Learning networks are about social search

      Filed under: Articles   

Damon Horowitz and Sepandar Kamvar recently published a paper – referentially entitled “The Anatomy of a Large-Scale Social Search Engine” (PDF) – in which they nicely lay out the social search problem and the Aardvark solution. As I was reading this paper, one thought kept surfacing: Learning networks are about social search.

It seems reasonable to claim that since Q&A is about learning, Aardvark’s network-based Q&A site can be described as a type of learning network. The less obvious but, in my opinion, more interesting claim is that the design of learning networks can (and should!) be viewed as a social search problem. Illich argued this, Meetup demonstrates it quite nicely, and Grockit is moving in that direction, too. In an email exchange published last year, I suggested that the future of AI in Education will have a lot to do with social search:

I think the AIED systems of the future will be less about teaching directly, and more about providing guidance: when and how a student would benefit from working with someone else (perhaps a teacher, tutor, or peer.) When I get stuck solving a particular type of problem, who (that’s online and available) can best help me understand it? A good system will have predicted the frustrating challenge, and will have already lined up the person best-suited to explaining it to me in a way that I will understand. After I’ve demonstrated that I mastered the necessary skills, who can I then explain it to, both to help them and to clarify it for myself? A good system will be able to seamlessly coordinate this process. Through these interactions, the system will unobtrusively be learning more about me – both as a learner and as a peer-tutor – in order to improve with time.

Downright Aardvarkian.



ari wrote on February 8, 2010
One other observation: Social search is human computation. While I'm having trouble pinning down the reference right now, I'm fairly certain that I've seen or heard Luis von Ahn describe human computation as a technique by which an AI-hard problem can be reduced to an HCI problem. After reading this paper, I think it's fair to say that Aardvark is an attempt at doing exactly that. The piece that I'm not understanding is the why: Why bother responding to a question? Most HC applications that I've read about seem to offer something compelling to the human in exchange for their computation: cash money (MTurk), fun (gwap), porn (pornornot), not getting your form submission rejected yet again (recaptcha). So what is the compelling motivator with Aardvark? Perhaps it's something more subtle: to show others what you know, to help a friend of a friend, or to voice an opinion (even if just to an audience of one). If you have thoughts on this, please share.

ari wrote on February 12, 2010
and... sold.

Ari wrote on September 10, 2011
and... shut down.