I'm also skeptical of the flexibility that news on demand can achieve. Any kind of filtering involves editorial choices, and the danger of burying these choices in software (such as in choices of search terms) is that end-users don't realize those choices were made. It worries me that one system reported in the Communications of the ACM searches for topic labels "drawn from a set of over 5,500 possible topics known to the system." Another depends on "extensive lists of persons, companies and locations, as well as contextual information." All those lists and heuristics depend on human choice -- and I wasn't the human who was consulted. Other systems, I'm glad to see, use dictionaries that provide a lot of breadth and relative neutrality. But I worry about whether such systems can scale up to covering thousands upon thousands of broadcasts viewed by billions of people.
I also think that contexts for terms are very hard to establish, so search and classification systems will have trouble figuring out whether a particular item is really want the user asked for. A service would be worse than useless if a search for "cross-country air travel" returned the movie Sleepless in Seattle. Come to think of it, any search would be worse than useless if it returned Sleepless in Seattle.
I'm disappointed that the Communications of ACM articles didn't even touch on any of the issues I've raised. The ACM has taken a leading role on social and political issues in computing, but this particular feature shows none of that influence. The articles focus narrowly and repetitiously on particular technical hurdles faced by the researchers on various projects.
And so I laugh when I see so many research systems struggling to classify news broadcasts through speech recognition. Speech recognition is a critical research area, but this particular application doesn't seem worth the effort. News broadcasts consist mostly of staff reading from written scripts. Spontaneous comments from people interviewed on the air make a critical contribution to the shows, but the topics and context can easily be gathered from the scripts. Haven't the people developing search services thought of asking news studios for their scripts?
More generally, I like the idea of being exposed to news and ideas I didn't think were important before. That's why I still read my local papers, not to mention listening to news shows on the radio and reading a lot of technical publications online and offline. One of the big advantages of a well-edited publication is that you learn things that somebody else thinks is important, not just things you already knew were important. Being an editor myself, I'm gratified to see that editors still play a key role in keeping people educated. I look forward to news on demand, but I hope everybody will recognize that it embodies the assumptions of its developers, just like any other technology, and that they use it to complement traditional publications rather than supplant them.
Join the discussion about this article in the O'Reilly Network General Forum.
Andy Oram, firstname.lastname@example.org, is an editor at O’Reilly & Associates and moderator of the Cyber Rights mailing list for Computer Professionals for Social Responsibility. This article represents his views only.