Social media, media streaming services and gig economy platforms all use recommendation systems to make guesses about what we like by analyzing ours and others behavior. These systems can make abundance of options seem trivial and sometimes even so obvious that we don’t even think of them as choices. This is comfortable and saves us time, but will result in the loss of something else.
Recommendation Systems are forced to evolve in an environment that is our ever shrinking filter bubble. It can only learn from what it exposes us to which makes the early guesses very significant. I keeps trying to learn but all the future guesses will be contaminated by the error of all the previous guesses. The validity of the recommendations will thus decrease with every recommendation made. There are attempts to solve this issue. (Sinha, Gleich & Ramani, 2016) suggests creating a mathematical representation of the users true preference by subtracting the biased impact these so called feedback loops make. There are however a number of assumptions that need to be accepted for this to make sense. Also it does not address the validity problems that follow from the fact that some objects that are not liked by the first exposed users disappears from the systems radar. They will fall into an arbitrary viral shadow.
IDENTITY CONSTRUCTION AFTER THE FACT
Another problem is that we are risking to gradually make ourselves more similar to the prototype of us that the recommendation system constructs. To make ourselves make more sense we will feel a need to rationalize the choices we made. There is a concept in social psychology that is known as post decision bias, that in short describes our tendency to be more likely to find evidence that support that the decision we made was correct, than evidence of the contrary, after the decision has been made. We feel better when we can reassure ourselves that the choice we made was the only right one because it increases positive emotions like pride and decreases unpleasant feelings such as remorse. This means that we will look for reassurance in our external environment as well as changing or adding facts to the story about ourselves. Thoughts and statements like: I choose A and it was the right choice because I am the kind of person that likes A. So the recommendation system gives us what it thinks that we want, and at the same time we are drawn to becoming more like the person that wants just that. That way our human brains underlying mechanics is helping the system to appear more accurate than it actually is.
CONFIRMATION BIAS TURBO
Another related concept in social psychology is confirmation bias. The fact that we tend to be more attentive to signs around us that support what we already believe. Personalized search results and feeds enhances this effect by also changing our actual environment; subjecting it to a sort of mechanical confirmation bias, even before it reaches our cognition. That decreases even further our chances of discovering facts that contradict our beliefs.
EXISTING IMPERSONAL OPTIONS
One option for those who wish to make searches without getting personalized results is the search engine DuckDuckGo. There are also social media platforms that don’t personalize your feed or add customized advertising and instead just gives you chronologically sorted posts like the Instagram clone Pixelfed. I have listed more options on the page: [INDIE CHOICES].
1) FILTER UNBUBBLING
It would be interesting to develop a search engine, that instead of amplifying our confirmation bias would work against it, by using our preferences to select search results that will widen our views. A search engine that makes us surprised, curious, sometimes confused but always less simple minded. Or a recommendation system that gives you: This is what others, that are not at all like you, found interesting.
2) ANTIVIRAL MEDIA
In so called qualitative research the aim is to expose all the nuances of an investigative material, rather than to weigh these nuances fairly according to how common they are. We could build a social media platform based on that principle. All posts go through a classification process that creates categories based on similarity to other posts. The goal of the network is then to expose the users to posts that are as varied (heterogeneous) as possible. In concrete terms: As many different categories and as few from the same category. You can’t follow or be followed. You can like posts, in which case that poster gets notified, but it does not change your future feed. Posts will not, through sharing and feedback phenomena become viral. What you see when you look into the network is not what is trending or what is most commonly posted (cats and babies doing cute things), but a balanced overview of everything that is happening on the network.