Investigating receptiveness to sensing and inference in the home using sensor proxies

Abstract

In-home sensing and inference systems impose privacy risks and social tensions, which can be substantial barriers for the wide adoption of these systems. To understand what might affect people's perceptions and acceptance of inhome sensing and inference systems, we conducted an empirical study with 22 participants from 11 households. The study included in-lab activities, four weeks using sensor proxies in situ, and exit interviews. We report on participants' perceived benefits and concerns of in-home sensing applications and the observed changes of their perceptions throughout the study. We also report on tensions amongst stakeholders around the adoption and use of such systems. We conclude with a discussion on how the ubicomp design space might be sensitized to people's perceived concerns and tensions regarding sensing and inference in the home.

"What Can't Data Be Used For?": Privacy Expectations about Smart TVs in the U.S.

“…Smart TV manufacturers could also follow the guidance emerging from research on sensors found in smart-homes, for example designing signals that clearly communicate the collection and transmission of sensor data [54], [30], [28], [7], [21]. Our study reinforces prior findings that users' expectations vary according to different types of data; we therefore support recommendations to consider how to communicate that different types of data are being collected, when applicable.…”

“…However, they note that differences based on recipient were generally weaker than differences based on data type. In contrast, interviews and in-home experiments by Choe et al [7] found that participants' acceptance of sensor recording depended more on purpose and recipient-and on how data would be recorded and processed-than on data type. In qualitative interviews, Binns et al [4] found that participants' concerns about datasharing were dependent on their pre-existing relationships with and knowledge about the specific parties receiving the data.…”

How Good is 85%?

“…These deployments are very resource-intensive, however, and thus we aim to reduce the overhead of assessing the acceptability of accuracy before such systems are built. Finally, other researchers have proposed methods of formative assessment of ubicomp systems through the concepts of sensor proxies [3] and experience sampling [4], but these methods still require in person interaction with participants, and do not provide explicit guidance on the acceptability of accuracy of inference systems. We believe our method can complement these existing approaches.…”

“…For respondent k on scenario j in application i, with p drawn from a categorical distribution over (-1,0,1) corresponding 3 While we considered using an ordinal or a multinomial logistic regression instead of a binomial regression, ultimately the question when evaluating a classifier here becomes "how many people said the accuracy was acceptable at all? ", in which case this threshold would be applied after regression anyway, so the simpler model suffices while invoking fewer assumptions.…”

Attitudes toward vehicle-based sensing and recording

“…Many technologies include recording and sensing, including home sensing and recording systems [9,18,24], personal and wearable devices ranging from mobile phones and firstperson video devices to personal healthcare devices [22,26,28], as well as CCTV and other infrastructure-based cameras. The increasing ubiquity of such technologies can lead to recording and sensing in varied spaces.…”