Ideal Recommender Systems are dead, long live Surprise-Me Buttons

[Sometimes I paraphrese bits and pieces of this book's intro] How perfection killed recommender systems.


The market for news, entertainment, and information has been finally perfected. In 1995, Nicholas Negroponte prophesied the emergence of the Daily Me – a communication package that is personally designed, with each component fully chosen in advance. If recommender systems know a little about you, they can tell you what “people like you” tend to like – and they can create a Daily Me, just for you, in a matter of seconds.

Perfect! Not really. There are serious dangers in the Daily Me, serious problems to democracy. In his new book “Republic.com 2.0″, Cass Sunstein shows why.

The dark side of Social Computing
Sunstein says that a well-functioning system of free expression must meet two distinctive requirements:

1) People should be exposed to materials they wouldnot have chosen in advance.

2) Most citizens should have a range of common experiences.

If those two requirements aren’t satisfied, there are risks of
fragmentation and, more worryingly, of extremism. Recommender
systems allow their users to bypass general-purpose intermediaries
and restrict themselves to opinions and topics of their own choosing. If recommender systems were to be perfect, millions of people would listen to louder echoes of their own voices, they would find it hard to understand one another, and they would engage into extremism, which is produced by any situation in which like-minded people speak only with themselves.

How to satisfy those requirements? It’s important to maintain the
equivalent of “street corners” or “commons” where people are exposed to things quite involuntarily. There are many apparent “street
corners” in the Internet, but they are highly specialized, limited
to particular views. Lives should be structured so that people 1)
come across views and topics that they have not specifically
selected; and 2) share common experiences upon which to form a
social glue.

The case for Surprise-Me Buttons
Does all this call for a counter research agenda in social computing? One in which imperfect recommender systems (perhaps with surprise-me buttons) are more than welcome?

5 Responses to “Ideal Recommender Systems are dead, long live Surprise-Me Buttons”

  1. Neal Lathia says:

    Perhaps I wouldn’t call a surprise me button.. that seems to imply some kind of serendipitous discovery, but keeping a ‘street corner’ doesn’t seem to be quite the same thing, no?

  2. By contrast, street corners are gonna end up to be specialized (to be niches). My point is that serendipity should be built into rec systems in one way (surprise-me buttons) or another

  3. Another way to go about it: At CMU, they ” develop a computer system that can automatically identify highly biased television news and encourages audience to seek news stories from contrasting viewpoints.”
    http://lastlaugh.inf.cs.cmu.edu/alex/lin08vp.pdf

  4. [...] algorithms may be not good for us (previous post). Few points from the Guardian [...]

  5. [...] for books or DVDs may encounter similar problems. I’ve been discussing the possibility of a “surprise me” button, proposed by Daniele Quercia. The idea is that sometimes as a user I would like to be suggested [...]