Domain
stringclasses 9
values | Quality Characteristic
stringlengths 5
32
| Definition
stringlengths 54
95
| Guidance
stringlengths 44
166
| Standard
stringclasses 1
value | Curator
stringclasses 1
value |
|---|---|---|---|---|---|
Functional Suitability
|
Functional completeness
|
Degree to which the set of functions covers all specified tasks and user objectives.
|
Implement rigorous requirements gathering to map all user tasks to system functions. Validate coverage through user acceptance testing (UAT). Ref: ISO/IEC/IEEE 29148.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Functional Suitability
|
Functional correctness
|
Degree to which the AI system provides the correct results with the needed degree of precision.
|
Establish ground truth datasets. Validate against accuracy metrics (F1-score, precision). Use adversarial testing. Ref: NIST AI RMF (MEASURE Map).
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Functional Suitability
|
Functional appropriateness
|
Degree to which the functions facilitate the accomplishment of specified tasks and objectives.
|
Conduct task analysis and user studies. Prioritize features based on user value. Ref: ISO 9241-210 (Human-Centered Design).
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Functional Suitability
|
Functional adaptability
|
Degree to which the AI system can be adapted for different specified tasks and environments.
|
Design systems with configurable parameters. Use feature flags and modular architecture for retraining hooks. Ref: MLOps retraining pipelines.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Performance Efficiency
|
Time behaviour
|
Degree to which response/processing times and throughput rates meet requirements.
|
Set SLOs for latency. Optimize models (quantization, pruning). Ref: Google's ML Testing Rules.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Performance Efficiency
|
Resource utilisation
|
Degree to which the amounts and types of resources used meet requirements.
|
Monitor compute/memory usage. Right-size infrastructure and auto-scaling. Ref: Green AI principles.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Performance Efficiency
|
Capacity
|
Degree to which the maximum limits of a product parameter meet requirements.
|
Load/stress testing for maximum users/transactions. Implement rate limiting. Ref: ISO 25010.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Compatibility
|
Co-existence
|
Degree to which an AI system performs efficiently while sharing a common environment.
|
Test in staging environments mirroring production. Ensure no resource monopolization. Ref: ISO 25010.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Compatibility
|
Interoperability
|
Degree to which systems can exchange information and use it.
|
Adopt standard formats (ONNX, PMML) and APIs (REST, gRPC). Schema validation. Ref: NIST AI RMF (Develop).
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
Appropriateness recognisability
|
Degree to which users can recognize whether an AI system is appropriate for their needs.
|
Provide Model Cards/Fact Sheets explaining capabilities and limitations. Ref: MIT Model Cards.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
Learnability
|
Degree to which the system enables the user to learn how to use it effectively.
|
Intuitive UI, interactive tutorials, contextual help. Ref: Nielsen Norman Group Heuristics.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
Operability
|
Degree to which an AI system is easy to operate and control.
|
Consistent UI/APIs. Effective error messages. Automate complex tasks. Ref: ISO 9241-110.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
User error protection
|
Degree to which an AI system protects users against making errors.
|
Input validation, undo functionality, constraints on invalid inputs. Ref: NIST AI RMF (Govern).
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
User interface aesthetics
|
Degree to which the UI enables pleasing and satisfying interaction.
|
Apply design systems (e.g., Material Design). Clean interfaces. Ref: ISO 9241-12x.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
Accessibility
|
Degree to which an AI system can be used by people with the widest range of capabilities.
|
Follow WCAG 2.1 (screen readers, keyboard nav, alt text). Ref: W3C WCAG.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
User controllability
|
Degree to which the user can control the AI system's behavior.
|
Settings for confidence thresholds. Allow override of AI decisions. Ref: NIST AI RMF (Human Oversight).
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Usability
|
Transparency
|
Degree to which functions and decisions are understandable to the user.
|
Implement XAI (LIME, SHAP). Document training data/algorithms. Ref: EU AI Act.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Reliability
|
Maturity
|
Degree to which an AI system meets needs for reliability under normal operation.
|
CI/CD with automated testing. Track MTBF. Canary deployments. Ref: ISO/IEC 25010.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Reliability
|
Availability
|
Degree to which an AI system is operational and accessible when required.
|
Redundancy across availability zones. Monitor uptime/SLAs. Ref: Site Reliability Engineering (SRE).
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Reliability
|
Fault tolerance
|
Degree to which an AI system operates as intended despite faults.
|
Retries with backoff, circuit breakers, fallback mechanisms. Ref: Azure AI Design Patterns.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Reliability
|
Recoverability
|
Degree to which an AI system can recover data and state after failure.
|
Automated backup/restore. Disaster Recovery (DR) plans. Track MTTR. Ref: NIST SP 800-184.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Reliability
|
Robustness
|
Degree to which an AI system functions correctly in the presence of invalid inputs or attacks.
|
Test with noisy/adversarial inputs. Adversarial training. Ref: NIST AI 100-2.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Security
|
Confidentiality
|
Degree to which data are accessible only to those authorized.
|
Encryption (rest/transit). RBAC. Anonymization/Pseudonymization. Ref: ISO/IEC 27001.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Security
|
Integrity
|
Degree to which unauthorized modification is prevented.
|
Hashing/Digital signatures for models. Immutable audit trails. Ref: NIST CSF.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Security
|
Non-repudiation
|
Degree to which actions/events can be proven to have taken place.
|
Secure logging. Digital signatures for attribution. Ref: NIST SP 800-57.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Security
|
Accountability
|
Degree to which actions can be traced uniquely to a responsible entity.
|
Clear model ownership. Audit trails of decisions. Ref: NIST AI RMF (Govern).
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Security
|
Authenticity
|
Degree to which identity of subject/resource can be proved.
|
MFA. Provenance verification of training data/models. Ref: NIST SP 800-63.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Security
|
Intervenability
|
Degree to which an AI system allows for human intervention.
|
Human-in-the-loop (HITL) processes. Kill switches/pause functions. Ref: EU AI Act.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Maintainability
|
Modularity
|
Degree to which changes to one component have minimal impact on others.
|
Loosely coupled architecture. Well-defined interfaces. Ref: Google ML Architecture.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Maintainability
|
Reusability
|
Degree to which an asset can be used in other systems.
|
Package models/features as assets. Containerization (Docker). Ref: IEEE 1517.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Maintainability
|
Analyzability
|
Degree to which one can assess the impact of an intended change.
|
Comprehensive logging/monitoring. Data lineage documentation. Ref: Observability practices.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Maintainability
|
Modifiability
|
Degree to which an AI system can be modified without defects.
|
Version control (code/data/model). Feature toggles. Ref: Martin Fowler.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Maintainability
|
Testability
|
Degree to which test criteria can be established and performed.
|
Isolated test environments. Automated regression tests. Ref: Google ML Testing Rules.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Portability
|
Installability
|
Degree to which an AI system can be successfully installed/uninstalled.
|
Docker/Helm charts. Automated deployment scripts. Ref: DevOps practices.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Portability
|
Replaceability
|
Degree to which an AI system can replace another for the same purpose.
|
Standard interfaces/protocols. Exportable data/models. Ref: ISO 25010.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Portability
|
Adaptability
|
Degree to which an AI system can be adapted for different environments.
|
Environment-agnostic design. Externalized configuration. Ref: 12-Factor App.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Effectiveness
|
Degree to which accurate and complete results are achieved.
|
Metrics aligned to user goals. Task success rate tracking. Ref: ISO 9241-11.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Efficiency
|
Degree to which results are achieved with appropriate resources.
|
Measure time-on-task. Optimize workflows. Ref: ISO 9241-11.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Usefulness
|
Degree to which the system is capable of being used to achieve specified goals.
|
Task analysis. Validate usefulness via feedback. Prioritize high-value features.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Trust
|
Degree to which the user has confidence the system will behave as intended.
|
Reliability, transparency, fairness. Clear limitations. Ref: NIST AI RMF.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Pleasure
|
Degree to which the user obtains pleasure from fulfilling personal needs.
|
User-centered design. Usability testing. Reward user actions.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Comfort
|
Degree to which the user is satisfied with physical comfort.
|
Ergonomic principles. Readable text. Assistive tech support. Ref: ISO 9241.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Transparency (Use)
|
Degree to which the user can understand functions, decisions, and outputs.
|
Natural language explanations. Actionable explanations. Ref: ISO/IEC TR 29119-11.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Economic risk mitigation
|
Degree to which the AI system mitigates potential economic risks.
|
Cost-benefit analysis. Safeguards against financial loss errors. Ref: NIST AI RMF.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Health/safety risk mitigation
|
Degree to which the AI system mitigates health and safety risks.
|
FMEA. Fail-safes. Compliance with IEC 61508.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Environment risk mitigation
|
Degree to which the AI system mitigates environment-related risks.
|
Optimize energy footprint. Renewable energy sources. Ref: Green AI.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Societal/ethical risk mitigation
|
Degree to which the AI system mitigates societal and ethical risks.
|
AI Ethics board. Bias testing. Impact assessments. Ref: EU AI Act.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Context completeness
|
Degree to which the system functions across all intended contexts.
|
Test all contexts. Diverse datasets. Monitor context drift.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
Quality in Use
|
Flexibility
|
Degree to which the system can adapt to new, unanticipated contexts.
|
Modular architecture. Transfer learning capability.
|
ISO/IEC 25059:2023
|
Prof. Hernan Huwyler
|
This dataset establishes a standardized taxonomy of Quality Objectives for AI systems, based on ISO/IEC 25059:2023 (Quality models for AI systems). It adapts the classic software quality model (ISO 25010) specifically for Artificial Intelligence contexts, covering domains such as Transparency, Robustness, Bias Mitigation, and Intervenability.
It is designed to help AI Architects and GRC leaders define "Non-Functional Requirements" (NFRs) and control frameworks.
This framework was curated and adapted by: Prof. Hernan Huwyler, MBC, CPA
This dataset synthesizes the ISO/IEC 25059 standard with actionable guidance for implementation.
The dataset contains the following fields:
GRC teams can import this list to create a control baseline.
Engineering teams use this to ensure they are building the right system.
Auditors can use this checklist to verify if an AI system meets quality standards required by the EU AI Act (which heavily overlaps with ISO 25059).
| Domain | Quality Characteristic | Guidance |
|---|---|---|
| Reliability | Robustness | Test with noisy, out-of-distribution, and adversarial inputs. Use adversarial training. |
| Usability | Transparency | Implement Explainable AI (XAI) techniques (e.g., SHAP). Document model purpose. |
| Security | Intervenability | Design human-in-the-loop processes. Ensure the system can be paused or stopped safely. |
If you use this dataset in research or tooling, please cite:
Huwyler, H. (2024). AI Quality Objectives (ISO/IEC 25059). Hugging Face Datasets.
