To counter personalization technology’s narrowing of our minds, let’s design for choice and serendipity.
Researchers at Google’s sister company, DeepMind Technologies, are using deep learning techniques to uncover how algorithms contribute to the problem of filter bubbles and echo chambers.
In a new paper, researchers analyzed how different recommendation algorithms can speed up or slow down both phenomena, which the researchers define separately. Echo chambers, they say, reinforce users’ interests through repeated exposure to similar content. Filter bubbles, by comparison, narrow the scope of content users are exposed to. Both are examples in academic-speak of “degenerate feedback loops.” A higher level of degeneracy, in this case, refers to a stronger filter bubble or echo chamber effect.DeepMind is asking how AI helped turn the internet into an echo chamber
Locally, Not Globally, Optimal
Personalization algorithms organize information in ways that optimize for
One of the problems that optimization algorithms run into is what’s known as a non-optimal, local maximum. To understand this challenge, think of optimization as a topographical map, where higher elevations represent more optimal solutions and tweaks to an algorithm are made based on how much they ‘move’ it to a higher altitude. The problem occurs when you start in a place on the map where the algorithm pushes you up a decent-sized “hill,” while prohibiting you from ever going ‘downhill’ to get to an even larger nearby mountain:
The solution for getting around the local maximum problem is tricking the algorithm into accepting some ‘temporary’ loss of optimization in order to get to an even better solution. Often, that entails throwing in a little randomization to your starting place on the map. In practice, what this means is that a personalization algorithm is going to be a little more hit and miss. As a result, you might get a “Hello Kitty” lunchbox thrown into your product recommendations from time to time.
In other words, to prevent filter bubbles and echo chambers, we need to be more open to the possibility of mistakes. We must more fully embrace serendipity if we are to open ourselves to the possibility of making desirable discoveries by accident. We need a little faith in luck and the possibility that by cranking serendipity up to an eleven, we create room for an optimization that is bigger than our conscious understandings of our preferences.
Efficiency and Exploration: A Choice
There are times when we need to prioritize efficiency, of course, just as there are times when we are more open to exploration and discovery. Today’s online services have their dials set to an eleven on efficiency because that is what companies believe is best for moving people through user interfaces at Internet scale. Even the designs for getting you to click and stay on Facebook are optimized for giving you exactly what the algorithms think you’ll want. There’s no dial for loosening that optimization and opening up the power of serendipity.
There are cases where companies open up windows for more serendipity:
These are laudable moves away from the tyranny of filter bubbles and echo chambers, but by themselves they are not enough. I want my search results page to have a slider that allows me to control the intensity of algorithmic optimization in real-time. With Twitter, I don’t want to have to adjust the level of algorithmic filtering through my settings as an all-or-nothing choice. I want to be able to open and close serendipity in real-time as I’m looking at my timeline, based on what kind of mood I’m in at that particular moment.
It’s not just a case of me being lazy or overly persnickety here. We need this kind of control over algorithmic strength to be a prominent feature in the user experience design if we want people to actually use it. Hiding an option for serendipity deep in the user settings is simply checking a box because the company feels it has to. But doing the hard work to make these choices feel simple and natural requires real effort in user experience design and, heaven forbid, even adding more customer support to deal with the added complexity.
Filter bubbles and echo chambers won’t magically disappear. They are naturally occurring phenomena that are a direct result of today’s user experience designs. The only way to curtail them is through changing those designs.