Using AI for Generating Visual Vignettes

Vignette experiments, which generally present survey participants with textual descriptions of hypothetical situations consisting of several core dimensions, are a powerful tool for studying decision-making in controlled environments. While they allow us to study decisions which are difficult or impossible to observe through observation or classical experiments, they suffer from limitations in realism and engagement. To overcome these limitations, a number of studies have applied visual or video vignettes which provide participants with scripted videos or photos which vary in specific dimensions; however, such stimuli can be difficult and expensive to produce.
Recent advances in generative AI offer novel opportunities to overcome the aforementioned challenges. This presentation will highlight the capabilities of state-of-the-art image-generating technologies and how they can be used for experimental social science research. Specifically, I will discuss and show how AI technologies can be used to create visual vignettes where key dimensions can be systematically varied, much like textual descriptions. While these visual vignettes offer significant advantages in enhancing realism and engagement, they also introduce new challenges. Visual stimuli can be context-dependent, leading to variations in interpretation across different participants. Additionally, issues of external validity and over-reliance on visual cues may arise, potentially skewing decision-making processes or limiting the generalisability of results. I will discuss these pitfalls and explore strategies for mitigating them when integrating AI-generated visuals into vignette-based research.