Automating Activity Analysis

Wed. Feb. 2, 2005, 12pm

John Canny

Most knowledge work is collaborative, and collaboration in most workplaces has a number of known frictions and breakdowns. More generally, for context-aware computing its important for a system to be aware of the activities that people are engaged in, not just the location, time etc. For instance, to the question "where are you", people often report an activity rather than a place. Many attempts have been made to model the tasks being performed by users on a computer, but they have been largely unsuccessful. Tasks are short-lived, often interleaved, may involve improvisation, differ in performance from one person to the next, have inconsistent ordering, etc. On the other hand "activities" are long-lived, have a stable imprint, involve a common set of objects between participants. And activity analysis is isomorphic to "topic" or "theme" analysis of a normal text, which works quite well.

This is not a chance observation, but an example of a linguistic principle: "language as symbolic action" - i.e. texts can be viewed as "action logs" of fictional or real characters. I'll describe our work on Activity analysis, how we collect the data and what we do with it. Then I'll list some applications to document sharing, security, and information retrieval. Finally, I'll describe our work on privacy techniques for activity analysis. Over the last 4 years, we have developed an efficient and strong set of cryptographic methods that allow us to do activity analysis with full protection of personal data.

Improving Aviation Safety with Information Visualization: Airflow Hazard Display for Helicopter Pilots

Wed. Feb. 16, 2005, 12pm

Cecilia Aragon

New advances in aviation sensor technology are introducing vast amounts of data into aircraft cockpits. Information visualization interfaces developed by the HCI community are usually assumed to be used in office environments which have fundamentally different user requirements than the cognitively overloaded environment of an aircraft cockpit. Similarly, the field of human factors in aviation has focused on psychological studies and measurements of attention and performance and is only beginning to deal with the overwhelming flood of information that computers are bringing to the cockpit. There is a clear need for data visualization algorithms that are appropriate in the cockpit and do not compromise aviation safety. The synergy of applying HCI techniques to human factors in aviation can benefit both fields.

In this talk, we present the results of a usability study of an airflow hazard visualization system that significantly reduced the crash rate among experienced helicopter pilots flying a high-fidelity rotorcraft flight simulator into hazardous conditions.

We focus on one particular aviation application, but the results may be relevant to user interfaces in other operationally stressful or cognitively overloaded environments, such as emergency response, as compute power becomes available and ubiquitous. We also suggest guidelines for the design of data visualization interfaces in situations where attention to task is critical.

For more information, please see http://www.sims.berkeley.edu/~aragon/.