Exploring the Meaning of Trustworthy and Ethical AI through the Lens of Knowledge Graphs for Explainable AI Infrastructure.

The potential for both intentional and unintentional misuse exists in every technology. Models, ranging from linear regressions to deep neural networks, are built on assumptions about the world. When these assumptions are incorrect, models can produce misleading or biased outcomes. Complex Artificial Intelligence (AI) models are increasingly deployed in critical, high-stakes areas such as climate science, healthcare, supply chain management, critical infrastructure protection, and cybersecurity. To ensure AI is trustworthy and ethical, it must be designed, implemented, and utilized in ways that uphold societal values of fairness and individual rights. Trustworthy and ethical AI is not merely an outcome but a continuous process and practice, emerging from human interactions with AI. Concepts like fairness, bias, trustworthiness, and explainability gain significance only through these interactions. Ethical AI encompasses multiple dimensions, from development to deployment, within the context of mission needs, legal frameworks, governance processes, and technological support. AI replicates certain human functions, including advanced computational analysis and logic-based techniques for machine learning and machine reasoning, to interpret events, support decision-making, and automate actions. As AI models grow more complex, their explainability diminishes, as seen in the limitations of black-box machine learning models and Large Language Models (LLMs). Explainability remains a significant challenge in deploying AI systems. Knowledge graphs have emerged as foundational technology for AI infrastructure, offering a pathway to more explainable, ethical, and transparent systems. There is a notable gap in methods and techniques for mitigating bias in AI. Applied ontology techniques, which provide a semantic representation of data through knowledge graphs, can support the generation of various explanation types. Ontology, the discipline of precisely representing knowledge, constructs frameworks from data, context, and relationships, projecting them onto a knowledge graph. By extending ontology from international standards and backing it with first-order logic, software can validate assertions in explainability in a manner that is itself explainable. Ontologies can be developed to help system designers incorporate different explanation types into AI-enabled systems. Recent successes indicate that ontologies can be operationalized using off-the-shelf technology within six months by leveraging publicly available ontologies and government-funded work. A knowledge graph is a collection of interlinked descriptions of entities—objects, events, or concepts. Unlike traditional databases, knowledge graphs encode not only data but also the relationships between them, using a graph-structured model. Built on semantic technology principles, they utilize ontologies to define relationships between concepts in ways that are understandable to both machines and humans. Knowledge graphs are a sophisticated form of knowledge representation, a critical aspect of AI that involves formalizing information so machines can process it. Accurate and efficient knowledge representation determines an AI system's ability to reason, learn, and make decisions. In AI, machine learning and machine reasoning are distinct yet interconnected disciplines. Machine learning involves algorithms learning from data to make predictions or decisions without explicit programming. Such models, especially deep learning, are often seen as "black boxes" due to their opaque processes and results. Machine reasoning uses logic and rules to simulate human reasoning, offering more transparency and explainability because the decision-making process follows clear logical steps and known rules. Knowledge graphs support both machine learning and machine reasoning by providing a structured, interpretable framework of data and relationships. For machine learning, knowledge graphs enhance learning by adding context and relationships, improving accuracy and robustness. For machine reasoning, they offer a clear, logical structure that aids in tracing decision-making processes, thus enabling explainability. The explainability of AI involves the ability to describe, understand, and predict AI systems' behavior. Knowledge graphs contribute to AI explainability by encoding knowledge in a format accessible to humans and operational for machines. This infrastructure allows for tracing the lineage of AI decisions, offering insights into how conclusions are reached. Knowledge graphs stand at the intersection of machine learning and machine reasoning, providing robust infrastructure that supports the creation of more explainable and interpretable AI systems. They are crucial for applications where transparency is essential, such as healthcare, finance, and autonomous vehicles. By leveraging semantic technologies and structured data, knowledge graphs form the backbone of AI infrastructure, facilitating a deeper understanding of machine behaviors and making AI more accessible and trustworthy. To keep pace with rapid AI advancements, developers, experts, executives, auditors, and users must maintain technical currency. A consensus among experts suggests that AI will continue to advance rapidly, necessitating constant and continuous technical upskilling. This research aims to demonstrate how exploring the meaning of trustworthy and ethical AI requires examination through the lens of applied ontology, semantic technology, and knowledge graphs for explainable AI infrastructure.