Operating Systems practice questions

All members from the legal department

Why wrong: Legal focuses on compliance, not broader ethics.

IT department head and data scientists

Why wrong: Lacks ethical training and independence.

Only senior executives from the company

Why wrong: Executives may have conflicts of interest and lack ethics expertise.

Obtaining a CE marking of conformity through self-assessment only

Why wrong: High-risk systems typically require third-party conformity assessment unless a harmonised standard covers all aspects.

Acquiring liability insurance for AI-related harms

Why wrong: Insurance is not a mandatory requirement of the EU AI Act.

Publishing the training dataset in an open repository

Why wrong: The Act does not require public release of training data; data governance is required but confidentiality may be maintained.

Explicit consent from all data subjects

Why wrong: Consent is not always required; other legal bases may apply.

Data retention period of at least 10 years

Why wrong: Retention periods vary; no universal 10-year requirement.

Remove all demographic features from the model.

Why wrong: Removing features may not eliminate bias if other correlated features exist.

Use a different algorithm that is inherently unbiased.

Why wrong: No algorithm is inherently unbiased; bias depends on data and deployment context.

Hire more diverse data scientists.

Why wrong: Diversity in team is beneficial but does not directly fix model bias.

Cloud-based deployment for scalability.

Why wrong: Deployment method is operational, not governance.

Model accuracy thresholds for production deployment.

Why wrong: Accuracy is a performance metric, not a governance component.

Use a different dataset from a similar hospital without checking demographics

Why wrong: Using a different dataset without validation may perpetuate or introduce new biases.

Manually increase the weight of age-related features in the model

Why wrong: Manual weighting can introduce new biases and lacks systematic validation.

Replace the neural network with a decision tree to simplify decision logic

Why wrong: Model simplification does not guarantee fairness and may reduce performance.

Cross-validation

Why wrong: Cross-validation is a method to evaluate models, not a cause of bias.

High regularization

Why wrong: Regularization reduces variance, not bias.

The event will be stored in a dead-letter queue automatically.

Why wrong: A dead-letter queue requires explicit configuration, not automatic.

The event will be accepted but user_id will be set to null.

Why wrong: Required fields cannot be null.

The event will be accepted with a default user_id of 0.

Why wrong: No default value is defined in the schema.

Obtain explicit consent once and use data indefinitely.

Why wrong: Consent must be specific and can be withdrawn; indefinite use is not allowed.

Store personal data permanently for model improvement.

Why wrong: GDPR mandates data minimization and right to erasure, so permanent storage violates these principles.

Use only aggregated data without any individual records.

Why wrong: Aggregated data may still be considered personal if individuals can be re-identified, and it might not meet business needs.

Developing algorithms

Why wrong: Algorithm development is typically done by data scientists, not the ethics board.

Managing cloud infrastructure

Why wrong: Cloud infrastructure is an IT operations function.

Marketing AI products

Why wrong: Marketing is separate from ethics governance.

Obtain certification from an ISO 27001 auditor

Why wrong: ISO 27001 is for information security, not specifically for AI systems.

Publish the model's source code publicly

Why wrong: Open-source is not a requirement; proprietary systems are allowed.

Register the AI system with the national data protection authority

Why wrong: Registration is not specifically required for all high-risk systems; conformity assessment is.

Recommendation system model

Why wrong: Recommendation systems suggest items, not detect anomalies in server metrics.

Linear regression model

Why wrong: Linear regression predicts a value based on input features, not suitable for detecting anomalies in multivariate metrics.

Image classification model

Why wrong: Image classification is for labeling images, not for time-series or metric anomaly detection.

Reinforcement learning

Why wrong: Reinforcement learning is for sequential decision-making in dynamic environments, not static recommendation.

Unsupervised clustering

Why wrong: Clustering can segment users but does not directly predict which product a user will buy.

Self-supervised learning

Why wrong: Self-supervised learning is typically used for pre-training representations, not directly for recommendation.

Provide a few-shot example of a valid JSON response in the prompt

Why wrong: Few-shot examples guide the model but may not be sufficient for complex constraints; the model can still deviate.

Include a system prompt that says 'You are a helpful coding assistant.'

Why wrong: A system prompt sets the assistant's role but does not enforce a specific output format.

Use chain-of-thought prompting to have the model reason step-by-step before answering

Why wrong: Chain-of-thought improves reasoning quality but does not guarantee structured JSON output.

Use a separate AI model to audit the primary AI's decisions weekly

Why wrong: Auditing is a monitoring measure, not real-time human oversight for each decision.

Allow applicants to appeal AI decisions through a customer service process

Why wrong: Appeals after the fact do not constitute 'human oversight' during the decision-making process as required by the Act.

Provide a dashboard showing the AI's confidence score for each application

Why wrong: Showing confidence scores helps transparency but does not provide the ability to intervene before the decision is made.

Ensure training data is relevant and representative

Why wrong: Data governance requirements mandate data quality and representativeness.

Establish a risk management system

Why wrong: Risk management is a core requirement for high-risk AI systems.

Provide human oversight mechanisms

Why wrong: Human oversight is required for high-risk systems to allow override.

Full transparency by publishing the model's source code and training data

Why wrong: The Act does not require full source code disclosure; it mandates transparency documentation such as model cards.

Annual third-party audits of the model's energy consumption

Why wrong: Energy consumption reporting is recommended but not a mandatory requirement for high-risk systems.

Obtaining explicit consent from each applicant to process their data

Why wrong: While consent may be needed for data processing, it is not the specific human oversight requirement for high-risk AI systems.

Enabling model parallelism

Why wrong: Model parallelism is a performance optimization, not a security or fairness control.

Increasing the model's learning rate

Why wrong: Learning rate affects convergence but does not address bias or security.

Using a larger batch size

Why wrong: Batch size is a training hyperparameter, not a security practice.

Obtain explicit consent from all affected individuals

Why wrong: Consent is a GDPR requirement but not specific to high-risk AI systems under the AI Act.

Open-source the model's code to the public

Why wrong: Open-sourcing is not a requirement; proprietary systems can be high-risk.

Freeze the model to prevent any changes

Why wrong: Freezing the model would ignore drift and degrade performance.

Roll back to a previous model version if performance degrades

Why wrong: Rollback is a reactive measure, not a proactive management strategy.

Manually review all model predictions

Why wrong: Manual review is not scalable or efficient for drift management.