 (b) 1 mark for each correct row to max 4. 4 Allow shading, or letters or words
etc to show understanding of each
run.
W W W W B W W W W
W W W B B B W W W
W W B B B B B W W
W B B B B B B B W
How to answer it
OCR A-Level CS: AI Training Ethics & Run Length Encoding
What This Question Tests
This two-part question assesses Section 1.5 (Legal, Moral, Cultural and Ethical Issues) and Section 1.1.3 (Data Representation & Compression):
- Part (a) [9 Marks]: An extended-response essay evaluating the definition of AI, legislation covering web-scraping (CDPA, DPA/GDPR, CMA), moral dilemmas (consent, privacy, bias/representation), and formulating a justified conclusion.
- Part (b) [4 Marks]: Lossless compression technique using Run Length Encoding (RLE) to decompress run-count pairs into a 2D pixel grid.
Legal & Moral Implications of AI Web-Scraping
Extended response discussing ethics, UK legislation, and deployment justification
💡 Key Knowledge (AO1 & AO2)
What is Artificial Intelligence?
- Software/systems engineered to perform tasks that typically require human intelligence.
- Learns by identifying patterns in massive datasets, using trial/experience, and generalising to generate novel outputs.
Relevant UK Legislation:
- CDPA 1988 Copyright, Designs and Patents Act: Scraping artists' and photographers' creative works without licenses infringes intellectual property.
- DPA 2018 / GDPR Data Protection Act: Photos containing identifiable individuals constitute personal data; scraping without consent breaches data privacy rights.
- CMA 1990 Computer Misuse Act: Bypassing site security measures, paywalls, or bot mitigation to scrape data constitutes unauthorised access.
🧠 Exam Technique (Level of Response)
To reach Mark Band 3 (7–9 marks), you must structure your answer into four distinct pillars:
- Define AI clearly: Mention pattern recognition, machine learning from paired inputs/outputs, and decision making.
- Legal Specificity: Name exact laws ( CDPA , DPA/GDPR , CMA ) and explain precisely how scraping breaks them.
- Moral Debate: Cover two angles:
- Individual level: Lack of consent, privacy violations.
- Societal level: Data bias/lack of cultural diversity resulting in distorted outputs.
- Justified Conclusion: Do not just repeat facts; offer solutions (e.g., using curated public-domain/royalty-free datasets, offering opt-outs).
✅ Model Answer Plan (Top-Band Response)
1. Definition of AI: Artificial Intelligence refers to computer systems capable of performing tasks that historically require human cognitive abilities. In this scenario, machine learning models process millions of text-image pairs, extracting statistical features to generate novel visual imagery based on natural language prompts.
2. Legal Implications:
• Copyright, Designs and Patents Act (1988): Millions of online photos are protected by copyright. Scraping and duplicating these images without the explicit licence or permission of the rights holder is copyright infringement.
• Data Protection Act (2018) / UK GDPR: Photographs containing identifiable people count as personal data. Storing and processing this data without explicit consent or a lawful basis infringes data subject rights.
• Computer Misuse Act (1990): Automated web-crawlers that circumvent a server’s security measures, anti-bot mechanisms, or terms of service can be classed as unauthorised access.
3. Moral Implications:
• Consent & Privacy: People uploading photos to social media or personal blogs have a reasonable expectation of privacy; having their likeness used to train commercial/generative tools without knowledge feels exploitative.
• Algorithmic Bias & Representation: Random internet scraping frequently ingests unrepresentative, stereotypical, or offensive imagery. This leads to biased outputs, cultural stereotyping, or poor diversity representation across demographics.
4. Conclusion & Recommendation: OCR Photo should not use the unrestrained, random-scraping approach described. While AI requires vast training datasets to generalise accurately, indiscriminate harvesting exposes the company to severe legal liability and ethical backlash. Instead, OCR Photo should train its AI exclusively on royalty-free/Creative Commons datasets, licensed stock libraries, and public-domain imagery, combined with strict content-filtering algorithms and an opt-out mechanism for artists.
❌ Common Pitfalls to Avoid
- Vague legal naming: Stating "it breaks the copyright law" or "data privacy law" without citing the formal act names ( Copyright, Designs and Patents Act , Data Protection Act ).
- One-sided argument: Focusing entirely on copyright and ignoring the moral issues (consent, societal bias, stereotyping).
- Missing Conclusion: Forgetting the final bullet point of the question. You cannot score Band 3 without an explicit conclusion justifying whether they should proceed.
Decompressing Run Length Encoding (RLE)
Reconstructing an uncompressed binary pixel grid from run-count data
📐 Step-by-Step Decompression
The image is encoded using letter-number pairs ( W = White, B = Black):
- Row 1: W4 B1 W4
Decompresses to: 4 White, 1 Black, 4 White → Total = 9 cells. - Row 2: W3 B3 W3
Decompresses to: 3 White, 3 Black, 3 White → Total = 9 cells. - Row 3: W2 B5 W2
Decompresses to: 2 White, 5 Black, 2 White → Total = 9 cells. - Row 4: W1 B7 W1
Decompresses to: 1 White, 7 Black, 1 White → Total = 9 cells.
✅ Completed Grid (1 Mark Per Row)
Either letters (W/B) or shaded boxes are fully credited in the exam:
| W | W | W | W | B | W | W | W | W |
| W | W | W | B | B | B | W | W | W |
| W | W | B | B | B | B | B | W | W |
| W | B | B | B | B | B | B | B | W |
🧠 Verification Tip
Always count the total number of cells in each decompressed row before moving on. In this question, the grid width is exactly 9 cells wide. Checking that each row's values sum to 9 (e.g. 4 + 1 + 4 = 9 and 2 + 5 + 2 = 9) guarantees you haven't shifted an encoded block accidentally.
Topics
1.3 Exchanging data · 1.5 Legal, moral, cultural and ethical issues · 1.3.1 Compression, Encryption and Hashing · 1.5.1 Computing related legislation · 1.5.2 Moral and ethical Issues
Question and mark scheme from the OCR A-Level Computer Science examination, Computer systems (01), June 2025. QuestionVault is an independent revision resource; questions remain the copyright of the awarding body.