For the May and November 2027 IB Computer Science Paper 1 case study, you need enough outside research to explain the relevant technology accurately, compare possible solutions, and support ethical judgments with identifiable evidence. There is no official requirement to read a particular number of sources or produce a bibliography in the examination. As a practical target, most students should build a focused evidence bank from approximately 8 to 12 credible sources, rather than collecting dozens of articles they cannot remember or apply.
The 2027 case study, Generative AI for image creation -- a diffuse vision, concerns Visionary Studios and its evaluation of generative image models. Both SL and HL students need to understand the scenario, diffusion models, model evaluation, and the associated ethical issues. This guide explains how to research those areas efficiently, take citation-ready notes, and reference academic research in an exam answer.
What Paper 1 actually expects
Under the curriculum first assessed in 2027, the case study is included in Paper 1 for both SL and HL. According to the official IB Computer Science subject brief, Paper 1 assesses Theme A, including machine learning, and contains three questions related to the case study.
The assessment arrangements differ by level:
LevelPaper 1 durationWeightingRecommended case-study teaching timeSL1 hour 15 minutes35%5 hoursHL2 hours40%18 hours
These teaching hours are not an official quota for private research. They indicate the relative depth expected, particularly at HL. The IB's Computer Science curriculum update also confirms that the former HL-only Paper 3 has been removed and that the case study now appears in Paper 1 for both levels.
Outside research matters because the booklet is a stimulus, not a complete textbook. It establishes the scenario and terminology, but students must develop the technical understanding and real-world evidence needed to analyse unfamiliar examination material.
How much outside research is enough?
A useful research portfolio normally contains the following:
Source roleSuggested numberWhat it should provideFoundational technical research2-3How diffusion and related models workEthics and governance frameworks2-3Bias, privacy, copyright, transparency, safety, and accountabilityApplied examples2-3Evidence from creative industries or deployed image-generation systemsCritical or comparative analysis1-3Limitations, competing interpretations, or mitigation strategies
The 8-12 source range is advice, not an IB requirement. A student who can accurately use eight carefully chosen sources is better prepared than one who has bookmarked 40 pages without extracting any evidence.
Your research is sufficient when you can do four things without notes:
-
explain diffusion using terms such as noise injection, iterative denoising, neural denoiser, and latent space;
-
compare model choices using quality, computational cost, controllability, reliability, and suitability for Visionary Studios;
-
analyse ethical concerns from more than one stakeholder perspective;
-
identify several studies or frameworks by author or organization, year, finding, and relevance.
SL students should prioritize accurate explanations and clearly applied evidence. HL students need a larger bank of comparisons, limitations, counterarguments, and justified recommendations because they receive more teaching time and face a longer Paper 1. Both levels should avoid treating ethical discussion as an unsupported list of advantages and disadvantages.
Finding credible sources on diffusion models
Begin with original research rather than blogs describing commercial products. Search Google Scholar, a school library database, conference proceedings, or DOI records using phrases such as:
-
denoising diffusion probabilistic models original paper -
latent diffusion models computational cost -
diffusion model image generation limitations -
text-to-image model bias training data study
A strong foundational source is Ho, Jain, and Abbeel's Denoising Diffusion Probabilistic Models, published at NeurIPS in 2020. It formalized a widely used diffusion approach in which a model learns to reverse a gradual noising process.
For efficiency and high-resolution generation, use Rombach and colleagues' High-Resolution Image Synthesis with Latent Diffusion Models. Their 2022 CVPR paper explains how operating in a compressed latent representation can reduce computational demands while preserving useful image detail. This gives you a concrete study for evaluating whether a model is practical for a creative company.
Do not assume that an arXiv upload is automatically peer reviewed. Check whether the work was subsequently published at a recognized conference or in a journal, inspect the methodology, and distinguish the authors' measured results from your interpretation.
Researching generative AI ethics
Ethical research should use frameworks that define risks precisely. The NIST Generative AI Profile, released in 2024, identifies risks including harmful bias, data privacy, environmental effects, information integrity, intellectual property, and abusive synthetic content.
The UNESCO Recommendation on the Ethics of Artificial Intelligence provides a human-rights perspective emphasizing fairness, transparency, privacy, human oversight, and environmental sustainability. The OECD AI Principles, updated in 2024 to address developments including generative AI, add useful concepts such as accountability, robustness, safety, and information integrity.
For every ethical issue, connect the abstract principle to the scenario:
-
Bias: Could the training data cause stereotyped or culturally narrow advertising images?
-
Copyright: Were creators' works used in training, and could outputs imitate protected material?
-
Transparency: Can clients and audiences tell that an image was AI-generated?
-
Environmental impact: Does improved productivity justify the computing resources required?
-
Accountability: Who checks outputs and accepts responsibility for harmful or misleading material?
A credible ethics answer also considers mitigation. Dataset documentation, human review, output testing, provenance records, content labels, licensing controls, and complaint procedures are more useful than simply stating that a model should be “ethical.”
How to evaluate each source
Apply a short credibility test before adding a source to your notes:
-
Authority: Are the authors qualified researchers, standards organizations, regulators, or recognized institutions?
-
Publication: Was the study peer reviewed or issued through an accountable official body?
-
Evidence: Does it explain its data, experiment, evaluation method, and limitations?
-
Currency: Is it recent enough for the claim, or is it an older foundational paper?
-
Purpose: Is the source informing, selling a product, defending a policy, or promoting a model?
-
Corroboration: Can an important claim be confirmed independently?
Jojo AI may help you formulate search terms, simplify a difficult abstract, or quiz you on your notes. It should not be treated as the source of a technical or ethical claim. Generative AI can invent papers, authors, and findings, so open every cited publication and verify that it exists. The IB's statement on AI and assessment emphasizes that AI-generated material is not automatically the student's own work and must be handled in accordance with academic integrity expectations.
Taking citation-ready notes
Do not copy whole abstracts. Convert each source into a compact evidence card that you can revise and reproduce from memory.
FieldExample noteSource identityRombach et al., 2022, latent diffusion paper, CVPRMain contributionPerforms diffusion in a compressed latent space rather than directly across all image pixelsRelevant findingReduced computational requirements while retaining strong image quality and flexible conditioningCase-study linkCould reduce the resources needed for Visionary Studios' high-resolution image generationLimitationLower computational demand does not remove dataset bias, copyright concerns, or all inference costsMemory cue“Rombach 2022: compress, diffuse, decode”
Keep quoted language in quotation marks while taking notes, even if you expect to paraphrase it later. Record authors, year, title, publisher or conference, URL or DOI, relevant pages, and the date accessed. This protects you from accidental plagiarism in coursework and makes later verification much easier.
The RevisionDojo Computer Science resources for the 2027 syllabus can support retrieval practice after you have built your evidence cards. Use flashcards for terminology, then move to extended responses in the Questionbank so that research becomes usable exam knowledge.
How to reference a study in an exam answer
A Paper 1 answer is not a formal research essay, so do not spend time constructing a full Harvard or MLA bibliography unless an examination instruction explicitly asks for one. A useful exam citation identifies the source sufficiently and, more importantly, explains what the evidence demonstrates.
A weak reference merely name-drops a paper:
Rombach et al. researched latent diffusion models.
A stronger answer integrates the evidence:
Rombach et al. (2022) showed that diffusion could be performed in the compressed latent space of a pretrained autoencoder, reducing computational requirements compared with pixel-space diffusion while retaining strong image quality. For Visionary Studios, this supports choosing a latent diffusion system when producing high-resolution campaign images at scale. However, efficiency alone would not address biased training data or uncertainty over image licensing, so human review and provenance checks would still be necessary.
This structure works because it contains four elements: source identity, relevant finding, application, and qualification. If you cannot remember the full title, an accurate author or organization and year are more useful than an invented detail. Never fabricate a statistic, quotation, page number, or research conclusion.
Common research mistakes
The most frequent problem is over-researching the technology while neglecting application. Students may memorize the diffusion process but fail to explain why speed, cost, control, or output reliability matters to Visionary Studios.
Other avoidable mistakes include:
-
using company marketing as independent evidence of model quality;
-
citing generative AI output without locating the original research;
-
memorizing statistics without methodology or context;
-
discussing copyright law as if it were identical in every jurisdiction;
-
giving only one stakeholder's perspective;
-
naming a study without explaining its relevance;
-
writing a pre-planned essay that does not answer the actual question.
Practise adapting the same evidence to different command terms. “Explain” requires a developed reason, while “evaluate” requires a supported judgment that weighs strengths, limitations, and context. RevisionDojo's guide to preparing for HL Papers 1 and 2 can help with exam structure, while Computer Science predicted papers provide timed application. The 2027 Computer Science exemplars are also useful for studying how technical claims and evaluation can be expressed precisely.
Conclusion
Effective outside research for the case study is focused rather than exhaustive. Build a balanced set of approximately 8-12 credible sources, including foundational diffusion research, current ethics frameworks, applied examples, and critical perspectives. Reduce each source to a memorable claim, its relevance to Visionary Studios, and at least one limitation.
In the examination, cite studies briefly and use the saved time to explain their significance. RevisionDojo's Computer Science notes, Questionbank, predicted papers, and Jojo AI feedback are most useful after the research stage, when you need to turn source knowledge into concise, question-specific Paper 1 answers.
Sources and referenced URLs
-
IB Diploma Computer Science subject brief, first assessment 2027
-
Ho, Jain, and Abbeel: Denoising Diffusion Probabilistic Models
-
Rombach et al.: High-Resolution Image Synthesis with Latent Diffusion Models
-
UNESCO Recommendation on the Ethics of Artificial Intelligence
-
RevisionDojo IB Computer Science resources for first assessment 2027
-
RevisionDojo guide to preparing for Computer Science HL Papers 1 and 2
-
RevisionDojo Computer Science exemplars for first assessment 2027
