Bursting the Bubble part 2: Artificial Assistance
In “Bursting the Bubble part 1: Worlds of Illusion”, I outlined a number of ways in which what we take to be objective reality is an artificial construct. More correctly, we are living in a hierarchy of such constructs. The most fundamental level, our direct perception of the world, our experience of “now”, is, so far as we can tell, quite accurate within its limitations, and we have systems for correcting it.
It is the higher levels of remoteness, meaning and value, where these constructs begin to really diverge, leading me to say in part 1 that “we are not living in the same world”. Our location, social status, age, nationality, political outlook, and racial, ethnic, sexual and gender identities mean that our answers to “What significant things happened in the world today?” can vary hugely, as can our interpretations of specific events that we are mutually aware of.
Three well-known and related mechanisms contribute to this tendency towards different understandings and perceptions of the world and our construction of the social “multiverse”. They are the “telephone game”, “echo chambers” and “filter bubbles”. In telephone, a message is passed from person to person, subtly changing as it travels along the chain until the final version is radically different from the original. In an echo chamber, a relatively homogeneous group of people repeat the same or similar stories to each other, acting to reinforce each other’s beliefs. Both of these phenomena are as old as society. The filter bubble is a newer, similar phenomenon driven by technology, whereby automated search and recommendation services, by giving each user results tailored to their previous actions and choices, tend to isolate them into insular groups, reinforcing existing assumptions, biases and opinions.
These phenomena combine with an interesting and related aspect of human memory to help us build and inhabit different “worlds”. Neuroscience tells us that when we remember a past event, we recall it, interpret it in our current context and write it back into our memory (see this PMC article, for instance). This means that our memory of the past changes as we come to understand the world “better” or at least differently.
As our communications and perceptual horizons expand to global levels, and we come to live more within the different worlds of our bubbles, society becomes more fragmented and polarized. For me, this raises the question of whether technology can be used to reduce the isolation, to burst the bubble, rather than reinforce it.
I suppose that this question begs another, “Why should we even try to burst the bubbles?” If we are getting what we want, getting answers that are in keeping with our expectations, past behavior, and interests, isn’t that a good thing? My answer to that is, yes, it is good when Google can infer what I am a really looking for with my query and not just mindlessly match a few words, and yes, it is good when Amazon or Netflix can say that given the books, movies and TV shows I’ve enjoyed in the past, here are some more that I will like.
Still, an unvaried diet wears thin, after a time. Growing up, each year, for our birthdays, my father and I had pretty much the same celebratory dinners: lobster, artichokes, corn, mashed potatoes and strawberry shortcake. It was a special treat, but if every day, 365 days a year, dinner was the same old lobster and artichokes, etc., it would have warn pretty thin. I’d have started pointing out the old Massachusetts law that you can’t force your servants to eat lobster more than three times a week. Variety is important. A human who is really good at recommendations will not only recommend things that are just like everything else we’ve liked, but will pepper their recommendations with things that are unexpected.
Perhaps more importantly than that, though, is the social impact of people becoming too isolated, having too little in common with others. If all my entertainment is made up of science fiction, fantasy and comic books, and someone else’s is nothing but sports, we begin to have nothing in common, but if we learn that we both love Breaking Bad, The Good Wife, The Americans or Homeland, or that we love Disneyworld, hiking in the mountains, or Thai food, then we can begin to understand each other’s worlds.
Today we find that our politics, our culture, our understanding of current events divide us, break us into tribes. Our choice of computer or smartphone OSes or cable news channels separate us. Major politicians who have the support of around half the people are regarded by the other half as totally untrustworthy and pathological liars, and you can quickly pick counterbalancing examples. It is important that we not all agree 100%, but it is just as important that even when we disagree on some matters, we have enough in common that we have a basis for conversation and understanding.
Returning to the question of whether technology can help to burst the bubbles, the answer, it seems to me is, “yes”. Just as human teachers can help us explore new ideas, autonomous virtual assistants and AI search optimizers and recommendation engines can be constructed to augment the narrowly filtered results with appropriate additional information. Take the following approach, for example:
Suppose that an AI-driven search filter could not only recognize my interests, preferences, habits and tastes, but could identify groups and clusters of users according to the similarity along these dimensions. Rather than just limiting the horizon of my bubble, it could be aware of the many different bubbles there are among users. It could also measure the conceptual “distance” between various bubbles, identifying those that are most similar and those that are quite different.
Equipped with this information, my search results could be chosen with an eye towards not merely those results that most closely resemble my past choices, interests and behavior, but augmented with a few items meant to expand the horizon. For instance, suppose that each page of my search results included 20 items. The first 16 might be the ones that I am most likely to be interested in according to past behavior. Three more might be chosen because they rank very highly according to the tastes of users in nearby bubbles. One could be chosen because it is strong in a bubble that is quite distant from mine.
This could be further enhanced by having the system keep track of which items outside a given bubble are chosen when offered. This could lead to the system discovering which items are most likely to be able to attract users unexpectedly. A system that can recommend something unexpected that I might well not have found on my own becomes a delighter to me as a customer, more so than something that recommends the same old things over and over.
The exact mix of in-bubble, peripheral, and long stretches would want to be balanced experimentally, but a strategy of occasionally offering unexpected choices that expand our limits is more interesting and potentially delighting to the user, makes the filtering system more engaging and helps to broach barriers that act to separate us socially and culturally.
