ed and displayed in useful and attractive visualizations if people are to be able to interpret and use it appropriately. This must ultimately be linked to an understanding of the particular kinds of tasks people want to carry out with these systems, and the goals they have in mind as they use the data. When the MyLifeBits project began, there were about 30,000 named items placed in about 1,500 folders. Retrieval was principally by folder location and file name. This quickly turned out to be unwieldy. One possible alternative to this might be to store everything in one large folder and retrieve items using a search engine with knowledge of the folders' content, and with no attention to location. However, full text search is not always enough. Many items require other attributes in order to be found. Unfortunately, with the quantities of information we are dealing with, users are not just unwilling to classify, but are also unable to do it. We are currently exploring a middle ground that lies between the intricately hand-crafted, and arguably brittle, foldering scheme and no organization. To avoid having to become professional curators constructing our own personal classifications, we have become interested in classification sharing. We are experimenting with hierarchical classifications that will be developed by others to be downloaded by the user, and which contain extra information such as synonyms and descriptions to ease their use. These hierarchical classifications should allow scoping searches in a meaningful way and should lead to systems that scale more gracefully. But even with convenient classifications and labels ready to apply, we are still asking the user to become a filing clerk-manually annotating every document, email, photo, or conversation. More must be done automatically. The first, easy step is to stop throwing out any potentially useful metadata. Time is probably the most important attribute in our database, yet some photo-editing programs erase the value for date taken. Even capture itself must be more automatic on this scale, so that the user is not forced to interrupt their normal life in order to become their own biographer. While these are some of the general issues we are tackling, focusing on particular domains and applications is helping us to understand some of the specific technical and design issues. Two examples are (1) visualization to support personal reflection, and (2) monitoring and managing personal health information. Visualization to Support Personal Reflection Reporting tools with appropriate visualization form a class of very useful applications supporting reflection. A simple query-based tool can be remarkably insightful and useful-whether it is "how I spend my time" or "count the space PART II: SOLUTIONS FOR PIM used" by different items. In the simplest case, the amount of work on a document, spreadsheet, or Web page can be logged and displayed. Alternatively, reports can track what the user is working on or perhaps even thinking about-for example, by plotting the "budget" or "nominating committee" tasks against time. Programs that can assist in the creation and visualization of trip diaries and key lifetime event stories will considerably increase use, especially for future viewers who have no familiarity with the content. For example, a fishing trip diary with a timeline, animated maps, and annotations is substantially more valuable to us and our progeny than a collection of unlabelled photos in a labeled folder. Visual autobiographies that summarize a person's life could also be envisioned. Monitoring and Managing Personal Health Information Another possible application of these personal lifetime stores is in the area of monitoring and managing one's health (Oliver & Flores-Mangas 2006). This would imply that our system should hold all of a person's health records, associated health financial transactions, and wellness information in order to assist users, caregivers, and health-care providers. Wearable physiological and environmental monitoring is becoming increasingly pervasive, allowing for the continuous, noninvasive collection of many of these signals, such as heart rate, body temperature, and other vital signs. These records could be maintained by an individual and form a baseline indicator of histher health. Deviations from the average pattern in these metrics could flag potential changes in health. Important components of such a monitoring system are machine learning and data mining algorithms for finding correlations, trends, and deviations from these trends. These trends would be useful not only in keeping a centralized repository of one's health information, but could also allow the system to provide reminders or actions that would promote better health for the user. The explosion of multisource, heterogeneous, and continuous data in general poses interesting challenges for sense-making utilities. 6.4.4 Security, Privacy, and Access Control As we evolve to record everything for all time, we find significantly more opportunities for research around security, privacy, and access control, as well as ownership and the control of what bits get stored and which get thrown away. If the system is built as a stand-alone computer that stores and retrieves digital records, security is of little concern. Locking a hard drive is not difficult. However, ease of access makes it desirable to connect personal digital memories to a broader virtual network, even the Internet. The network will inevitably