Just thinking out loud....
If I had any lingering doubts about the future possibility of automatic metadata extraction, they were banished last week watching Watson, the IBM supercomputer, playing Jeopardy (and beating the humans). So, the computational processing of natural language is one thing that's looking more promising.
Another issue is the future of embedded metadata in ebooks (the EPUB format already has embedded metadata as one of its features). What happens to cataloging when the majority of the library's books (and other resources) already comes with metadata?
So, I think it's fair to ask the question: Will we need to catalog ebooks (and other digital resources) in the future? These are not new ideas. Some of the experts in our field have already been working on them (for example, the Final Report for the AMeGA (Automatic Metadaa Generation Applications) Project [pdf]. But I have to say, they are inching every closer to being more a reality and less a pipe dream.
Vanessa Harris asks a similar question in her recent post, Challenging Metadata Surrogacy processes:
Consider if we could start from scratch what would we do? How can this be a reality without chaos ensuing? If xml and rdf instead of MARC alone allows for greater cross-searchable data, we could consider the use of interoperable automated applications. It is time to challenge traditional processes, although to be fair it has been the time to challenge the legacy of our library cataloguing forefathers since the original concept of the world wide web in 1989 by Tim Berners-Lee.