Abstract
The technical project is a part of ongoing research conducted for the Defense Advanced Research Projects Agency (DARPA) to design and develop reliable disease detection analytics through data collected from smartphones. The ultimate goal is to design a disease detection system to be deployed for military personnel stationed in combat zones, but the immediate focus of the technical project is to design efficient methods of data collection without a reduction in data richness. The primary research consists of a three-week study, where each week users run a different data collection strategy on their smartphones. Prior to this study, different methods data collection methods were developed, including an adaptive sensing model which pings all smartphone sensors periodically and then turns on those returning non-zero data and a machine learning model that listens to the phone’s accelerometer for a small interval and then predicts whether the phone is in use, turning sensors off/on accordingly. The motivation behind these methods is that data collection from sensors is incredibly battery-intensive, and so minimizing the time spent collecting while still maintaining a rich dataset makes for a much more efficient system. While the study is currently still underway, initial results indicate that both the adaptive sensing model and the machine learning model have improved battery usage over an “always-on” collection strategy, and the machine learning model, in particular, is able to prioritize when the phone is actually in use, and thus lose very little important data in the process.
The STS research paper explores how user-generated data from mobile and smart devices is collected and analyzed for the purposes of online advertising. Companies like Facebook, Amazon, Apple, and Google all collect user data through social media sites and devices and then use this data to make predictions on a user’s preferences and traits. These traits can range from the benign, such as food and drink preferences or preferred clothing styles, to the potentially sensitive, such as political leanings, sexuality, or even predisposition to mental illness. Simultaneously, users are generally unaware of the potentially sensitive insights this data can generate, as well as what other companies, organizations, or individuals have access to this data, either legally or through security breaches. Though this paper does consider user-generated data generally, it focuses on two specific use cases: Instagram and the Amazon Echo, also called Alexa. Both of these technologies collect and store user-generated data and heavily employ that data to gather insights about users. By performing documentary and case study research with the backing STS theories of Actor-Network Theory and Technological Momentum, this paper has uncovered many risks to users, including the risk of breach of privacy by bad actors, harassment based on their characteristics determined through machine learning algorithms, and exposure to discriminatory policies from private companies or governments. There are solutions, regulatory, by companies, and by users, but they are piecemeal, and drastic change would need to happen for users to be truly protected from these risks.
Notes
School of Engineering and Applied Science
Bachelor of Science in Systems and Information Engineering
Technical Advisor: Laura Barnes
STS Advisor: Bryn Seabrook
Technical Team Members: Erin Barrett, Cameron Fard, Hannah Katinas, Charlie Moens, Lauren Perry, Blake Ruddy, Shalin Shah, Ian Tucker
Wilson, Tucker Wilson
