How to trick AI - Observing users' coping strategies and mental models to influence personality perception by intelligent algorithms

Description

A growing body of research shows that personality traits can be automatically inferred from users' digital texts, e.g. conversations with chat bots or social media posts. Like other personal data, we think that users' personality data might be captured in the future without users' knowledge and out of their control. This raises, in our opinion, ethical and privacy concerns and might potentially lead to misuse. We assume that depending on the context, users might not be comfortable with being profiled and its potential consequences. For example, people tend to adapt their behaviour to specific situations to elicit different personality assessments in real life, e.g., people might want to be perceived as very conscientious when being interviewed for a job. Nonetheless, little research has yet examined how users can control the perception of their digital personality profiles, i.e. ``trick'' the algorithm. Hence, we want to investigate the following two research questions: (1) What is the user's mental model of automatic personality assessment? (2) Which coping strategies do users employ to change the system's perception of their personality?

Tasks

The thesis comprises of the following tasks:

• Conduct an extensive survey of related research with a focus on personality assessment and attitude towards data sharing.
• Design a suitable empirical study survey based on previous research
• Recruit n = 30 participants and conduct the user study
• Evaluate the findings using qualitative research analysis methods.