2025 IB DP Digital society 模擬試題連答案詳解

To ensure accountability, governments can use policy interventions such as:
1. Algorithmic audits: Mandating that third-party, independent bodies assess the facial recognition software for accuracy, demographic bias, and error rates before and during its deployment.
2. Transparency registries: Requiring agencies to maintain public registries detailing where cameras are placed, who has access to the data, how long data is stored, and the specific legal frameworks under which they operate.

Award 1 mark for each valid policy intervention identified up to a maximum of 2 marks.
- Acceptable answers include: independent algorithmic audits, public registers/databases of active surveillance systems, strict limits on data retention periods with automatic deletion, requiring human-in-the-loop verification for matches, and establishing an independent oversight ombudsman.
- Do not accept vague technical answers like 'improving the AI algorithm' or 'encrypting the database', as the question asks specifically for policy interventions.

Algorithmic bias often originates from the data preparation phase in two main ways:
1. Historical human bias: If historical resumes and hiring decisions from a time when systemic discrimination existed (e.g., gender bias in tech roles) are used to train the model, the algorithm will identify these patterns as positive features and continue to filter out candidates based on those biased historical indicators.
2. Underrepresentation (Sampling bias): If the training dataset predominantly consists of resumes from a specific demographic profile, the model will lack sufficient representative data to accurately assess candidates from underrepresented backgrounds, leading to skewed predictions.

Award 1 mark for each distinct, valid description of how bias is introduced during data preparation, up to a maximum of 2 marks.
- Accept: historical dataset bias, sampling/representation bias, labeling bias (where target variables are defined using biased proxies, like 'performance score' determined by biased managers).
- Do not accept: programmer prejudice (unless linked directly to manual data labeling), post-deployment feedback loops, or security breaches.

Persuasive design features exploit human psychological vulnerabilities to maximize screen time:
- Infinite scroll / Auto-play: By continuously loading content or automatically starting the next video, platforms eliminate the 'stopping cues' that would normally prompt users to close the app, leading to compulsive use and sleep deprivation.
- Push notifications: Using behavioral psychology principles of intermittent reinforcement, notifications (such as likes, mentions, or alerts) trigger dopamine loops, disrupting attention and causing anxiety regarding social validation (FOMO).

Award 1 mark for each valid persuasive design feature outlined, up to a maximum of 2 marks.
- Accept: Infinite scroll/autoplay, push notifications (social validation loops), gamification elements (streaks, daily rewards), read receipts/typing indicators, variable reward algorithms (pull-to-refresh).
- Do not accept: general utility features such as 'having a search bar', 'private messaging', or 'dark mode' unless they are explicitly contextualized as coercive engagement mechanisms.

Automated decision-making (ADM) systems used in public administration can severely impact human rights, particularly the right to due process.

**Way 1: Lack of transparency and explainability ("black box" problem)**
Many ADM systems rely on complex machine learning models where the internal logic is not easily interpretable by humans. If a citizen is denied a crucial public service (such as social welfare, housing assistance, or a visa) by an algorithm, the lack of transparency makes it impossible for them to understand *why* they were rejected. Without a clear explanation or justification, the citizen cannot construct a meaningful appeal, thereby undermining their procedural right to contest administrative decisions.

**Way 2: Absence of meaningful human oversight and bias propagation**
ADM systems are trained on historical datasets that often contain human biases or systemic inequalities. When deployed, the system may systematically discriminate against marginalized communities. If these systems operate without a "human-in-the-loop" who has the authority to override algorithmic outputs, citizens are denied an individualized, empathetic review of their circumstances. This automated rigidity violates the principles of administrative fairness and natural justice.

For each of the two ways explained:
- **1 mark** for identifying a valid way/challenge to due process (e.g., lack of explainability, data bias, lack of human-in-the-loop, inability to appeal).
- **1 mark** for explaining how this challenge directly impacts or infringes upon the citizen's right to due process (e.g., explaining that a lack of explanation prevents a legal challenge or appeal, or explaining how automated bias leads to systemic denial of rights without recourse).

**Maximum total: 4 marks.**

*Note to examiners:* Candidates may refer to specific real-world case studies (such as the Dutch childcare benefits scandal or Robodebt in Australia) to support their explanation, which should be highly credited.

The evaluation of Maruval's NDISTS requires a balanced analysis of competing interests: government efficiency versus individual human rights. Arguments in favor of the system focus on administrative efficiency, fraud reduction, and enhanced security. Centralized digital identity systems can streamline welfare distribution, reduce identity theft, and improve data-driven policymaking. The 'Trust Score' could incentivize positive civic behaviors and lower transaction costs in economic activities. However, the system poses profound ethical and governance challenges. First, surveillance and privacy are severely compromised. Linking biometric, financial, and social media data creates a panoptic state capable of continuous monitoring, violating the fundamental right to privacy. Second, there are critical risks of discrimination and social exclusion. Algorithmic bias in the 'Trust Score' calculations may disproportionately penalize marginalized groups, reinforcing existing inequalities without transparent channels for appeal. Third, this concentration of power in state hands threatens democratic expression, as citizens may self-censor their social media activity to maintain high scores (a 'chilling effect'). In conclusion, while the NDISTS offers significant administrative utility, its current design heavily compromises human rights and values. For the system to be ethical, the Maruval government must implement robust regulatory oversight, independent judicial review, transparent algorithmic criteria, and strict data protection laws to protect citizens from state overreach.

An evaluation essay is assessed out of 8 marks based on the following level descriptors: [7-8 Marks] Demonstrates excellent understanding of digital society concepts (such as power, values, and governance). Offers a balanced, highly detailed evaluation of both positive and negative implications of the NDISTS. Explicitly addresses the tension between security/efficiency and human rights, supported by relevant real-world parallels. Structure is logical, leading to a well-justified conclusion. [5-6 Marks] Demonstrates good understanding of digital society concepts. Offers a structured evaluation with some balance, but may focus more heavily on either the benefits or the risks. Provides clear arguments, but the synthesis or concluding judgment may lack depth. [3-4 Marks] Descriptive response with limited evaluation. Understands basic concepts of digital identity or surveillance but fails to critically analyze the implications of the 'Trust Score' system. The response may be one-sided or unstructured. [1-2 Marks] Superficial response. Lists isolated points about technology, privacy, or government without clear connection to the scenario. Minimal or no evaluation present.

Introduction: Establish the context of State A's elections and the threat of synthetic media to democratic integrity. State the recommendation, prioritizing either Option 1 (regulatory-algorithmic approach) or Option 2 (civil-voluntary approach) with a brief justification. Evaluation of Option 1 (Legislative Bans and Automated Filtering): Pros include rapid response times of automated classifiers to stop viral misinformation before it influences voters and clear legal accountability for platforms. Cons include high rates of false positives in algorithmic detection of synthetic media, which can lead to over-blocking and censorship of legitimate political satire or political speech. This impacts freedom of expression, a core human right, and centralizes power in state-approved filtering mechanisms. Evaluation of Option 2 (Self-Regulation and Media Literacy): Pros include protecting freedom of speech from state overreach, fostering an active and resilient citizenry through media literacy, and allowing platforms to adapt policies faster than rigid legislation can. Cons include the voluntary nature of industry self-regulation, which often lacks enforceable penalties for non-compliance, and the slow, long-term nature of media literacy programs, which may not protect the immediate upcoming election cycle. Synthesis and Recommendation: Compare the options using digital society concepts. Discuss how Option 1 highlights issues of power, governance, and algorithmic bias, while Option 2 relies on digital citizenship, ethics, and decentralized action. Formulate a final recommendation. For example, recommend a phased approach that uses Option 2 as the foundation to avoid authoritarian overreach, but supports it with narrow, legally-binding transparency mandates on digital platforms to ensure accountability without resorting to blanket automated censorship.

Level 1 (1-3 Marks): The response is mainly descriptive, offering basic definitions of synthetic media or censorship with limited focus on the scenario. The recommendation is weak or unsupported. Level 2 (4-6 Marks): The response outlines both options and lists some advantages and disadvantages. There is some attempt to link the discussion to digital society concepts such as security or human rights, but the analysis lacks depth. A recommendation is made but not fully justified. Level 3 (7-9 Marks): A balanced evaluation of both options is provided, discussing the trade-offs of automated filtering (Option 1) and media literacy (Option 2). The response uses relevant terminology (e.g., algorithmic bias, digital governance, freedom of expression). The recommendation is clearly stated, structured, and supported by logical arguments. Level 4 (10-12 Marks): A comprehensive and critical evaluation of both options. The response demonstrates a sophisticated understanding of the tensions between state security, algorithmic capabilities, and human rights. It integrates relevant real-world examples or digital society concepts (e.g., power, expression, governance). The final recommendation is fully justified, addresses potential counterarguments, and explicitly tackles the tension between short-term election security and long-term democratic values.