figura.directory
Katharina Schneider is creating, experimenting, and researching at the intersection of design and technology.
> figura.directory > projects
Mirror, Mirror on the Screen –
Personality Extraction
with ChatGPT (2024)
// Node.js, OpenAI-API, p5.js, Socket.io
Inspired by the idea of inferring personal information through interactions with a large language model (LLM), this thesis explores how effectively ChatGPT can extract personality traits as defined by the Big Five model from conversation. Three of the five personality traits were considered: openness, conscientiousness, and extraversion.
illustration
process of ChatGPT's personality assessment
In a study, participants (N = 13) first self-assessed their personality using an abridged Big Five inventory (serving as ground truth), then engaged in short dialogs with ChatGPT on four different topics. Afterwards, ChatGPT attempted to estimate each participant’s personality dimensions using two distinct strategies:
- pD: analyzing the entire dialog as a whole
- pN: analyzing each message individually
ChatGPT’s performance was then compared to the participants’ self-assessments. Both the pD strategy (MAE = 0.61; SD= 0.51) and the pN strategy (MAE = 0.61; SD = 0.49) showed similar accuracy and consistency overall, though each exhibited distinct strengths in the estimating specific personality traits.
To support the study, a custom application was developed using Node.js, p5.js, and the OpenAI API, allowing participants to interact with ChatGPT.
6 elements | various tags