From Opaque Black Boxes to Dialogues: Integrating Mechanistic Interpretability with Human Interpretation in Hermeneutic Literature Reviews.
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This dataset contains complete supplementary materials for the ECIS 2026 paper "From Opaque Black Boxes to Dialogues: Integrating Mechanistic Interpretability with Human Interpretation in Hermeneutic Literature Reviews." (Paper ID 1565). This codebook emerged through three grounded theory coding cycles following Klein and Myers (1999), Strauss and Corbin (1990), Gioia et al. (2013), and Senyo et al. (2022). The systematic reduction process demonstrates interpretive rigor: Cycle 1 (Open Coding): 29 first-order concepts from line-by-line analysis. Cycle 2 (Axial Coding): 9 second-order themes through pattern recognition (69% reduction). Cycle 3 (Selective Coding): 4 aggregate dimensions through theoretical integration (86% reduction). All codes are derived from a publicly available 935-line transcript of the Anthropic interpretability team interview (Anthropic, 2025). Final Distribution of 9 Axial Codes: Mechanistic Interpretability (MI): 3 codes (33.3%).Dialogical Hermeneutics (DH): 3 codes (33.3%). Neohumanism (NH): 2 codes (22.2%). Cross-Cutting Themes (CT): 1 code (11.1%).
Fields of Study & Subjects
Explainable Artificial Intelligence (XAI)
Ethics and Social Impacts of AI
Computational and Text Analysis Methods
Publication details
- Journal / source
- AIS Electronic Library (AISeL)
- Publication date
- 30 March 2026
- Language
- EN