Epistemological Debates in Qualitative Software and AI

This handout is a rough summary of the ongoing academic discussion regarding how Computer-Assisted Qualitative Data Analysis Software (CAQDAS) and Artificial Intelligence (AI) influence the research process. It presents two perspectives out of many possibilities on the relationship between technology and the nature of knowledge.

Perspective 1: The Risk of Positivistic Drift

This viewpoint cautions that the technical structure of software and AI can prioritize “objectivist” outcomes over interpretive depth.

• The mechanical process of “chopping up” transcripts into discrete codes may strip away the narrative flow, treating human experiences as isolated data points rather than holistic stories (Kelle, 1997).

• Software tools like word clouds and coding matrices visually privilege quantity. This may lead a researcher to assume that the most frequent code is the most significant, potentially overlooking rare but critical insights (Bazeley & Jackson, 2019).

• Using AI to “identify” themes relies on mathematical probability. Critics argue this creates a “black box” where meaning is calculated by a model rather than being cocreated through the researcher’s lived experience (Christou, 2025).

• Software and AI use statistics to determine codes.

Perspective 2: The Enhancement of Qualitative Rigor

This viewpoint argues that software is a neutral tool that, when used reflexively, increases the transparency and “trustworthiness” of the findings.

• Software provides a permanent record of the researcher’s decision-making process. Features like “Query Histories” and “Coding Stripes” allow others to see exactly how a researcher moved from raw data to final themes (Bazeley & Jackson, 2019).

• Digital memos and annotations provide a dedicated space for researchers to “bracket” their biases and document their philosophical journey, ensuring the analysis remains grounded in their specific epistemological framework (Kelle, 1997).

• AI as a “Synergistic Partner”: Rather than replacing the researcher, AI can act as a “sparring partner.” It can identify fine-grained patterns or alternative perspectives that a human might miss, which the researcher then evaluates and interprets (Sinha et al, 2024).

Interpretivist/Constructivist Perspective

This is a personal decision/belief. Some researchers never code.

• Many qualitative researchers argue that a theme being identified due to having more data is a mistake. Instead, a theme’s importance is not due to how many times it appears, but by its salience (its power to explain the phenomenon).

• In hand-coding or cutting and pasting in digital docs, a single, deeply insightful quote from one participant might completely change how you understand the other 19 interviews.

• Unlike statistics, which requires a large “n” (sample size) to be “true,” qualitative work often relies on analytical induction. You aren’t looking for how often something happens, but how it happens and what it means.

• When researchers code “ by hand” we are often looking for saturation—the point where you are no longer seeing new information.

• While this may feel like a statistical “tipping point,” it is actually a theoretical one.

• Researchers aren’t counting the largest percent; they are waiting until the “story” the data is telling feels complete and coherent.

Reflective Questions for Researchers

• If a participant mentioned a concept only once, but it provided the “missing link” for your entire theory, would you still include it as a major finding? Why or why not?

• Are you paying as much attention to what participants didn’t say as to what they repeated most often? (Statistically, silence is “zero,” but qualitatively, it can be a significant finding).

• When you call a theme “major,” are you basing that on the number of people who said it,or the emotional/theoretical weight it carries in the context of the study?

• If two people coded this same data by hand, would they find the same “preponderance”? If not, does that make the findings less “objective,” or simply more “interpretive”?

References

Bazeley, P., & Jackson, K. (2019). Qualitative data analysis with NVivo (3rd ed.). SAGE.

Messner, R., Smith, S., & Richards, C. (2025). Artificial intelligence and qualitative data analysis: Epistemological incongruences and the future of the human experience. International Journal of Qualitative Methods, 24, https://doi.org/16094069251371481

Kelle, U. (1997). Theory building in qualitative research and computer programs for the management of textual data. Sociological Research Online, 2(2), 55–69. 10.5153/sro.86

Sinha, R., Solola, I., Nguyen, H., Swanson, H., & Lawrence, L. (2024). The role of generative AI in qualitative research: GPT-4’s contributions to a grounded theory analysis. In Proceedings of the 2024 Symposium on Learning, Design and Technology (LDT ’24) (pp.45–56). Association for Computing Machinery. https://doi.org/10.1145/3663433.3663456