The Hermeneutic Algorithm Audit: recovering accountability from the calculative register

Hermeneutic Algorithm Audit

The Hermeneutic Algorithm Audit (HAA) sits on the same Heideggerian/Gadamerian ground as my methodology, rather than being added on as ethics-in-the-abstract. Like the AI-Assisted Triple Hermeneutic, the HAA evolved iteratively as I worked through the practice of data collection and interpretation, including the AI-assisted reflexive work, compelled by the centrality of auditability within hermeneutic phenomenological research (HPR).

Indeed, in applied HPR, auditability is understood as an attuned audit trail of interpretive decision-making, in contrast to the tradition of mechanism for quantitative verification. As I carried out my hermeneutic inquiry and developed my evolving framework for reflexivity and rigour (see The Researcher HAA Audit), that traditional reliance on the iterative movement of the hermeneutic circle, where understanding shifts between the part and the whole, required an ethical account of the AI-assisted work. In HPR, auditability ensures that the researcher's interpretive turns, pre-understandings (Vorverständnis), and emerging insights remain transparent, traceable, and trustworthy to the reader. For my work, this required an integrated ethical framework accountable to both the focus on algorithms as phenomena in my study and the disclosive dialogic of my methodology. As with the AATH, this practical requirement led to a return to the literature and a consideration of an ontological orientation to auditing in my AI-assisted processes.

The HAA rests on recovering ¹audīre (to hear) from beneath the calculative sense the word often carries. Where algorithmic auditing as a field audits algorithmically (technical, sociotechnical, metric-driven, audit-as-inspection), my HAA hears as a philosophically informed comportment. Participants' lived experience of algorithmic governance becomes the mode of understanding. In addition, the researcher fore-structures are explicated, not bracketed, such that my own situatedness is also part of what is heard. The consequent relocation of algorithmic accountability from the calculative register into the hermeneutic-phenomenological one is a two-part methodological claim as it unfolded:

1. Participants' lived experience of algorithmic governance becomes the mode of understanding.
2. The researcher's practice-based experience of algorithmic governance becomes a point of critical reflective auditability.

Philosophical and theoretical underpinnings

In Heideggerian philosophy, technology is not merely instrumental but constitutes a way of revealing the world, one that, through enframing (Gestell), orders all beings as standing-reserve (Bestand), resources to be calculated and controlled (Heidegger, 1954/1977). Enframing conceals itself, so what is required is not only measurement from within the technological, but a clearing and listening that lets the concealed show itself. Within this, the etymology of audit carries both senses of the concept: audit as a disposition (attentive hearing, an ethical way of being with the material) and audit as a communal event (accounts rendered orally and heard together at my assessments).

In Patricia Benner's (1994) pioneering application of Heideggerian phenomenology to healthcare inquiry, she advocates for explicit audit trails and interpretive research teams so that external readers can follow the logical pathway from raw text to thematic articulation without reducing phenomenology to subjective impulse. Additionally, the literature addresses institutional standards of rigour by detailing how researchers must document their reflective mulling, how initial biases are surfaced, and how interpretative choices are made within the hermeneutic circle (Kahn, 1993; Cohen et al., 2000). Often cited in qualitative evaluative criteria for research quality, Lincoln & Guba (1985), as constructivists, build their foundational work on trustworthiness and dependability via a clear audit trail that remains the standard adapted by phenomenologists to demonstrate analytical rigour. In IPA, often utilised in the field of psychology, Smith et al. (2022) emphasise traceability through systematic documentation (e.g., reflective journals and transcript annotations), ensuring claims are visibly grounded in the data.

Max van Manen (2014), however, offers a critical view, warning against bureaucratic, mechanical audits. Instead, he calls for hermeneutic rigour, where the writing itself allows the reader to trace and intuitively validate the disclosed meaning.

It is against this philosophical and theoretical grounding that the disclosive dynamics already showing themselves in my AI encounters within the hermeneutic circle came into view and could be named: pressurising, challenging, and revealing and deepening.

Pressurising: AI as an historically situated horizon

An LLM is not neutral; it is a concentrated, algorithmic compression of vast historical text and cultural discourse. Consequently, the AI as "pressurising" the researcher's interpretation is naming how the huge, historically sedimented weight of its training data pushes against the researcher's personal horizon. It brings the accumulated cultural "sediment" of language into immediate tension with our own pre-understandings.

Challenging: dialogic provocation to the researcher

Gadamer argues that understanding only happens when our pre-judgments (Vorurteile) are brought into question through an encounter with an "Other" or an unfamiliar text. Applied critical reflexivity is mindfully used as challenge through critical reflective questions that confront the researcher's horizon and the historical weight of the data corpus. The challenge forces the researcher to ask: "Is the AI reflecting a dominant cultural bias, or is it exposing a blind spot in my own interpretation?" It brings the researcher's implicit assumptions into view.

Revealing & Deepening: Aletheia within the Hermeneutic Circle

Together, the tension between the historical pressure of the corpus and the deliberate challenge to the researcher's biases creates the conditions for unconcealment (aletheia) and further interpretive possibilities, resulting in a deepening of human understanding while the study is still underway.

A methodological framework for the HAA

The three registers of the hearing form my methodological practice and my presentation strategy. This does not imply three uses of one word; they are three scales of a single hearing, the audit conducted at each:

• Register I · Researcher hearing participants: hearing the phenomenon. The originating audit; women's lived experience as the reckoning of the algorithm.
• Register II · Researcher hearing herself: my reflexive turn. My own engagement with the study's algorithm phenomenon as part of my horizon in the hermeneutic circle, and my algorithmic engagement (AI as Disclosive Horizon, the AI-Assisted Triple Hermeneutic) held to the same hearing-standard through a specific set of critical reflective questions for accountability in AI-assisted practice, so the methodology is consistent.
• Register III · Community hearing the researcher: the communal hearing. The audit-as-event, enacted in my viva, with the reader enabled to walk the traceable path as a deferred member of that hearing-community. It becomes ethics I can show: journal extracts, AI use sorted and evidenced rather than declared, chat logs, and my decision memos as slide presentations and articles, leaving a path someone can walk back along.

References

• Benner, P. (Ed.). (1994). Interpretive phenomenology: Embodiment, caring, and ethics in health and illness. SAGE Publications.
• Cohen, M. Z., Kahn, D. L., & Steeves, R. H. (2000). Hermeneutic phenomenological research: A practical guide for nurse researchers. SAGE Publications.
• Gadamer, H.-G. (2004). Truth and method (J. Weinsheimer & D. G. Marshall, Trans.; 2nd rev. ed.). Continuum. (Original work published 1960)
• Heidegger, M. (1977). The question concerning technology. In The question concerning technology and other essays (W. Lovitt, Trans., pp. 3–35). Harper & Row. (Original work published 1954)
• Kahn, D. L. (1993). Ways of discussing validity in qualitative nursing research. Western Journal of Nursing Research, 15(1), 122–126. https://doi.org/10.1177/019394599301500111
• Lincoln, Y. S., & Guba, E. G. (1985). Naturalistic inquiry. SAGE Publications.
• Smith, J. A., Flowers, P., & Larkin, M. (2022). Interpretative phenomenological analysis: Theory, method and research (2nd ed.). SAGE Publications.
• van Manen, M. (2014). Phenomenology of practice: Meaning-giving methods in phenomenological research and writing. Left Coast Press.

AI Use Declaration