On 1 October 2015, OCLC announced with a tweet the printing of the last catalogue card and informed us that the service of supplying catalogue cards to libraries – a total of about two billion – had lasted 44 years. It is just one of the many changes that have occurred in recent years in – to use an expression of Elaine Svenonius – ‘the organisation of information’.

From Cataloguing to Metadata Creation is a long journey between founding principles, objectives and technological evolutions of the ‘information organisation’ service offered by libraries. At the end of this journey, we can identify endeavours in three directions with more awareness.

The first one concerns human–machine co-operation in the creation of metadata. Nowadays, some experiences in the adoption of technologies related to artificial intelligence (AI, also called machine learning) for subject cataloguing (or semantic indexing) are still active. We should remember that artificial intelligence is an ‘algorithmic product completely lacking in flexibility’ (Floridi, 2014) and that ‘the effectiveness of an access system varies according to the intelligence invested in organizing information’ (Elaine Svenonius). In other words, in this context, the levels of skills required by a librarian are higher and higher and it is always his or her ‘intelligence’ that makes the difference. The expected benefits from these ongoing experiences essentially concern the chance to expand the coverage (given the significant increase in resources that require bibliographic control) and not to reduce the costs of cataloguing. Work in libraries – and especially cataloguing – cannot be replaced by robots or unqualified professionals.

The second direction concerns the methods of publishing metadata. We shouldn't forget a critical stage of the journey: MARC. With MARC, libraries were the first sector to understand that, thanks to the intermediation of information technology, the metadata produced by a library could be reused and enjoyed in other libraries (and not only in libraries). We could say that MARC is for bibliographic information what the Semantic Web (or web of data) is for the web tout court (or web of documents). As we have seen, this journey also takes into account the work of libraries on the Semantic Web and linked data. In summary, we could say that the entry of bibliographic data into the web of data will still require a lot of work (if the goal is to guarantee the same diffusion that MARC has now).