Human-assisted search gets flushed: how would you do it right?


News: Human-assisted search gets flushed: how would you do it right?

  1. According to a link from TechCrunch, ChaCha search has announced that they're ditching their assisted-search model, where a human guides the search. ChaCha was quite popular when announced, but their integration points apparently didn't work.
    The idea is that you do a search on ChaCha and a real person works with you via a chat interface to give you results. In theory those results would be better than Google. In reality, they weren’t [snip!], and ChaCha still had to pay all those guides. Today, according to an email sent to ChaCha's guides titled "The Future Is Here," they announced that guided search will be discontinued in favor of the one product they offer that isn’t monumentally dumb - mobile search. They claim that "new users are growing at a staggering rate every day" (most likely due to cell phone spamming). So what happens to all the guides who worked on desktop search? Some of them, at least, can now apply for new positions on the mobile product.
    Mechanical Turks aren't new, but ChaCha built their business model around it. That said, guided searches were only ten percent of the interactions - and some, including a snippet included with the article, are hilarious. However, a company that happens to be about ten doors down from ChaCha Search - Interactions uses something similar to a Mechanical Turk - humans managing small parts of the interactions, instead of the entire interaction. The results are amazing, and successful. An example offered by the CTO at TheServerSide Java Symposium included a pizza ordered by a drunk, irate customer over the phone. The call started with the drunkard at a party, calling an automated system; the party was loud, with one person repeating everything the caller said to the rest of the party, with its attendant background noise. The system walked him through getting his address and his name, sounding quite natural for an automated system - and the customer, apparently familiar with automated voice systems, was rather irate, even though it got his information correct ("Your name is spelled B - E - N, right?" is an example of the automated nature of the call, where a human might just say "Ben.") At one point, he even called the system a foul name. However, as he worked with the system, he eventually actually asked it if it was in fact automated - at the same time asking for two drinks. The system responded that it was, in fact, an automated system and what drinks would he like? If you think about it, that last interaction is amazing - he asked two questions at once, and the system answered them both; in addition, he knew it was an automated system, but was actually questioning his own knowledge due to the system's interactivity. Now, that's service. I've seen the system in action, and it's incredibly workable and quite impressive (which is part of why the CTO was invited to TSSJS) - but I'd like to point two things out here: ChaCha Search is apparently unsuccessful with the Mechanical Turk approach; Interactions is successful with it. How would you approach leveraging human intelligence in an enterprise application?
  2. To be accurate, ChaCha still does human guided search, they just lost the web chat aspect. Given the competition in the desktop space, I think it just made more sense to move into the mobile instead. Especially since with SMS you only want one result - the answer to your question. You don't want to page through a list of results (like you do with Yahoo!'s mobile SMS service). Basically a voice or SMS query is put into the system, a human guide gets it, they research the query then provide a single answer (with a corresponding reference). You get the single answer back via SMS, then you can browse your answer history online or check the references cited by the human guide.