Chapter 8: Timeshifting Life 325 by other persons. Moreover, automated handling of content will require machine-readable descriptions. • Machine-readable metadata is needed for automation Metadata creation: the trend is towards automatic metadata creation – especially when manual annotation does not pay. Online communi- ties will (semi-automatically) annotate content through tagging and ontologies, in cases of content that is meaningful to and desired by a large number of users. Interactions between people in general will turn out to be important sources of metadata. One major metadata creator will be the mass-market industry, spurning out huge amounts of meta- data via product information, advertisements, services, and so forth. • Communities will be a key source of metadata Metadata usage: basic metadata will be commonplace, being adapted to people’s everyday lives. Raw metadata will be mostly invisible to the users, yet it will be everywhere, deeply embedded in applications, operating systems, data, and physical objects. Occasionally people will see glimpses of metadata as the tip of the iceberg, when automatic rating or summarization factors are presented to the user, for example. Metadata will be multi-form and used in unexpected novel ways. Meta- data support becomes a competitive “must have” for applications. Application interworking will be the norm, not the exception. • Good metadata is invisible! Technology: The Semantic Web will make breakthroughs within limited domains. Generic ontologies will still be largely just dreams. Tags and folksonomies will run into complexity and scale-up problems. Decomposition of media items will become commonplace: each indi- vidual media fragment can be uniquely addressed, re-used, and refer- enced. Automatic metadata extraction, with various techniques over multiple content types, becomes the norm. • Upcoming tech: domain-specifi c ontologies, folksonomies, decomposable content, metadata extraction Risks: compromised privacy and bogus metadata will turn out to be the largest initial concerns. Automated metadata extraction from content will improve, but will still guess wrong in cases of more diffi cult content. Non-compatible metadata formats will create a need for