Earlier this year, Eli Pariser asked two friends to type “Egypt” into Google and show him the first page of results. One of them got articles on the Egyptian protests and the Arab Spring; the other got travel guides with no mention at all of the political unrest in the country. That’s just one example from Eli’s recent lecture at the London School of Economics on the “Filter Bubble“.
Facebook, Google and many other websites are engaged in a race for relevance. Analysing records of your past online behaviour their algorithms attempt to predict what you mean and what your intent actually is. This is used to give you personalised content based on what that code thinks is most relevant to you right now. Clearly useful, when, for example, I want to find the speaker’s slides from a session I attended last week. I want the single right answer, even if I search generically on “social media seminar”. However, this degree of personal relevance might well prevent me from seeing the other search results I actually need, such as discussion from an earlier session I missed or the details for the next one in series. Goodbye discovery, serendipity and synchonicity.
Remember all that small print on investment products that says “past results are not an indicator of future performance”. The same is true for past learning needs. The very things that make social learning using social media so valuable also make it highly susceptable to the Filter Bubble.
Of course, bias is everywhere. If you tune into Fox News or read The Guardian, the bias of Glenn Beck or Polly Toynbee are absolutely “in your face”. You are also making a conscious choice, are probably well aware of the editorial line and everyone watching/reading gets the same thing. Online personalisation uses a membrane of invisible filters you are not aware of. The bias is passive, you don’t know the criteria for selecting what’s in or out and the view you get is unique to you.
All three of key problems Eli cites for the Filter Bubble apply to social learning:
- Distortion: you see more of what is familiar to you and your close/strong tie network. Stuff you agree with, very little that is challenging to your point of view.
- Psychological Obesity: Recommendations only from “people like you” based on immediate relevance are the equivalent of a junk food diet for the mind.
- Control: Personalised search is making choices for you and you have little or no control over it. Kranzberg’s first law of technology says: Technology is neither good nor bad; nor is it neutral. The personalisation code cannot be pure or neutral – who does it work for?
Looking at some specific examples and potential remedies:
- Like button: The Like button is a key mechanism for propagating content in Facebook or Yammer, etc. “Like” is a postive affirmation easily applied to positive news and comfortable views. So that’s the stuff that moves quickly and forms the “common” context in your network. It is much harder to “like” a complex or challenging article or point of view on, say, trusting staff to use open access to social media wisely. We aspire to open and honest online discussion, but unless the organisational culture really supports it, we’ll not discuss uncomfortable topics for fear of making a “career limiting move”. So the safe options make it to the top 10 and get “liked” re-inforcing them as the concensus.
- Remedy: Eli suggests moving away from relevance and “like” as the sole rating and recommendation indicators. How about adding other options such as “important”, “uncomfortable”, “challenging”,”other point of view”. If we’re stuck with Like, then also add a comment on why you like it (because it reminded me of…, got me thinking about…, was well argued even if I don’t agree)
- Remedy: Reward and recognise people who consistently work to link the tribes in your organisation (and externally, especially with stakeholders)
- Personal Learning Network: A key reason for corporate social network is to help teams share and learn together but the dangers of re-inforcing silos and supporting group-think are clear if we only follow each other and the usual suspects in our area.
- Remedy: Emphasize Harold Rheingold’s notion of what “Explore” means for a PLN: “it’s not just about knowing how to find experts, co-learners, but about exploration as invitation to serendipitous encounter
- Access and moderation: Social media tools are feature rich, yet often in the corporate environment many of these features are overly restricted. An organisation I worked with introduced a community-based collaboration tool which included blogs. Unfortunately, permission to blog on a number of groups was restricted to people who “were well-known” in that discipline and comments were only displayed after moderation. There were no clear criteria for what “well-known” actually meant. Blogging as a way to build peer recognition was denied.
- Remedy: Trust your coporate communities to self-moderate and open the full feature set to everyone
Social learning uses the very tools that generate Filter Bubbles. You can’t turn it off completely (and you probably don’t want to) but you can dial down its bad side effects. Simply remembering that the map is not the territory is a good start. Otherwise the filters will gradually exclude more and more until instead of easy access to a web of everyone, everything, connected, you have a web of one. Yourself, multiply reflected.

