Publicação
When Artificial Intelligence Reads the Human: Potentials and Limits of AI-Assisted Qualitative Research
| Resumo: | The integration of generative artificial intelligence (GenAI) into qualitative research has sparked debates over the balance between automation and human discernment, challenging the epistemological foundations of interpretive analysis. This exploratory study critically examines the potential and limitations of human–AI collaboration in qualitative categorisation, based on experiences developed within the PIC-Edu 2024/2025 project. The research was guided by two central questions: (Q1) to what extent can GenAI support exploratory processes of qualitative data analysis, and (Q2) what are its main epistemological, ethical, and operational limitations. Methodologically, the AbductivAI model — abductive in nature — was applied to the categorisation of 324 scientific abstracts from the World Conference on Qualitative Research (WCQR), using ChatGPT-4.0 through interative prompt refinement cycles and cross-validation by two researchers. The results show a mean agreement rate of 72% between automated and human coding, with higher precision in controlled semantic categories but recurrent hallucination errors and classificatory instability in complex interpretive contexts. It is concluded that GenAI is effective for repetitive and pre-analysis tasks but insufficient for contextual and ethical interpretations, reaffirming the indispensability of human mediation. The study proposes a replicable methodological protocol, based on systematic documentation, cross-validation, and ethical supervision, and recommends governance policies ensuring transparency, traceability, and epistemic equity. The conclusions highlight the importance of integrating GenAI as an analytical and formative partner, ensuring methodological rigour and ethical responsibility in contemporary qualitative research. |
|---|---|
| Autores principais: | Mateus, Sara Cristina |
| Outros Autores: | de Lira, Yakamury Rebouças; Costa, António Pedro |
| Assunto: | Inteligência artificial generativa Investigação qualitativa AbductivAI Ética da investigação Colaboração humano–IA Inteligencia artificial generativa Investigación cualitativa AbductivAI Ética de la investigación Colaboración humano–IA Generative artificial intelligence Qualitative research AbductivAI Research ethics Human–AI collaboration |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | artigo |
| Tipo de acesso: | unknown |
| Instituição associada: | Universidade de Aveiro Centro de Investigação Didática e Tecnologia na Formação de Formadores |
| Idioma: | português |
| Origem: | Indagatio Didactica |
| Resumo: | The integration of generative artificial intelligence (GenAI) into qualitative research has sparked debates over the balance between automation and human discernment, challenging the epistemological foundations of interpretive analysis. This exploratory study critically examines the potential and limitations of human–AI collaboration in qualitative categorisation, based on experiences developed within the PIC-Edu 2024/2025 project. The research was guided by two central questions: (Q1) to what extent can GenAI support exploratory processes of qualitative data analysis, and (Q2) what are its main epistemological, ethical, and operational limitations. Methodologically, the AbductivAI model — abductive in nature — was applied to the categorisation of 324 scientific abstracts from the World Conference on Qualitative Research (WCQR), using ChatGPT-4.0 through interative prompt refinement cycles and cross-validation by two researchers. The results show a mean agreement rate of 72% between automated and human coding, with higher precision in controlled semantic categories but recurrent hallucination errors and classificatory instability in complex interpretive contexts. It is concluded that GenAI is effective for repetitive and pre-analysis tasks but insufficient for contextual and ethical interpretations, reaffirming the indispensability of human mediation. The study proposes a replicable methodological protocol, based on systematic documentation, cross-validation, and ethical supervision, and recommends governance policies ensuring transparency, traceability, and epistemic equity. The conclusions highlight the importance of integrating GenAI as an analytical and formative partner, ensuring methodological rigour and ethical responsibility in contemporary qualitative research. |
|---|