SBE: Medium: Towards Personalized Privacy Assistants

无论是在智能手机上、浏览器上还是在社交网络上，人们都面临着越来越多的难以管理的隐私设置。我们需要的是一种新的、更具可扩展性的范式，使他们能够重新获得对数据收集和使用的控制。人们在智能手机上下载的手机应用尤其如此。事实证明，这些应用程序会收集和共享各种各样的敏感数据，而用户却无法跟上。如果所有用户对移动应用程序的数据收集和共享实践都有同样的看法，那么很容易让应用程序预先配置为只允许他们感到舒服的做法。不幸的是，之前的研究表明情况并非如此。这个项目正在研究一种新技术，旨在大大简化配置隐私设置的过程，比如那些与移动应用程序相关的设置。美国有近2亿智能手机用户，平均每人设备上有近50个移动应用程序，这个项目可能会对美国人的日常隐私产生重大影响。具体来说，本研究利用隐私偏好建模、机器学习和对话技术的最新进展，开发个性化隐私助手，能够学习人们的隐私偏好，并代表他们半自动地配置许多隐私设置。研究人员正在评估个性化隐私助手的不同配置，特别关注人类受试者实验，旨在评估它们对隐私决策和用户行为的影响。评估的配置在与用户对话的风格和频率、使用机器学习驱动这些对话的方式以及配置隐私设置的自动化程度方面有所不同。人体实验研究的因素包括不同技术配置对用户隐私设置舒适度的影响，他们的整体意识和控制感，以及短期和长期的行为影响。其他重要因素包括用户负担、中断频率和总体用户满意度。

Whether it is on their smartphones, in their browsers or on social networks, people are confronted with an increasingly unmanageable number of privacy settings. What is needed is a new, more scalable paradigm that empowers them to regain control over the collection and use of their data. This is particularly the case for mobile apps people download on their smartphones. These apps have been shown to collect and share a wide variety of sensitive data, with users unable to keep up. If all users felt the same way about the data collection and sharing practices of mobile apps, it would easy to have the apps pre-configured to only allow for those practices with which they are comfortable. Unfortunately, prior research has shown that this is not the case. This project is studying novel technology intended to significantly simplify the process of configuring privacy settings such as those associated with mobile apps. With close to 200 million US smartphone users, each with an average of nearly 50 mobile apps on their devices, this project could have a significant impact on the privacy of everyday Americans. Specifically, this research harnesses recent advances in privacy preference modeling, machine learning and dialogue technologies to develop personalized privacy assistants capable of learning people's privacy preferences and of semi-automatically configuring many privacy settings on their behalf. The researchers are evaluating different configurations of personalized privacy assistants, focusing in particular on human subject experiments intended to evaluate their impact on privacy decision making and user behavior. Configurations being evaluated differ in the style and frequency of dialogues with users, the way in which machine learning is used to drive these dialogues and the level of automation in configuring privacy settings. Human subject experiments look at factors that include the impact of different configurations of the technologies on the level of comfort users have with their privacy settings, their overall awareness and sense of control, and both short-term and long-term behavioral effects. Other important factors include user burden, frequency of interruptions and overall user satisfaction.