To what extent can AI agents be fine-tuned to mimic human interactions?
Guilherme de Figueiredo Cameira · Portuguese National Funding Agency for Science, Research and Technology (RCAAP Project by FCT) · 2025
This thesis investigates the viability of employing AI-powered personas as a complement or alternative to traditional human focus groups in qualitative research. Leveraging VARTHEO’s proprietary methodology, the study compares interaction sessions conducted by human participants with those generated by AI-Powered Personas, both exploring consumer perceptions of product design and positioning. A mixed-methods approach is utilized: data from separate sessions are evaluated using advanced natural language processing tools, notably MiniLM-based models and an agentic evaluation via Google Gemini. The findings highlight the strengths of AI-powered personas in delivering scalable, consistent, and cost-effective insights, particularly useful for preliminary testing or situations with logistical constraints. However, the study also identifies limitations in replicating the full depth of human emotional nuance and creative expression as well as human interaction, suggesting that hybrid research designs may offer a more balanced approach. This research contributes to the evolving discourse on integrating AI in qualitative research and underscores the need for continued refinement in AI methodologies to better mirror complex human behaviours and interactions.