| Description |
Sensor-equipped systems that automatically infer attributes about a person (e.g., age and mood) already exist in real-world scenarios (e.g., smart vending machines). More of these kinds of systems are being developed and deployed as companies explore and push their ability to deliver experiences that are immediately customized to a person, whether or not the person is signed into the system. Even though these systems are being deployed more frequently, when and why they make people uncomfortable is understudied. Past research suggests that the perceived necessity of personal data for a system's purpose and the perceived value of that purpose both play a part in users' comfort levels. We use application scenarios (vignettes) on Amazon Mechanical Turk (N = 420) to investigate end-users' comfort and necessity ratings of attributes in the context of specific scenarios. We also ask users to rate how much they personally value the functionality provided by the described systems. Quantitative results from our dataset show links between scenario and attribute necessity and comfort, but also that neither necessity ratings nor scenario values fully explain attribute comfort ratings-attributes themselves still matter. While these quantitative results are perhaps expected based on related work, it is important to establish these relationships for inferred attributes and emerging systems. Diving into our qualitative data in order to better understand why participants reacted as they did to different attributes yields interesting discussion around deeper ideas of privacy expectations and violations. We consider themes expressed in our qualitative data and how that information might be used to draft a set of dimensions that might help reason about why people react more negatively or positively to specific attributes, even absent the context of a specific scenario. We connect our research to the theory of contextual integrity and open the question of how to best apply contextual integrity to research such as ours wherein it seems as though information is being created instead of merely shared. We also reflect on our quantitative analysis and the process through which we were able to better understand, interpret, and present compelling statistical evidence with effect sizes. iv |
