Assessing Ambiguous Crime Scenarios

Discrimination and prejudice on the basis of ethnic background is a persistent phenomenon in Western societies across different domains and devaluations of ethnic minorities have particular consequences within the legal justice system (e.g. police brutality against black men in the US). Real-life criminal situations often exhibit some ambiguity and decisions in these situations need to be made rapidly; following dual process theory, it could be expected that these decisions are thus more racially biased than if actors were able to take their time.
This presentation will present an empirical application in the form of an experimental visual vignette study (for which data will be collected in March 2025 through the German Internet Panel). Participants in this study will be shown an AI-generated image that depicts an ambiguous situation that could potentially involve criminal activity. The ethnicity and age of the person shown will be systematically varied and participants will be encouraged to respond to the image and rate the offender's level of suspiciousness and deviance either spontaneously, while being pressured with a countdown, or thoughtfully, after having written a description of the image. The experimental setup will allow to test whether participants’ responses to the images are more racially biased and stereotype-driven when respondents are encouraged to answer spontaneously (and taking into account potential non-compliance). The study thus aims to shed light on how cognitive processes involved in decision-making under time pressure might contribute to racial biases and to highlight how AI technologies can be used to provide respondents with a controlled but engaging stimulus.