Explainable Artificial Intelligence (XAI) has gained significant traction due to the increasing use of AI systems across various fields. XAI aims to make AI processes and decisions transparent and understandable to users. The paper defines 'explain' as providing clear reasons for a model's decisions, and 'explainable' as the AI's capacity to offer such comprehensible rationales. The authors note a gap in existing research, which often focuses on technical aspects of XAI rather than user understanding. They highlight that explanation and understanding are distinct concepts; explanation is the process of disclosure, while understanding is the cognitive achievement of the user. The paper emphasizes that XAI's goal is not merely to provide explanations but to facilitate genuine user understanding, which is crucial for acceptance and trust. The authors analyze the nature of understanding from philosophical perspectives, considering it as a form of knowledge, belief, or a state of perception in discourse, emphasizing its inherently relational and communicative nature. The debate between internalism and externalism regarding understanding is discussed, with the paper focusing on the relational aspect of understanding in the context of XAI.

The paper reviews existing literature on XAI, highlighting different approaches to explainability, such as transparency requirements, case-based explanation methods, natural language explanation, and counterfactual explanation. It critiques existing user-centric XAI frameworks for failing to adequately consider the perspectives of both AI creators and users. Various philosophical viewpoints on the nature of understanding are examined, including understanding as knowledge, belief, or a perceptual state within discourse. The authors explore the internalist and externalist approaches to understanding, concluding that the relational aspect is key to their analysis of XAI understanding. The paper also touches on the existing challenges in reconciling the efficiency of complex AI models with the ethical imperative for explainability.

The paper proposes a novel framework: the Ritual Dialog Framework (RDF). The RDF is based on the idea that understanding of XAI is not just about providing information, but also about building trust and fostering a sense of shared understanding between AI creators and users. The paper divides user understanding of XAI into three modes: contrastive, functional, and transparency understanding. **Contrastive understanding** focuses on explaining AI model results by comparing actual outcomes with counterfactual scenarios (what-if analysis). It emphasizes causal traceability and avoids deep alterations to the model's internal structure. The authors discuss the advantages of its non-intrusive nature, but also highlight limitations such as overlooking feature explanations within the model, a deterministic attitude, and a relativistic conclusion due to the model-specificity of different approaches. **Functional understanding** centers on the AI model's function and assesses whether the output accomplishes its intended purpose. It emphasizes the model's efficacy in achieving a goal. However, the paper notes that functional understanding might not always lead to effective explainability, can ignore model features unrelated to the output, might not adequately address internal systemic responsibility, and can be biased due to reliance on human explanations and a focus on steady-state processes. **Transparency understanding** focuses on disclosing the internal workings of the AI model to ensure users' right to know. It aims to build trust by reducing barriers to understanding and establishing the model's reliability. The authors identify the challenges in balancing user understanding with privacy concerns and highlight that transparency explanations are susceptible to biases from preconceptions in public opinion, and cognitive penetration. The RDF, inspired by anthropological concepts of ritual, aims to create a structured dialog between AI creators and users, promoting trust and shared understanding. The authors emphasize that this dialog is not just an information exchange but a symbolic interaction that builds trust and legitimizes the technology.

The paper's main contribution is the proposal of the Ritual Dialog Framework (RDF) as a solution to the explainability dilemma in XAI. The RDF addresses the shortcomings of current XAI approaches by focusing on building trust and fostering a shared understanding between AI creators and users through structured dialogue. The paper argues that the current lack of user understanding stems from a failure to consider the social and cultural context of AI interaction and the importance of trust-building mechanisms. The three modes of understanding—contrastive, functional, and transparency—are analyzed, revealing inherent limitations in each approach, particularly concerning the balance between technical efficiency and ethical transparency. The authors argue that a purely technical approach to explainability is insufficient for achieving user trust and acceptance. They analyze the complexities of achieving adequate user understanding, highlighting issues such as the selective nature of human explanation preferences, the potential for biased explanations, and the challenges of reconciling different philosophical perspectives on understanding. The RDF is positioned as a sociotechnical intervention designed to bridge the gap between AI creators and users, promoting mutual understanding and trust.

The RDF addresses the central problem identified in the paper: the current lack of a framework for establishing trust between AI creators and users. The authors argue that building trust is crucial for the social acceptance of XAI. The RDF aims to achieve this by structuring the interaction between the two parties as a ritualistic dialog. This structured interaction helps to address the limitations of the three modes of understanding discussed earlier, providing a more holistic approach to XAI explainability. The emphasis on dialog and trust-building goes beyond simply providing technical explanations, acknowledging the social and cultural factors that influence user acceptance of AI. By emphasizing mutual understanding and commitment, the RDF aims to promote responsible AI development and deployment. The success of the RDF depends on several factors, including the willingness of both creators and users to participate in the dialog, the establishment of a shared objective, and the ability to predict and integrate new information during the interaction.

This paper analyzes user understanding of XAI through contrastive, functional, and transparency perspectives, identifying inherent dilemmas in achieving widespread user acceptance. The key contribution is the proposed Ritual Dialog Framework (RDF), designed to foster trust and shared understanding between AI creators and users through structured dialog. RDF transcends technical explanations, addressing social and cultural factors influencing trust in AI. Future research could focus on empirical testing of RDF efficacy and exploring cultural variations in its implementation.

The RDF is a theoretical framework, and its practical effectiveness requires further empirical testing. The study does not account for diverse levels of technical expertise amongst users, potentially affecting their participation and understanding within the RDF. Cultural factors influencing the interpretation of ‘trust’ and participation in dialogues remain largely unexplored within the framework.