Responsible Research

How to Use AI Ethically in Qualitative Research and Data Analysis

Speaker

Dr. Lorita Tauginienė (Council Member, COPE)

Dr. Lorita Tauginienė (Council Member, COPE)36 min

Quick answer

When using SciSpace in qualitative research, AI may assist with organization and tentative coding only when consent, ethics approval, data protection, and interpretive context are protected. De-identify material, limit access, preserve reflexive notes, review every theme against the source data, and disclose the tool’s role without outsourcing interpretation.

About this session

Dr. Lorita Tauginienė discusses the evolution of research ethics and the integration of generative artificial intelligence in qualitative studies [L1, L5]. The presentation examines historical developments in informed consent and research protocols, specifically addressing the ethical complexities introduced by using modern AI technologies for data collection and analysis [L5].

Qualitative research depends on context, meaning, reflexivity, and an ethical relationship with participants. This session explores where AI can assist with data preparation or pattern discovery and where it may flatten nuance, reproduce bias, expose confidential information, or obscure the researcher’s interpretive role.

What this tutorial covers

Evaluate consent, confidentiality, de-identification, and data-sharing conditions before uploading qualitative material.
Use AI to support transcription cleanup, coding suggestions, organization, or comparison without outsourcing interpretation.
Preserve reflexive notes and an audit trail showing how codes and themes were developed.
Check whether generated categories erase minority views, cultural context, ambiguity, or researcher positionality.

Key concepts explained

Contextual integrity: A participant’s words cannot be separated safely from consent, setting, and intended use.
Reflexivity: Researchers should document how their position and the tool influence interpretation.
Human-led thematic analysis: AI may suggest patterns, but themes require conceptual judgment and evidence.
Auditability: Prompts, codebook changes, disagreements, and validation decisions should be recorded.

How to try it in SciSpace

1
Review consent, ethics approval, data sensitivity, and whether the tool is authorized for the material.
Review consent language, ethics approval, data sensitivity, participant expectations, and whether the tool is authorized. Identify excerpts that should never leave the approved environment.
2
De-identify the dataset and remove unnecessary personal or contextual identifiers.
De-identify the material with a written log and remove unnecessary contextual clues, not just names. Confirm that the resulting data still preserve the meaning needed for analysis.
3
Create a preliminary human codebook or analytical framework before requesting AI suggestions.
Create a preliminary human codebook or analytical framework before requesting AI suggestions. Define code meanings, inclusion boundaries, and examples so suggestions can be compared consistently.
4
Compare AI-proposed codes with raw excerpts, negative cases, and the research team’s interpretation.
Compare proposed codes with raw excerpts, negative cases, memo notes, and the research team’s interpretation. Do not accept frequency or semantic similarity as a substitute for contextual meaning.
5
Document changes, unresolved disagreements, limitations, and how AI assistance was disclosed.
Document changes to the codebook, unresolved disagreements, limitations, and the role of AI. Preserve reflexive notes and describe the assistance accurately in the final methods or disclosure.

Worked example

A qualitative study includes interviews with survivors of domestic violence. The team confirms that consent and ethics approval permit the proposed use, removes identifiers, restricts access, and does not upload raw transcripts until governance review is complete. SciSpace is used only to help organize a de-identified excerpt set and suggest tentative codes. Researchers compare every code with the source context, preserve reflexive memos, and make the final thematic interpretations themselves.

Practice prompt

Using the de-identified excerpts, suggest candidate descriptive codes only. For each code, quote the supporting excerpt, note ambiguity or counterexamples, and do not infer identity, diagnosis, intention, or causality. I will approve the codebook and construct themes.

Key takeaway

AI may assist with organizing qualitative data, but ethical responsibility, reflexive interpretation, and participant protection cannot be automated.