Chapter 4: Metadata Magic 85 metadata is usually defi ned in content production houses. In addi - tion, some metadata can be fi lled in automatically without human intervention. The effort that is invested into entering metadata for a piece content is often inversely proportional to its uniqueness. We use uniqueness as one measure of the return-of-investment analysis for metadata. The uniqueness of an object’s metadata is related to whether the object is self-created or available to a wide audience. Commercial content usually comes in abundant copies. There are numerous reasons for this, such as the nature of our economical system (mass production lowers prices and boosts sales), and digital content is fast and easy to duplicate and distribute. To date, commercial interest has always implied large-scale duplication and distribution. On the other hand, photos are usually unique. Even if two photog- raphers are taking photos at a birthday party, it is highly unlikely that they shoot identical photos. As a consequence, the images feature differences in orientation, lighting, people, objects – even if the photographers take the images exactly at the same time and same place. So even when the physical circumstances match exactly, the contents cannot be the same due to variations in camera electronics and optics, which will render the resulting binary information different from each other. This makes every image unique. Of course, the per- sonal values attached to an image further distinguish it from other similar images. In essence, each self-created content object exists in small numbers, is valuable only to a few people, and thus there are often only a handful of people who can really say something meaningful about such content. On the other hand, the world is full of duplicates of commercial content objects, which implies that there is a larger pool of people that 20 can describe the content. Individual people can then benefi t from others’ metadata. The other users may gather bits and pieces of meta- data from here and there as somebody else has already entered it. In addition, everybody may contribute more metadata (section 4.1.1). Naturally, this kind of sharing of distributed metadata entry will suffer not only unintentional metadata disintegration, but also mischief by people that abuse or try otherwise to spread false data intentionally (we will return to this phenomenon in section 4.8). 20 However, this necessarily introduces noise in the metadata, as different people may enter similar metadata in different ways. Witness artist and song names in P2P fi le sharing networks: there must be 50 ways to spell “Paul Simon”.