An ontological approach to foster the convergence, interoperability and operationalization of frameworks for Trustworthy AI

Abstract: AI systems are consistently evolving in terms of both capability and autonomy with an holistic social impact. In this context of proliferation and fast technological evolution, the scientific community is actively engaged to assure Trustworthy AI. However, in general terms, AI safety research is significantly slower and is facing critical challenges in terms of strategy, consensus and operationalisation. This paper presents AI-Ethics Ontology (AI-EO) which, by leveraging Semantic Technologies on the Web infrastructure and ontology-based knowledge representations, provides an abstracted semantic infrastructure to foster the convergence, interoperability and operationalization of the different frameworks for Trustworthy AI. The current implementation results from the analysis of two relevant case studies to establish a dynamic development process in fact, as well as to enable its iterative evolution according to a formally-defined methodology. The version 1.0 of the Ontology is freely available and has been designed to be conceptually close to target applications, in a context of interoperability, adaptability as a natural response to change and usability.

• The paper presents a semantically rich AI-Ethics Ontology (AI-EO) that converges and operationalizes diverse frameworks for Trustworthy AI.
• It employs an iterative ontology engineering methodology combining manual domain analysis with AI-driven keyword extraction and semantic enrichment.
• The ontology facilitates cross-framework alignment, automated compliance mapping, and dynamic extensibility to support evolving AI governance needs.

An Ontological Infrastructure for Convergence and Interoperability of Trustworthy AI Frameworks

Introduction

The paper presents the AI-Ethics Ontology (AI-EO), which addresses the convergence, interoperability, and operationalization of heterogeneous frameworks for Trustworthy AI. The necessity for such a solution is highlighted by the rapid proliferation of AI capabilities and their impact, juxtaposed with the relative lag and fragmentation in AI safety research and ethical operationalization. Positioning AI-EO within the context of the Semantic Web and leveraging ontology-based formalism, the work intends to serve as a semantic backbone facilitating alignment, traceability, cross-framework interactions, and dynamic extensibility of principles, requirements, and operational guidelines for Trustworthy AI.

Motivation and Related Work

The ongoing diversification of AI ethical frameworks—epitomized by Australia’s AI Ethics Principles and the EU’s Ethics Guidelines for Trustworthy AI—has resulted in a landscape wherein core concepts (e.g., principles, requirements, dimensions of trustworthiness) are specified with differing taxonomies, levels of abstraction, and domains of application. The absence of unified semantic infrastructure stifles strategy alignment, impedes consensus, and inhibits the translation of principles into actionable system-level requirements.

Previous ontological efforts, such as AIPO, TAIR, AIRO, and domain-tailored interventions in robotics ethics, establish the relevance of OWL-based knowledge representations for this domain. However, these offerings are either firmly rooted in specific regulatory or organizational perspectives or narrowly scoped (e.g., role taxonomy or risk management), lacking a flexible, application-proximate federation mechanism across frameworks.

Methodological Approach

AI-EO is developed through an iterative ontology engineering (OE) process facilitated by standard tools (e.g., Protege, OWL 2, Pellet, HermiT), with each iteration anchored in the detailed analysis of a distinct AI ethics framework. This spiral, Agile-inspired methodology encompasses four core stages:

1. Knowledge Structure: Manual domain analysis leads to an ontological schema reflecting central abstractions in the target framework.
2. Knowledge Extraction: AI-driven keyword extraction is combined with human-supervised abstraction and validation, ensuring fidelity and semantic coherence.
3. Semantic Enrichment: Entities are augmented with context, provenance, and method annotations, supporting transparency and traceability.
4. Knowledge Consolidation: Semantic equivalences and disjointness constraints are asserted, reducing redundancy and enhancing internal harmonization.

The content model systematically supports semantic saturation and cross-framework convergence, with incremental knowledge gains diminishing as core consensus patterns emerge. Notably, AI-EO’s iterative design accommodates ongoing input from additional frameworks, reflecting the dynamism of the Trustworthy AI domain.

Ontological Schema and Semantic Mechanisms

The v1.0 AI-EO schema is organized into three conceptual clusters:

• Central Concepts: Framework, Principle, Requirement, FundamentalRight, and AI_Dimension encapsulate the highest-level abstractions.
• Materialization Constructs: Application, UseCase, Scenario, and Example instances are designed for mapping framework elements to concrete contexts.
• Analytical/Classification Concepts: Hierarchies of keyword types (e.g., Risk, Organizational, Developmental), which facilitate domain-specific semantic annotations.

Object properties are rigorously specified, capturing the multiplex relationships (e.g., principle-to-framework, requirement-to-framework, use case equivalence between different contexts) and enabling explicit semantic equivalence and disjointness.

AI-EO employs annotation properties for method, reference, human-readable labelling, and short definitions, supporting external traceability and interpretability. Conventions for equivalence (e.g., cross-framework fairness principles or varying instantiations of accountability across principle/requirement axes) are formalized, enabling not only harmonization but also explicit mapping of ontology-level heterogeneity.

Notably, the ontology is designed as descriptive rather than prescriptive, providing maximal adaptation potential for differently scoped frameworks and use cases without imposing rigid structural mandates.

Applications and Integrability

AI-EO supports complex federated querying across frameworks, surfacing both correspondence and divergence in the conceptualization and operationalization of Trustworthy AI. This supports applications including compliance checking, scenario-based risk/impact analysis, and automated mapping of AI system requirements to evolving regulatory or standards-based expectations.

The design is inherently user-centric, with application-level abstractions directly exposed and annotation mechanisms supporting knowledge graph-based visualization and integration. This positions AI-EO as a foundational asset for knowledge-driven tools in AI governance, risk management, and certification pipelines.

Limitations and Future Research Pathways

AI-EO in its current form is a research-grade prototype, covering two principal frameworks as a proof-of-concept. Integration with external ontologies and vocabularies—both within AI ethics and intersecting domains (e.g., data protection, safety engineering)—is identified as a critical path for scaling interoperability and reuse.

Full automation of knowledge extraction, deeper AI-powered consolidation of semantic clusters, and robust external validation (via deployment in practical AI governance tooling) remain as primary axes for future work. The openness and modularity of AI-EO afford straightforward extension in response to framework evolution, emergent requirements, and community validation feedback.

Conclusion

AI-EO operationalizes an abstracted, semantically rich infrastructure for synthesizing disparate frameworks governing Trustworthy AI, implemented atop Semantic Web standards. Its iterative, user-centric design supports both convergence and adaptability—critical in a domain marked by rapid change, regulatory heterogeneity, and complex stakeholder requirements. While presently a mature prototype, the ontology demonstrates scalable potential as the foundation for cross-framework, application-level AI governance and compliance systems. Future research should focus on multidomain integration, automation of extraction/enrichment, and empirical validation in real-world Trustworthy AI operationalization contexts.