Charles Arthur On Technology: A misfire in the search engine

GOOGLE, THE name that has become a byword, even a verb, for "online search", launched a video search service last week. In its quest to "organise the world's information" (and make a bit of money), the company keeps coming up with new things. There was "news" (Google News, which still says it's a "beta" despite being available for three years). Then came "Gmail", a webmail service, also still beta. Now video: you'll find the service, inevitably a beta (meaning "you find the mistakes") at http://video.google.com.

As with most of Google's ideas, it's not the first to offer it. The start-up company Blinkx (www. blinkx.com) has offered video searching for some time, and also offers searches of TV programmes via www.blinkx.tv. AOL has SingingFish (www.singingfish.com), while Yahoo has a video-search beta at http://search.video.yahoo.com. AltaVista (www.altavista.com) and AllTheWeb (www.alltheweb.com) have video and audio search tools. Their advantage over Google is that they will search - as far as they can, which we'll discuss - multimedia files all over the web. Google's is restricted to a rather limited set of TV shows, and US ones at that. Gary Price of searchenginewatch.com says he's "underwhelmed".

The search is comparatively limited because it relies on the "closed captions" used in the TV broadcasts. These are a boon to the deaf for watching TV; and now they provide a useful adjunct for search. The entire spoken content - and any descriptions - can be indexed just like any flat web-page, allied to a timer the captions have to tell you what was said and when.

Do we want video search? You bet. It's obviously something you'd want to do. If a picture is worth a thousand words, video footage is worth a million - or at least takes up proportionally as much space. And some of the most enduring images of our time are in video; even terrorists have realised that the moving picture carries more power than the still one.

The problem with video searching is the same as with picture searching: how do you know what's in the footage you're looking for? If a human has watched it and laboriously put in keywords, you can search those keywords, using standard search methods (which are fast, on text). But you'll never be able to pull more information out than the person responsible for the item has catalogued.

Those programmes showing stars before they were famous are the result of meticulous tagging when the film was made, including the names of every performer. Lost film of David Bowie will stay lost if people don't think to search for old clips featuring "David Jones". And might there be some piano-playing footage of a young Reginald Dwight, before he became Elton John? But what if you want to find people who were in an audience? Their names usually won't be tagged. Or amateur footage, such as of the Asian tsunami: how do you catalogue that?

Even the top computer scientists throw up their hands at questions such as these. Getting computers to recognise objects as we do has eluded artificial intelligence researchers for 40 years. A few years ago, a company said it had a picture-searching technology that could spot pornographic images. This turned out to be related to the amount of flesh tones in the image, so works of art such as Le Dejeuner sur l'herbe would set it off.

No, video searching, while fun, is not at the point where we can say "find me some film of a white horse running up a hill against a blue sky". Unless, of course, someone tagged it with a caption saying that, or added text tags. Nice to know that humans still have some uses.

www.charlesarthur.com/blog