ISTQB CT-AI
Quality Characteristics and Acceptance Criteria
Quality characteristics according to ISO/IEC 25059
- Which standard extends the quality model of ISO/IEC 25010 with AI-specific aspects? Easy
- From which two perspectives does ISO/IEC 25059 evaluate AI-based systems? Easy
- Why was the characteristic of functional correctness adjusted for AI-based systems? Easy
- What does the quality characteristic functional adaptability describe? Easy
- Which quality characteristic describes that a person or another system can step into the operation of an AI system in time? Easy
- In ISO/IEC 25010 of 2023, which term takes the place of what used to be called usability? Easy
- What does the quality characteristic transparency refer to in AI-based systems? Easy
- What does the quality characteristic AI robustness describe? Easy
- Which quality characteristic describes how well an operator can step into an AI system in time to avert damage or danger? Easy
- Which perspective does the quality characteristic societal and ethical risk mitigation belong to? Easy
AI and safety
- What characterizes safety-related systems? Easy
- How are the requirements of many AI-based systems implicitly provided? Easy
- Which tool is an example of explainable AI techniques? Easy
- How does the EU AI Act classify AI systems used as safety components, for example in aviation, medical devices or vehicles? Easy
- How do mature international functional safety standards deal with AI? Easy
Acceptance criteria
- What form do acceptance criteria for AI-based systems often take? Easy
- Which quality characteristic does the acceptance criterion "accuracy of 95% for an image recognition system" belong to? Easy
- An acceptance criterion states: "If an autonomous drone loses its GPS signal and reports an emergency, a supervisor can take over control within 0.5 seconds." Which quality characteristic is being tested? Easy
- An acceptance criterion requires the dashboard and API to return the unique version identifier of the deployed prediction model and a link to its documentation. Which quality characteristic is concerned? Easy
- An acceptance criterion states: "A chatbot must pass an internal red teaming assessment with at least 95% and refuse content that promotes violence, self-harm or hate speech." Which quality characteristic is being tested? Easy
Quality characteristics according to ISO/IEC 25059
- Which quality characteristics are new or modified in ISO/IEC 25059 compared with ISO/IEC 25010? Medium
- Which parent characteristic is functional adaptability assigned to as a new sub-characteristic? Medium
- Which parent characteristic is AI robustness assigned to as a new sub-characteristic? Medium
- Which parent characteristic is intervenability assigned to as a new sub-characteristic? Medium
- Which parent characteristics is transparency assigned to as a new sub-characteristic? Medium
- Under which circumstances should a robust AI system maintain its functional correctness? Medium
- Which areas does the characteristic societal and ethical risk mitigation cover, among others? Medium
- How do the quality characteristics of ISO/IEC 25059 affect testing? Medium
AI and safety
- Which additional challenges does AI bring to safety-related systems? Medium
- In connection with non-determinism, what can cause unexpected behavior even in rigorously tested AI models? Medium
- Why does self-learning undermine the safety case of a system? Medium
- Which is an example of a safety guard for self-learning systems? Medium
- Which limitations do explainable AI techniques such as LIME have? Medium
Acceptance criteria
- An acceptance criterion states: "A video streaming service adjusts its homepage to recommend at least 40% documentaries after a user has watched three full-length documentaries." Which characteristic is being tested? Medium
- An acceptance criterion states: "If the outdoor AI camera overheats, it switches to a simpler, power-saving model instead of failing." Which characteristic is being tested? Medium
- An acceptance criterion states: "Before the AI-controlled signal box carries out a switch setting marked as critical, the dispatcher has 20 seconds to stop it." Which characteristic is being tested? Medium
- Which acceptance criteria are examples for safety? Medium
- Which two standards provide the comprehensive framework for defining software quality in acceptance criteria for AI-based systems? Medium
- An acceptance criterion requires that the purchasing team of a bought-in ML model can trace where its training data comes from. Which characteristic is being tested? Medium
- Which acceptance criteria are examples for AI functional correctness? Medium
Quality characteristics according to ISO/IEC 25059
- A business unit requires for an ML classification system: "The system must not produce a single misclassification." How is this requirement to be assessed? Hard
- An image recognition system delivers good results on clean test images, but many errors on slightly noisy or deliberately manipulated images. Which quality characteristic is deficient? Hard
- A credit system rejects applications without applicants or clerks learning which model in which version is used and where the training data comes from. Which quality characteristic is affected? Hard
- How do user controllability and intervenability differ? Hard
- Which mappings of quality characteristic and perspective are correct according to ISO/IEC 25059? Hard
- A recommender system adapts to new user trends during operation, but loses accuracy against manipulated ratings. Which characteristics need to be weighed against each other here? Hard
AI and safety
- A manufacturer wants to demonstrate traceability from requirements to implementation for an AI-based brake assist system. Why is this difficult? Hard
- A rigorously tested AI model for an industrial control system reacts differently in the field to almost identical inputs. Which challenge is shown here? Hard
- A self-learning assistant in a control center was comprehensively tested before deployment. Which measures help so that no dangerous behavior creeps in during operation? Hard
- A team wants to use LIME for a safety-related AI system to make decisions traceable. What must be considered? Hard
- A medical device manufacturer wants to use an AI module as a safety component and rely solely on a mature functional safety standard. Which problems are there? Hard
- Why is developing safety-related systems with AI harder than without AI? Hard
Acceptance criteria
- An acceptance criterion states: "The farm control system notifies the farmer when a sensor's visual performance degrades by more than 30% and allows immediate manual override; without a response, it fully deactivates from 50%." Which characteristic does it belong to? Hard
- Which acceptance criteria test AI robustness? Hard
- Which acceptance criteria test intervenability? Hard
- An acceptance criterion for a nuclear power plant requires that 100% of the input-output relationships of the ML model can be mapped by an explainability tool with an average accuracy of at least 99.9%. Which area does it belong to? Hard
- An acceptance criterion for a sentencing system requires that it does not discriminate between population groups based on a specified fairness metric. Which statements about this are correct? Hard
- A project manager wants to carry out the acceptance of an AI system like classic software using individual test cases with "pass/fail". What speaks against this? Hard
- Which acceptance criterion would be suitable for the functional adaptability of an engine management system? Hard
- A defect prediction system is to identify at least 90% of the actually defective modules as defective. Which statements about this are correct? Hard