RAI Design Modelling

Summary: RAI design modeling can be useful for capturing and analyzing ethical principles in the design process.

Type of pattern: Process pattern

Type of objective: Trustworthiness

Target users: Architects

Impacted stakeholders: Developers, data scientists

Lifecycle stages: Design

Relevant AI ethics principles: Human, societal and environmental wellbeing, human-centered values, fairness, privacy protection and security, reliability and safety, transparency and explainability, contestability, accountability

Mapping to AI regulations/standards: EU AI Act, ISO/IEC 42001:2023 Standard.

Context: AI has been widely adopted across various domains, such as finance and HR. The widespread adoption of AI systems brings growing public interest and concerns with RAI. The autonomous and opaque decision-making of AI systems can lead to erroneous, unintended, or undesired outcomes, such as bias in hiring. To reduce these RAI risks, the development team needs to adhere to ethical principles during the design process of AI systems.

Problem: How can we ensure that ethical principles are incorporated into the design of AI systems?

Solution: Methods of design modeling can be extended and applied to support the modeling of AI components and ethical considerations. This can include using Unified Modeling Language (UML) to describe the architecture of AI systems and represent their ethical aspects, designing formal models that take into account human values, using ontology to model AI system artifacts for accountability, creating RAI knowledge bases to inform design decisions that consider ethical concerns, and using logic programming to implement ethical principles. UML can be an effective language for describing AI systems and providing the necessary information for all stakeholders to make the AI system responsible. An extension of UML could include a declarative graphic notation for AI system architecture, with additional stereotypes/metamodel elements for responsible-AI-by-design reference architecture. Use case diagrams can help define stakeholders and their purposes, which is crucial for achieving accountability. State diagrams are useful for analyzing system states and identifying states that may cause ethical failures. Design patterns, such as the AI mode switcher, can be implemented to shift the state of an AI system to a more human-controlled state. Sequence diagrams can describe human-AI interactions and ensure all necessary explanations are provided.

Benefits:

  • Enhanced the ability to test and validate: RAI design models can provide a complete and clear representation of an AI system’s design, helping to identify RAI issues before deployment and making it easier to test and validate the system.
  • Better communication: RAI design models can help to make the design of complex RAI systems more understandable and allow for better communication and collaboration among stakeholders.

Drawbacks:

  • Time: One disadvantage when using modeling languages is the time required to create and manage the models.
  • Complexity: Creating accurate and comprehensive design models can be challenging as AI systems can be highly complex.

Related patterns:

  • AI mode switcher: An AI mode switcher can be utilized to initiate a state transition and alter the system’s state to a safe state.

Known uses: