A team from TUD Dresden, the Max Planck Institute for Human Development, and the University of Basel has developed a novel method that pairs observed choices with participants' written descriptions to reveal why people decide as they do. The approach, detailed by the Center Synergy of Systems (SynoSys), promises to move beyond traditional behavioral data.
By integrating large language models with free-text answers, researchers can now capture subjective reasoning that standard experiments miss. This allows them to study decision-making in richer detail, bridging the gap between quantitative data and qualitative insight.
The method relies on participants explaining their own thought processes in natural language, which LLMs then analyze alongside choice data. Early results suggest this combination yields more nuanced understanding of factors like risk tolerance, social norms, and cognitive biases.
If validated further, the technique could transform fields from economics to public health, offering a window into why people ignore warnings, choose certain products, or vote a particular way. The team plans to test the framework on larger, more diverse populations.
Critics caution that LLMs may still misinterpret ambiguous phrasing or embed their own biases into analysis, potentially skewing results. The researchers acknowledge these limitations and call for careful validation against controlled experiments.