Google’s latest offering,
is rather fun, but I’m not convinced that I will use it very often.
Compare search results like this:
The page on Wikipedia is much more useful. It seems that humans are better at making tables of data from diverse sources of information that computers are at this point. Will it always be this way?
Wikipedia has strict guidelines on how articles are written and how propositions should be backed by reliable sources. Could these guidelines be further formalised and pave the way for an algorithm that could write something like Wikipedia from scratch? Google seem to be attempting to build a system that can produce the pages on Wikipedia with names like “List_of_*”. For all I know, Google might have looked at all the articles on Wikipedia whose names match that pattern and used them to get their tables started.
Sport is a popular subject. It’s safe to say that there are lot of people who are willing to give up their free time to collate data on the subject. If some joker changed the Wikipedia table to say that Manchester United were relegated at the end of the previous season, this error would be corrected quickly as there is no lack of people who care deeply about the matter.
During a presentation for Wolfram Alpha, Stephen Wolfram was asked whether he had taken data from Wikipedia. He denied it and said that the problem with Wikipedia was that one user might conscientiously add accurate data for 200 or so different chemical compounds in various articles. Over the course of a couple of years, ever single article would get edited by different groups. The data diverged. He argued that these sorts of projects needed a director, such as himself. However, he said that his team had used Wikipedia to find out what people were interested in. If the article on carbon dioxide is thousands of characters long, is edited five times a day, has an extensive talk page, is available in dozens of languages, and has 40 references, it is safe to say that carbon dioxide is a chemical compound that people are interested in. This is true regardless of the accuracy content of the article. It would be pretty trivial for Google (or any Perl hacker with a couple of hours to spare and a few gigs of hard disk space) to rank all of the pages on Wikipedia according to public interest using the criteria that I just listed.
In many ways, an algorithmic encyclopaedia is to be preferred because of the notorious problems of vandalism and bias. However, tasks like condensing and summarising are not straightforward. The problem of deciding what to write about could analysing Wikipedia, as described above, and tracking visitor trends. Is there going to be a move to unseat Wikipedia in the coming years? How long before humans can be removed from the algorithm completely?