Source
urlhttps://doi.org/10.3389/frma.2024.1331589
rawraw/Williams_frma-09-1331589.pdf

TL;DR: A short but sharp perspective arguing that AI automation is structurally in tension with interpretivism’s epistemological assumptions, while supporting partial automation for mechanical tasks. Raises the hypothesis that AI might catalyze a new “computational paradigm” in qualitative research — neither positivist nor interpretivist but something shaped by the affordances of AI and digital data.

Problem

Technology has generally been absorbed into qualitative research without epistemological disruption. Recording interviews, using NVivo, conducting online fieldwork — these changed how qualitative research is conducted without challenging what it fundamentally is. The researcher remained the interpretive authority; technology assisted but did not substitute.

AI changes this. Natural Language Processing algorithms that automate theme identification and coding are not merely tools that help a researcher do qualitative analysis faster — they perform operations on data that qualitative analysis was previously reserved for the human researcher. The question is whether this constitutes a methodological advance, an epistemological violation, or the emergence of something new.

Williams’s entry point is the most basic version of the tension: Lincoln and Guba’s (1985) claim that “the instrument of choice in naturalistic inquiry is the human.” If this is true, what happens when the instrument is replaced by or shares responsibility with an algorithm?

Approach

This is a theoretical perspective piece — not an empirical study but a structured argument about the relationship between AI capabilities, interpretive research traditions, and possible future paradigm configurations. The paper synthesizes the existing literature on NLP in qualitative research and qualitative epistemology, drawing on the paradigm literature (Guba & Lincoln), the reflexivity tradition, and the CAQDAS literature.

The argument is organized around three positions:

  1. The tension is real. Automated programmes with clear rules and formulae do not work well under interpretivism’s assumptions. Contextual understanding requires embodied, relational, situated knowledge — rapport-building, cultural immersion, researcher reflexivity — that LLMs cannot replicate.

  2. Partial automation is viable. Mechanical tasks — transcription, initial code generation, pattern flagging — can be delegated to AI without epistemological compromise, as long as the researcher retains authority over contextual interpretation and final meaning-making.

  3. A new paradigm may be emerging. AI and digital data may be producing conditions under which a “computational paradigm” — neither positivist nor interpretivist — becomes a coherent methodological framework in its own right.

AI’s Role

AI is positioned in two distinct roles that reflect the two sides of the tension:

As a threat to interpretivism: when AI automates the coding and interpretation of qualitative data, it substitutes algorithmic pattern recognition for the contextual, relational, reflexive judgment that interpretivism regards as the source of qualitative knowledge. AI in this role is epistemologically disruptive.

As a partial automation tool: when AI handles transcription, initial coding, and pattern flagging while the researcher retains interpretive authority, it extends research capacity without substituting for the human’s epistemological role. This is the same position taken by anis-french-ai-qualitative-research-2023, brailas-ai-qualitative-research-2025, and most of the corpus.

The “computational paradigm” hypothesis imagines a third role: AI as a constitutive element of a new epistemological framework that has not yet been fully articulated.

Epistemological Stance

Interpretivist, with explicit openness to paradigm evolution. Williams situates himself within the interpretivist tradition — the human as the primary instrument, reflexivity as the quality mechanism, contextual understanding as the goal — but does not treat this as a permanent or natural state. The paradigm-shift hypothesis acknowledges that research paradigms evolve in response to technological and social conditions.

This historical consciousness distinguishes Williams from both the uncritical AI enthusiasts (who ignore epistemological tensions) and the strict interpretivists (who treat the tensions as grounds for rejection). Williams’s position: the tensions are real, some tasks can be delegated, and the field may be moving toward something new.

Rigor and Trustworthiness

As a perspective piece, this paper does not make claims that require evidential support in the conventional sense. Its contribution is conceptual: clarifying the epistemological tension, articulating the partial-automation compromise, and naming the computational-paradigm hypothesis as a possibility worth investigating.

The argument is well-supported by citations to the relevant literatures (Guba & Lincoln, Lincoln & Guba, Braun & Clarke, NLP research) and internally consistent. The computational-paradigm hypothesis is underdeveloped by design — it is raised as a future research question, not asserted as a finding.

Limitations

At under 3,500 words, the paper covers its ground quickly but does not develop any of its positions in depth. The partial-automation compromise is asserted but not operationalized: where exactly is the line between mechanical tasks (delegable) and contextual interpretation (non-delegable)? This is precisely the question that the rest of the corpus is trying to answer, and Williams does not provide a clear answer.

The “computational paradigm” hypothesis is the most novel contribution and the least developed. What would such a paradigm’s quality criteria be? What research questions would it address that neither positivism nor interpretivism can? What institutional conditions would need to exist for it to become established? These questions are not asked, let alone answered.

The paper was submitted in November 2023 (before many of the 2025 frameworks were published) and thus does not engage with CAAI, AbductivAI, GAITA, or other emerging frameworks that represent possible versions of the “computational paradigm” it anticipates.

Connections

What links here