Source
urlhttps://doi.org/10.1111/isj.12504
rawraw/Information Systems Journal - 2024 - Davison - The ethics of using generative AI for qualitative data analysis.pdf

TL;DR: A collective editorial by 11 Information Systems Journal editors, triggered by a specific incident: ATLAS.ti offering free AI analysis of research data in exchange for using that data to train its AI. Identifies the core ethical issues in AI-assisted qualitative research — data ownership, participant consent, transparency, and research integrity — and rejects the utilitarian argument that efficiency benefits justify ethical shortcuts.

Problem

The editorial is unusual in having a concrete origin: an Associate Editor noticed that ATLAS.ti was offering free AI-powered qualitative analysis on the condition that the researcher’s data would be shared with ATLAS.ti for AI training. This raised a specific, practically urgent ethical question: do participants who consent to their data being used for a specific research study also consent to that data being used to train commercial AI systems? Almost certainly not.

The ATLAS.ti incident is not an isolated case — any CAQDAS platform or AI service that processes research data and uses it for model training faces the same question. But it was specific and concrete enough to galvanize an editorial response. The 11 editors spent multiple rounds of deliberation before agreeing on the shared position the editorial represents.

The broader problem the incident reveals: the ethics frameworks that govern qualitative research (informed consent, data protection, participant anonymity, disclosure of analytic methods) were developed before AI was part of the research landscape. They need updating — but not by simply relaxing standards in favor of efficiency.

Approach

This is a collaborative deliberative editorial — not an empirical study or theoretical framework, but a considered collective position on a normative question. The paper’s authority comes from the institutional position of its authors (11 senior editors of a leading IS journal) and the deliberative process they undertook (email discussion, multiple drafts, shared document).

The editorial identifies four core ethical issues:

Data ownership and consent. Research participants consent to specific uses of their data — the study they participated in, the publication of anonymized findings, and nothing beyond unless explicitly agreed. Using their data to train commercial AI systems violates the scope of that consent. Even if data is “de-identified,” the use goes beyond what participants authorized.

Speed ≠ ethics. The utilitarian argument — AI’s efficiency benefits justify accepting some ethical compromises — is explicitly rejected. “The means that we employ to undertake research must be ethical, and must be seen to be ethical by our peers via the peer review system.” An efficiency gain cannot justify unauthorized use of participant data.

The peer review blind spot. If AI-assisted analysis is not disclosed in methods sections, reviewers cannot assess the quality, validity, or ethical propriety of the analysis. Transparency is not just a professional norm — it is constitutive of how the research community evaluates claims. Undisclosed AI use corrupts the review process.

Pattern recognition ≠ insight. Speed and comprehensiveness of AI pattern recognition do not translate to substantive helpfulness or appropriate understanding. Finding patterns faster is not the same as understanding meaning. The editorial resists the elision of speed with rigor that characterizes commercial AI marketing.

AI’s Role

AI is positioned primarily as a risk to research ethics and integrity — a powerful tool that creates new ethical obligations that researchers and software developers have not yet systematically discharged. The editorial does not reject AI categorically but insists that ethical frameworks must govern its use, not follow from its convenience.

The ATLAS.ti incident illustrates a specific AI role not usually discussed: AI as data consumer. When researchers upload transcripts to AI-powered CAQDAS for analysis, they are also potentially providing training data for those systems. This creates a commercial interest in AI adoption that researchers should understand and evaluate independently.

Epistemological Stance

Research ethics / normative, drawing on the information systems research community’s established professional ethics traditions. The paper does not take a position on whether AI can do good qualitative analysis — it addresses the ethical conditions under which any AI analysis must operate.

The explicit statement that the editorial is “entirely written by humans without any Generative Artificial Intelligence contribution or assistance” is itself a normative act: modeling the transparency the editorial calls for.

Rigor and Trustworthiness

The deliberative process — email discussion among 11 senior editors, multiple drafts, and collaborative editing — provides a form of rigor appropriate to an editorial. The diversity of the authors (11 different researchers from different traditions, with “contrasting views about whether GAI should be used in qualitative data analysis”) and the requirement to reach broad agreement strengthens the credibility of the shared positions.

The paper explicitly situates its scope: “Although many other topics related to the use of GAI in research could be discussed… we exclusively focus on the ethics associated with using GAI for qualitative data analysis.” This bounded focus is appropriate for an editorial.

Limitations

As a collective editorial, the paper cannot go as deep into any single issue as a dedicated scholarly paper would. The data ownership argument is compelling but brief. The specific legal and regulatory frameworks governing data use for AI training (GDPR in Europe, IRB requirements in the US) are not analyzed.

The paper focuses on the downstream ethics of AI use in research, but does not address upstream ethics — the ethics of the AI systems themselves (training data sourcing, labor practices, environmental costs). wheeler-technological-reflexivity-2026 briefly raises the environmental cost dimension; the ethics literature as a whole has not fully developed it.

The ATLAS.ti incident, while a useful concrete case, is presented without investigation of ATLAS.ti’s actual data use practices or subsequent policy changes. The editorial’s strength is in identifying the ethical principle; the investigation of specific corporate practice would require different methods.

Connections

What links here