92 Personal Content Experience: Managing Digital Life in the Mobile Age experiences. In case references to other people are encountered, the analysis can also be expanded to these persons’ personal content, if they so allow. Such rich networks that bind together social relationships and the related content objects truly enhance our abilities to enjoy and re-live past events. The attributes and relations of metadata together form a language for talking about data. In computer science, such languages have been pioneered in the subfi eld of artifi cial intelligence (AI), known as knowl- edge representation. That area of research seeks to defi ne ways of expressing data (or metadata) in such a way that certain computational problems become easier to solve. For instance, to solve a navigational problem on a map with multiple cities and roads connecting them, a useful data representation is a graph with cities as nodes and roads as arcs. Given the right knowledge representation, many algorithms become quick and powerful. Much of the research in AI continues to concentrate on fi nding suitable representations of the wonderfully complex world we live in. We feel it is useful to regard metadata formats as knowledge rep- resentation languages. While not really aimed at generic AI problems, such languages can be indispensable for the more specifi c tasks that are needed in manipulating content. Particularly, as we are trying to take the content manipulation from mere computation towards a more user-oriented semantic interaction, it will become necessary to capture some of the semantic information of the content objects. Relations are an essential part of that information, as are the attributes. Most metadata representations and formats have remained focused on attributes. Where relations are retained, they are frequently con- verted to text strings (or, at best, URLs) that are usually targeted for human consumption. That works if relational metadata can be omitted from automated analyses. We choose instead to encode both attributes and relations with machine-readable symbolic identifi ers for later auto- mated use. Chapter 5 describes our design approach in detail. 4.4.1 People as First-Class Metadata Ideally, people should be treated as fi rst-class metadata items (Figure 4-6). However, existing metadata formats normally reference people 23 as text strings. This works for simply displaying names, but falls short 23 There are good reasons for this. It is hard enough to deal with swapping fi rst names and last names, translating Chinese names to western representations, or track ladies’ surnames through six marriages. Plain (Unicode) text is the least common denominator for naming people. The technical solutions for identity representation are still in the works, hopefully surfacing from the Internet community in the next few years.