Roberts and Zuckerman presented their research in the use of circumvention tools for internet filtering.
Roberts first notes that the circumvention tools addressed here will not focus on filtering in the local-publisher level in the way that Rebecca MacKinnon addressed in her talk. Instead, they will address circumvention on an international and IP network level, through DNS (domain name server) queries, and on content-based circumvention through keywords on the online network.
All circumvention tools use similar mechanisms, be it processing between the user and server via a proxy (for IP and DNS filtering) or encryption (for keyword filtering). Circumvention also has several different challenges, including ensuring performance, developing sophisticated ways of keeping the proxy themselves from getting blocked, and building trust from users of the tools.
Roberts highlights three different models of hosting: (1) centralized hosting (e.g. UltraSurf), which has high trust; (2) p2p hosting (e.g. psiphon – although the software version has evolved since the time of testing); and (3) algorithm-based routing (e.g. Tor, also discussed in more detail in a later panel), which allows hosting more servers, but it is harder to find people on the servers.
Zuckerman then pointed to a few lessons learned based on their initial study. First, user models proved to be a complex and underemphasized aspect of their research. The funders of the study approached the project from the perspective of human rights activists (how can Burmese journalists stay secure as they report?), so they were looking for the most secure version of circumvention tools. Other users, however (high school students trying to get into Facebook), have a demand for less secure tools with lower stakes. So one weakness of the report is failing to reflect the mix of different user models for these circumvention tools. The current six criteria are utility, usability, security, promotion, sustainability, and openness. Zuckerman suggests that the report can be improved by including better criteria based on different circumvention needs.
One interesting finding was that the usage of single-hop proxies were increasing, and many were looking for single-hop proxies. There were sufficient proxies to be able to sell the right to be at the top of the directory page on their proxies. So the next round of work should be more egalitarian, looking at a much wider range of tools and proxies.
The next step in the project is to study what people are actually doing with their proxy servers by examining data from proxy servers or cyber cafes, analyzing search data for proxy servers, or conducting surveys within human rights and blogger communities. Zuckerman and the research team plan to collect behavioral data to determine user models, and apply their analysis within those models and a new set of tools.