IB DP · PastPaper.sampleTitle

MetadataPastPaper.sampleTitle

Thinka Nov 2025 SL IB Diploma Programme-Style Mock — Digital society

64 PastPaper.marks165 PastPaper.minutes2025
PastPaper.analysisPastPaper.samplePaper
An original Thinka practice paper modelled on the structure and difficulty of the Nov 2025 SL IB Diploma Programme Digital society paper. Not affiliated with or reproduced from IB.

Paper 1 Section A (Structured Questions)

Answer two questions out of four. Each question is worth 20 marks.
12 PastPaper.question · 36 PastPaper.marks
PastPaper.question 1 · Short Response
2 PastPaper.marks
Outline the difference between structured and unstructured data, providing an example of each.
PastPaper.showAnswers

PastPaper.workedSolution

Structured data is data that has been formatted and organized into a clearly defined schema or model, making it easily searchable by algorithms (e.g., a spreadsheet or SQL table containing customer names and phone numbers). Unstructured data is data that does not have a pre-defined conceptual structure or format, making it harder to collect and analyze using traditional methods (e.g., raw audio files, PDFs, or social media posts).

PastPaper.markingScheme

Award [1] mark for outlining structured data with a relevant example. Award [1] mark for outlining unstructured data with a relevant example.
PastPaper.question 2 · Short Response
2 PastPaper.marks
Describe how a feedback loop operates in an autonomous robotic system using a sensor and an actuator.
PastPaper.showAnswers

PastPaper.workedSolution

In an autonomous robot, a sensor continuously collects data from the environment (e.g., detecting an obstacle's distance). This input is processed by the control system, which decides on an action and sends commands to an actuator (e.g., a motor). The actuator executes the physical action (e.g., turning the wheels), and the new state is measured by the sensor again, closing the loop.

PastPaper.markingScheme

Award [1] mark for describing the role of the sensor in gathering environmental data and sending it to the controller. Award [1] mark for describing how the processor signals the actuator to adjust physical behavior based on that data.
PastPaper.question 3 · Short Response
2 PastPaper.marks
Outline how supervised learning differs from unsupervised learning in artificial intelligence systems.
PastPaper.showAnswers

PastPaper.workedSolution

Supervised learning utilizes training datasets that contain labeled inputs with their corresponding correct output targets (e.g., classifying images as 'dog' or 'cat' based on prior labeling). In contrast, unsupervised learning processes unlabeled datasets, where the algorithm attempts to identify patterns, associations, or groupings (e.g., clustering customer purchasing behaviors) without any prior training labels.

PastPaper.markingScheme

Award [1] mark for outlining supervised learning (focus on labeled data/known outcomes). Award [1] mark for outlining unsupervised learning (focus on unlabeled data/pattern discovery).
PastPaper.question 4 · Short Response
2 PastPaper.marks
Describe how algorithmic bias can be introduced into a machine learning algorithm through training data.
PastPaper.showAnswers

PastPaper.workedSolution

Algorithmic bias is introduced via training data in two primary ways: first, if the historical data used to train the system contains pre-existing human prejudices (e.g., historical hiring biases), the model will learn and replicate those biases. Second, if the dataset is skewed or unrepresentative (e.g., lacking diverse facial profiles), the algorithm will perform less accurately for underrepresented groups.

PastPaper.markingScheme

Award [1] mark for identifying a source of bias in the data (e.g., historical human bias, unrepresentative samples). Award [1] mark for describing how this input data causes the model to produce biased outputs or unequal performance.
PastPaper.question 5 · Short Response
2 PastPaper.marks
Describe two ways in which telemedicine can improve patient health outcomes in rural areas.
PastPaper.showAnswers

PastPaper.workedSolution

Telemedicine improves rural health outcomes in two main ways: 1. Access to Specialists: It allows patients in remote areas to consult with urban-based specialists via video conferencing, reducing travel times and delays in critical diagnoses. 2. Remote Monitoring: Wearable devices and online portals let doctors monitor chronic conditions (like diabetes) in real time, enabling early medical intervention before complications occur.

PastPaper.markingScheme

Award [1] mark for describing the first benefit (e.g., access to specialists, reduced travel). Award [1] mark for describing the second benefit (e.g., real-time remote monitoring, early intervention).
PastPaper.question 6 · Short Response
2 PastPaper.marks
Outline how the digital divide can impact an individual's participation in a democratic digital society.
PastPaper.showAnswers

PastPaper.workedSolution

The digital divide represents the gap between those with access to digital technology and those without. An individual lacking reliable internet or devices is excluded from online government services (e-governance), unable to access diverse news media for political awareness, and excluded from virtual civic debates, thereby marginalizing their political influence and democratic voice.

PastPaper.markingScheme

Award [1] mark for defining/outlining an aspect of the digital divide (e.g., lack of device/internet access). Award [1] mark for outlining how this directly limits civic or democratic engagement (e.g., inability to use e-voting, exclusion from public policy debates).
PastPaper.question 7 · Medium Response
2 PastPaper.marks
Explain one reason why training an artificial intelligence (AI) model on a biased dataset might lead to inequitable health outcomes in dermatology.
PastPaper.showAnswers

PastPaper.workedSolution

When an AI model is trained on a dataset that lacks diversity, such as dermatological images predominantly featuring lighter skin tones, the system fails to learn how skin conditions manifest across different skin types. Consequently, when deployed in clinical settings, the AI will have lower accuracy and a higher rate of false negatives for patients with darker skin tones, resulting in delayed diagnoses and poorer healthcare outcomes for these underrepresented populations.

PastPaper.markingScheme

Award [1] for identifying a specific bias or limitation in the training dataset (e.g., underrepresentation of diverse skin tones). Award [1] for explaining how this technical limitation directly results in an inequitable clinical outcome (e.g., misdiagnosis, higher rate of false negatives, or delayed treatment for specific demographic groups).
PastPaper.question 8 · Medium Response
2 PastPaper.marks
Explain one physical safety concern associated with deploying autonomous mobile robots (AMRs) in a shared human-robot warehouse environment.
PastPaper.showAnswers

PastPaper.workedSolution

In a shared warehouse, autonomous mobile robots (AMRs) rely on sensors like LiDAR and cameras to navigate and avoid obstacles. If these sensors become obscured by dust or experience software latency, the AMR may fail to detect a human worker who unexpectedly crosses its path, leading to a collision that can cause physical injury to the worker.

PastPaper.markingScheme

Award [1] for identifying a specific technical or environmental vulnerability (e.g., sensor occlusion, software latency, unpredictable human movement). Award [1] for explaining how this vulnerability leads to a physical safety hazard or collision within the shared environment.
PastPaper.question 9 · Medium Response
2 PastPaper.marks
Explain how the implementation of automated keystroke logging software to monitor remote employees can impact organizational trust.
PastPaper.showAnswers

PastPaper.workedSolution

Implementing highly invasive monitoring technologies like keystroke logging conveys to employees that management does not trust their professionalism or self-management. This constant surveillance can erode mutual respect, diminish employee morale, and lead to a culture of anxiety and resentment, ultimately weakening the overall trust within the organization.

PastPaper.markingScheme

Award [1] for explaining the perception of invasive surveillance or micro-management from the employee perspective. Award [1] for linking this perception directly to the breakdown of mutual trust, lowered morale, or negative organizational culture.
PastPaper.question 10 · Medium Response
2 PastPaper.marks
Explain one privacy risk that arises when a local government publishes open, anonymized public transportation datasets.
PastPaper.showAnswers

PastPaper.workedSolution

Even when transportation datasets are anonymized by removing names or account numbers, individual transit patterns—such as regular daily commute times and specific boarding stations—remain highly unique. An adversary could perform a linkage attack by combining this open dataset with external public data, such as social media check-ins, to re-identify specific individuals and reconstruct their private daily schedules.

PastPaper.markingScheme

Award [1] for identifying the method of re-identification (e.g., cross-referencing, data linkage, or correlating unique travel patterns with external data). Award [1] for explaining how this leads to a specific privacy violation (e.g., tracking a user's physical locations, exposing daily routines).
PastPaper.question 11 · essay
8 PastPaper.marks
Discuss the extent to which the use of artificial intelligence (AI) in recruitment and hiring processes can successfully eliminate human bias.
PastPaper.showAnswers

PastPaper.workedSolution

### Introduction
Artificial Intelligence (AI) in recruitment involves using algorithms, machine learning, and natural language processing to screen resumes, evaluate video interviews, and rank candidates. The primary promise of AI in this context is the reduction or elimination of human bias (such as gender, racial, age, or similarity bias).

### Arguments Supporting the Reduction of Bias
- **Standardization:** AI evaluates every candidate using identical, pre-defined metrics, eliminating the inconsistent and subjective 'gut feelings' of human recruiters.
- **Blinded Screening:** Algorithms can be programmed to ignore demographic signifiers (e.g., names, addresses, graduation years, genders) that often trigger unconscious bias in human reviewers.
- **Efficiency and Scale:** AI can process large, diverse talent pools, potentially surfacing qualified candidates who would have been overlooked due to human fatigue or limited regional networks.

### Arguments Outlining the Limitations and Challenges
- **Algorithmic and Historical Bias:** Machine learning models are trained on historical recruitment data. If a company's past hiring decisions favored a specific demographic (e.g., male software engineers), the AI will learn these patterns as 'success indicators' and systematically replicate this bias (as seen in Amazon’s discontinued AI recruitment tool).
- **Proxy Discrimination:** Even if demographic data is stripped, AI can identify proxy variables (e.g., sports played, clubs joined, or language style) that correlate with specific socio-economic or gender groups, leading to indirect discrimination.
- **Lack of Transparency ('Black Box' Problem):** Complex neural networks can make decisions that are difficult for HR professionals to interpret, making it hard to audit the AI for discriminatory patterns.
- **Narrow Criteria:** AI may struggle to assess soft skills, empathy, or unconventional career paths, favoring highly standardized profiles.

### Conclusion
To a limited extent, AI can mitigate conscious, individual human prejudices during initial screening. However, it cannot completely eliminate bias because it inherently codifies, automates, and scales systemic historical biases present in training datasets. True bias reduction requires a hybrid approach: using AI as an assistive tool while conducting rigorous, regular algorithmic audits and maintaining human accountability.

PastPaper.markingScheme

### Mark Allocation

- **7–8 marks:** The response displays a comprehensive and balanced discussion. It clearly evaluates both sides (mitigation of human bias vs. replication of systemic/algorithmic bias). Technical terms (such as training data, proxies, neural networks, or algorithmic bias) are integrated accurately. The conclusion is logical, well-synthesized, and directly addresses the 'extent' of the prompt.
- **5–6 marks:** The response provides a balanced discussion of AI in hiring, showing clear knowledge of benefits and drawbacks. It mentions technical concepts like training data, but the evaluation may lack depth. The conclusion is present but might not fully synthesize the arguments.
- **3–4 marks:** The response is primarily descriptive, listing the pros and cons of AI in recruitment with limited evaluation. Some concepts may be misunderstood or presented superficially.
- **1–2 marks:** The response shows minimal understanding of AI or bias, providing general or irrelevant comments without addressing the core of the prompt.
PastPaper.question 12 · essay
8 PastPaper.marks
To what extent do autonomous companion robots improve the overall well-being of elderly residents in long-term care facilities?
PastPaper.showAnswers

PastPaper.workedSolution

### Introduction
Autonomous companion robots, such as robotic pets (e.g., PARO the seal) or interactive humanoid robots (e.g., Pepper), are increasingly deployed in eldercare. Well-being in this context encompasses emotional, psychological, and social health, as well as physical safety.

### Positive Impacts on Well-being
- **Mitigation of Loneliness and Social Isolation:** Elderly residents often experience severe isolation. Companion robots provide continuous, low-demand social presence, offering comforting responses to touch and voice.
- **Cognitive and Psychological Stimulation:** Interacting with robots can stimulate memory, reduce agitation in dementia patients, and lower stress levels (cortisol), providing physiological benefits similar to animal-assisted therapy without the associated hygiene risks.
- **Assistance and Independence:** Some robots assist with daily reminders (medication, appointments) or physical support, enhancing residents' sense of agency and autonomy.

### Negative Impacts and Ethical Limitations
- **The 'Deception' and Authenticity Concern:** Many ethicists argue that companion robots foster a form of 'one-way' psychological attachment to an inanimate object that cannot feel or return genuine affection, which can be viewed as patronizing or deceptive to vulnerable individuals.
- **Substitution of Human Care:** There is a critical risk that care facilities may use robots as a cheaper alternative to human staff, leading to decreased genuine human interaction and potential social abandonment.
- **Privacy and Data Governance:** Autonomous robots contain sensors, cameras, and microphones to interact with users. This risks collecting, storing, or transmitting highly sensitive personal and medical data without robust, informed consent from cognitively impaired patients.

### Conclusion
Autonomous companion robots improve elderly well-being to a moderate extent. They are highly effective as supportive tools to alleviate acute feelings of loneliness, stimulate cognitive function, and offer comfort. However, they cannot replace genuine human empathy, touch, and relationship-building. Their implementation must be strictly monitored to ensure they complement, rather than substitute, human care workers.

PastPaper.markingScheme

### Mark Allocation

- **7–8 marks:** The response displays a highly structured, balanced evaluation of autonomous companion robots in eldercare. It thoroughly addresses both benefits (e.g., emotional comfort, cognitive stimulation) and limitations/ethical concerns (e.g., replacement of human care, privacy, emotional deception). A strong, nuanced conclusion directly answers 'to what extent'.
- **5–6 marks:** The response offers a balanced discussion, covering both positive and negative impacts. The technical and ethical aspects (e.g., sensors, authentic connection) are understood, though the evaluation or synthesis in the conclusion may be less refined.
- **3–4 marks:** The response describes companion robots and lists some pros and cons but is largely superficial or descriptive. It lacks structured analysis of the socio-ethical impacts or a clear conclusion.
- **1–2 marks:** The response shows limited understanding of autonomous robots or the context of eldercare, offering general assertions without development.

Paper 2 Section A (Source-based Questions)

Answer all questions based on the provided source booklet.
5 PastPaper.question · 24 PastPaper.marks
PastPaper.question 1 · Identify
1 PastPaper.marks
Refer to the source material describing a smart wearable health device for diabetic patients. The system continually tracks physiological data to manage insulin delivery. Identify one **input** received by this medical system to determine the required dosage of insulin.
PastPaper.showAnswers

PastPaper.workedSolution

The continuous glucose monitor (CGM) subcutaneous sensor tracks the user's interstitial fluid glucose levels. This real-time glucose level data is the primary input sent to the system's controller to calculate the precise insulin dosage needed.

PastPaper.markingScheme

Award [1] mark for identifying a valid input:
- Blood glucose levels / sugar levels / interstitial glucose readings.
- Manual meal-carb inputs entered by the user (if applicable as a bolus input).

Do not accept the hardware sensor itself (e.g. "the CGM sensor"), as the question asks for the input data received. Do not accept "insulin" (which is the physical output/substance administered).
PastPaper.question 2 · Identify
1 PastPaper.marks
Refer to the source material detailing an automated smart-home environmental system. The system uses motion sensors and ambient light sensors to optimize energy use. Identify one **output** produced by this system to minimize electricity consumption when a room is detected to be empty.
PastPaper.showAnswers

PastPaper.workedSolution

When the motion sensors detect no movement for a predefined period (input), the central smart controller processes this data and transmits an output command to turn off active smart bulbs or switch the HVAC system to an energy-saving eco-mode.

PastPaper.markingScheme

Award [1] mark for identifying any valid output action or signal:
- Command/signal to switch off the lighting system.
- Adjusting the thermostat setting (e.g., lowering heating or raising air conditioning thresholds).
- Turning off connected smart appliances/plug outlets.

Reject inputs such as "lack of motion" or "infrared sensor data".
PastPaper.question 3 · Explain
4 PastPaper.marks
With reference to the implementation of automated AI-based decision-making systems in human resource recruitment, explain two ethical concerns related to algorithmic bias that may arise during the candidate screening process.
PastPaper.showAnswers

PastPaper.workedSolution

Ethical Concern 1: Historical Data Bias
- Identification: The AI model can replicate and amplify historical human prejudices present in the training datasets.
- Explanation: If a company's historical hiring data reflects a lack of diversity (for example, predominantly hiring male software engineers), the AI system will learn that being male is a predictor of job success. Consequently, it will systematically downgrade resumes from female applicants, perpetuating systemic inequalities under the guise of objective, data-driven decisions.

Ethical Concern 2: Proxy Discrimination
- Identification: The algorithm may use benign data points as proxies for protected characteristics.
- Explanation: Even if explicit demographic variables such as race, age, or gender are removed from the dataset, the AI may find correlations in proxy data. For instance, using residential zip codes or attendance at specific schools can act as a proxy for race or socioeconomic status. The AI might then reject candidates from certain neighborhoods, resulting in indirect and opaque discrimination.

PastPaper.markingScheme

For each of the two ethical concerns, award marks as follows:
- 1 mark for identifying a valid ethical concern related to algorithmic bias in candidate screening.
- 1 mark for explaining how this bias manifests or impacts individuals in the context of recruitment.

Maximum of 2 marks per concern, up to a total of 4 marks.

Example points:
- Historical/representative bias (training data reflects past inequalities).
- Proxy discrimination (using correlated variables to indirectly discriminate).
- Lack of transparency/Black box issue (inability to audit how the bias occurred, preventing candidates from contesting unfair rejections).

[Note: Reject generic points about 'system errors' or 'hacking' unless explicitly tied to bias and ethical consequences in the screening process.]
PastPaper.question 4 · Compare and Contrast
6 PastPaper.marks
Refer to the scenario where a hospital is evaluating two diagnostic tools:

* **System A**: a deep learning-based neural network trained on millions of global patient medical images.
* **System B**: a rule-based expert system engineered using clinical guidelines and decision trees from local specialists.

Compare and contrast System A and System B in terms of their decision-making processes, explainability, and adaptability to new medical data.
PastPaper.showAnswers

PastPaper.workedSolution

### Similarities:
1. **Purpose**: Both System A and System B are clinical decision support systems designed to automate and assist medical professionals in diagnosing patient conditions from input data.
2. **Dependence on Input Quality**: Both systems are only as good as their foundation. System A is vulnerable to algorithmic bias if its training data lacks diversity, while System B is limited by the potential cognitive biases or knowledge gaps of the human experts who programmed its rules.

### Contrasts:
1. **Decision-making Process**:
* **System A (Deep Learning)** uses a bottom-up, sub-symbolic approach. It detects complex, non-linear statistical patterns across millions of input pixels without being explicitly programmed with medical guidelines.
* **System B (Expert System)** uses a top-down, symbolic approach. It processes input using explicit "if-then" logical rules and decision trees directly coded by human medical specialists.
2. **Explainability**:
* **System A** functions as a "black box". Its inner workings involve thousands of hidden layers and mathematical weights, making it extremely difficult for clinicians to trace the exact reasoning behind a specific diagnosis.
* **System B** is highly explainable ("white box"). It can provide a clear audit trail of the exact logical rules and decision paths that were triggered to reach a clinical conclusion.
3. **Adaptability**:
* **System A** is highly adaptable. It can automatically learn to identify new medical conditions or variations by being retrained on new datasets without human intervention to rewrite code.
* **System B** is rigid. Adapting it to new clinical research or novel diseases requires human experts and programmers to manually modify, test, and integrate new rules into the existing system codebase.

PastPaper.markingScheme

**[1-2 marks]**
* The response identifies basic features of either deep learning (System A) or expert systems (System B) but fails to structure a clear comparison or contrast. Alternatively, it only addresses one of the three required dimensions (decision-making, explainability, adaptability).

**[3-4 marks]**
* The response compares and contrasts the two systems, but lacks balance (e.g., only focuses on differences and omits similarities, or only addresses two of the three requested aspects with sufficient detail). Technical terminology may be inconsistent.

**[5-6 marks]**
* The response provides a balanced, well-structured comparison and contrast of both systems.
* It successfully addresses all three dimensions: decision-making processes, explainability, and adaptability to new medical data.
* Appropriate digital society terminology (such as "black box", "training data", "if-then rules", "symbolic vs sub-symbolic") is used accurately.
PastPaper.question 5 · Extended Discussion
12 PastPaper.marks

Source A: Case study of PulsePredict

PulsePredict is an artificial intelligence (AI) triage tool designed to assist doctors in urban clinics by predicting cardiovascular disease risks. It utilizes a deep learning neural network trained on over ten million historical patient health records. While clinical trials reported an overall diagnostic accuracy of 92%, recent independent reviews revealed that the tool consistently underestimated the risk of cardiovascular events in female patients and minority ethnic groups. Investigations showed that the training dataset was composed of 82% historical data from Caucasian male patients, leading to systematic algorithmic bias.

Question

With reference to Source A and your knowledge of digital society, discuss the ethical and social impacts of using artificial intelligence (AI) systems for automated decision-making in healthcare. In your response, you should analyze the tension between technological efficiency and algorithmic bias, and evaluate the responsibilities of both software developers and healthcare providers in addressing these challenges.

PastPaper.showAnswers

PastPaper.workedSolution

Exemplar Essay Structure & Content:

Introduction:

The integration of artificial intelligence (AI) and automated decision-making (ADM) systems in healthcare promises unprecedented gains in clinical efficiency, diagnostics, and patient throughput. However, as demonstrated by the PulsePredict case study in Source A, these technological advancements can codify and exacerbate pre-existing societal inequalities. This essay discusses the ethical and social impacts of automated healthcare triaging, analyzing the tension between technical optimization and systemic bias, and evaluates the ethical responsibilities of both developers and healthcare providers.

Tension Between Efficiency and Bias:

Technological efficiency is often measured by processing speed, resource optimization, and overall statistical accuracy (such as PulsePredict's 92% accuracy). In clinical settings, automated triage can reduce wait times and alleviate the burden on overstretched healthcare professionals. However, this focus on macro-level efficiency often masks micro-level inequities. Algorithmic bias emerges when the historical data used to train deep learning models does not reflect the diversity of the patient population. In Source A, the training dataset was heavily skewed (82% Caucasian male patients). Consequently, the AI optimized its diagnostic patterns for this dominant demographic, failing to recognize alternative clinical presentations of cardiovascular distress typical in female and minority ethnic patients. This reveals a fundamental ethical tension: optimization for the 'average' user in a biased historical dataset systematically disenfranchises marginalized sub-populations.

Ethical and Social Impacts:

The social impacts of such algorithmic failures are severe. First, there is the immediate risk of physical harm or death. Underestimating cardiovascular risk in female and minority patients means they may be denied life-saving preventative care or urgent intervention. Second, these systems entrench historical inequalities, turning past medical neglect (under-representation in clinical trials and historical health databases) into future algorithmic discrimination. Third, the deployment of biased systems erodes public trust in both medical technology and healthcare institutions. If marginalized communities perceive that diagnostic tools are not built to safeguard their health, they may disengage from formal healthcare systems altogether, widening the health disparities gap.

Evaluation of Developer Responsibilities:

Software developers bear a primary ethical responsibility during the design and training phases of the system lifecycle. First, they must ensure data representation. Developers must actively seek diverse datasets and employ algorithmic techniques to balance underrepresented classes. Second, they must prioritize transparency and explainability (XAI). Deep learning 'black boxes' make it difficult for clinicians to understand why a certain prediction was made; developers must design tools that explain the clinical features driving the AI’s output. Third, developers must subject their algorithms to rigorous, independent bias audits before commercial deployment, establishing clear thresholds for equitable performance across all major demographic groups.

Evaluation of Healthcare Provider Responsibilities:

Healthcare providers cannot outsource their ethical duties to an algorithm. They are responsible for clinical governance and patient safety during the deployment phase. First, they must enforce a 'human-in-the-loop' paradigm, ensuring that AI tools function purely as decision-support systems rather than autonomous decision-makers. Doctors must retain final diagnostic authority and be trained to critically question algorithmic outputs. Second, healthcare institutions must implement continuous local monitoring and post-market surveillance. Since clinical populations change, ongoing local auditing is necessary to catch demographic disparities in real-time. Finally, providers must ensure equity of access, ensuring that reliance on technology does not inadvertently disadvantage patients with lower digital literacy or those visiting underfunded clinics.

Conclusion:

Ultimately, while AI tools like PulsePredict offer profound potential to streamline healthcare delivery, they carry significant ethical risks when trained on unrepresentative data. Mitigating these risks requires a collaborative framework of responsibility: developers must build equitable and transparent models, while healthcare providers must maintain critical oversight, ensuring that automated efficiency never overrides the fundamental clinical commitment to equitable patient care.

PastPaper.markingScheme

Markbands (12 Marks Total):

Level 1 (1–3 marks):

• The response is mostly descriptive and shows a superficial understanding of AI or healthcare technologies.
• Limited reference to the source or digital society concepts.
• Minimal or no analysis of the tension between efficiency and bias, or the responsibilities of stakeholders.

Level 2 (4–6 marks):

• The response outlines some ethical and social impacts of AI in healthcare, referencing Source A.
• There is a basic attempt to address the tension between technological efficiency and algorithmic bias.
• The roles of developers and/or healthcare providers are mentioned but lack deep evaluation or distinction.

Level 3 (7–9 marks):

• The response discusses both ethical and social impacts with clear links to Digital Society concepts (such as Values and Ethics, Power, or Systems).
• There is a balanced analysis of the tension between technological efficiency and systematic bias, utilizing evidence from Source A.
• The response evaluates the responsibilities of both software developers and healthcare providers, though one may be addressed in more depth than the other.

Level 4 (10–12 marks):

• The response demonstrates excellent synthesis, critical thinking, and structured arguments.
• A sophisticated analysis of how automated decision-making creates tensions between clinical efficiency and systemic inequality.
• A comprehensive evaluation of the distinct and shared responsibilities of developers (design, data curation, auditing) and providers (governance, clinical oversight, patient safety).
• Consistently uses precise Digital Society terminology and integrates the source seamlessly.

PastPaper.sampleCTATitle

PastPaper.sampleCTADescription

PastPaper.ctaPrimaryPastPaper.ctaSecondary

PastPaper.sampleStickyMessage

PastPaper.stickyCtaText