Multimedia & Privacy
Image courtesy of Mitch Blunt
ICSI researchers are exposing a variety of ways it is possible to aggregate public and seemingly innocuous information in different media from multiple websites and other sources to attack the privacy of Internet users. Users who post photos, videos, and text to public websites often do not realize that personal information from one site can be correlated to information on another site, and that chains of inference can tell much more about them than they are aware they are revealing. We are investigating what these inference chains can show; for example, we have demonstrated how posts on different social-media sites can be linked to the same person using geolocation tags, timestamps, and/or writing style, even if they are using different usernames and trying to maintain separate personas.
This type of cross-referencing is particularly a concern as multimedia content retrieval becomes more sophisticated. Multimedia content retrieval allows users to find videos, photos, and other content on the Internet by searching on a variety of types of information, such as the text in tags, the sounds in audio tracks, visual cues, and automatically analyzed content.
ICSI’s work on global inference and online privacy grew in part out of our work on multimodal location estimation. As we were developing automated methods for establishing the location where a photo or video recording was taken — whether the person who posted the recording chose to reveal it or not — we began to discover the larger implications of the massive increases in the amount of multimedia data being posted to the Internet, and the corresponding massive increases in the capacity of researchers and others to process that data and obtain all sorts of information from it. In particular, our work on cybercasing sparked public interest in the possibilities — benign or problematic — of current global inference capabilities.
Going further into the implications, we are investigating the trade-off between the benefits of sharing information online and the risks of doing so, and exploring how Internet users can protect their privacy. As an outgrowth of this and related research, we are also developing a set of educational tools and exercises to help Internet users understand the privacy implications of the information they share and make more informed choices.
Exploring how public posts — images, videos, or text — across different venues can be correlated to draw inferences and expose information that the poster was not intending to reveal.
Developing educational materials and hands-on learning tools demonstrating what happens to personal information on the Internet, with the goal of empowering K-12 students and college undergrads in making informed choices about privacy.