How big is your filter bubble? What’s in it? What’s outside it?

Okay, next question: how can you tell?

In 2011, Eli Pariser defined a filter bubble as “that personal ecosystem of information that’s been catered by algorithms.” Unfortunately, he never defined a measure on filter bubbles. Like its name implies, a measure is a way to tell how big something is. Without one, we can talk qualitatively about filter bubbles, but if we want to get quantitative, we’re going to need to do some more work. We’ll start with a literature search.

Let’s assume that the continuum of human experience is a continuum in the mathematical sense. (Whether this is actually true has some interesting philosophical implications. If it is instead discrete, the analysis below is overkill, but still works.)

In the last few years, researchers have studied the frequency with which Facebook and web users, in aggregate, encounter news and opinions from a different ideological perspective. This is interesting, but doesn’t tell us anything about the size or interior of an individual user’s bubble. Dillahunt et al. 2015 defines a distance metric between search engine users based on the overlap between their personalized search results, but a metric only tells you how far apart two things are. If you want to know how much space a bubble takes up, you need a measure.

Nguyen et al. 2014 creates a similarity metric for movies from the Euclidean distance between vectors of descriptive tags, one vector per movie. The authors use this metric to evaluate a system that recommends movies based on titles a user has already rated. They take the top 15 recommendations for a user, compute their mean pairwise distance, and call that the content diversity of those recommendations. This still only tells us something analogous to the diameter of a filter bubble, though, and then only if you tilt your head and squint. However, because the set of all content that has ever been or ever will be created (and, thus, the set of topics that content has ever been or ever will be about) is finite (if unbounded), trivially a filter bubble in the real world is also a measure space. It’s not immediately clear to me whether there are any measures that tell us anything interesting about a filter bubble’s volume, but the Dirac measure δ_x(A) answers the question “is x inside or outside A’s filter bubble?” with 1 for “inside” and 0 for “outside.”

Facebook’s study found less of a difference between which news liberals see and which news conservatives see than Pariser expected. However, the researchers only looked at the liberal/conservative valence of articles, which they determined based on how the people who shared an article self-identified politically. They did not examine or identify the topics of articles. There’s been some work on topic identification in group filter bubbles, though, which still makes sense with our notion of filter bubbles as measure spaces. Here’s how.

Let T be all the topics a group discusses. The power set of T is all the possible subsets of T, so elements in the power set are also sets of topics. If each member of a group is only interested in some subset of the topics the overall group talks about, that means we can map each member of the group onto subsets of T. The resulting collection of subsets is a  σ-algebra, and along with T, it defines a measurable space: the topics in each member’s individual filter bubble constitute a measurable set which is an element of the power set of the overall set of topics. With our Dirac measure, the measurable space is also a measure space. (Mathematics nomenclature: not fucking helping since sometime before the 19th century.)

However, I have yet to see anyone attempt to empirically characterize topic filter bubbles outside of small political niches. What does it look like when, instead of along some increasingly meaningless two-party axis, you characterize groups by the topics they cluster around?

Well, it probably looks a little something like a continent and a bunch of islands:

Culture, in the part of the world in which I’ve been, and, for all I know, in other parts as well to which I cannot speak, has two rough parts: the Mainland and the Isles.

The Mainland is what calls itself the “mainstream” or “normal” culture.

You know… Mundania.

The Isles are everything else. Everything that’s not “mainstream” is an island.

Nobody knows how many Isles there are. They are wholly and utterly unmapped. Each one is its own subculture.

Some Isles are closer to the Mainland, and some further.

Some Isles are closer to others. Some are big. Some are small.

We — meaning I and a very large percentage of my readership — live in a collection of close Isles which form up an Archipelago. The SCA. Fandom. NERO. Etc.

This is the Archipelago of Weird.

No points for guessing I’m an Islander, obviously. I work on the Mainland, and I have acquaintances there, but I do my research and pretty much everything else that matters in the Isles. The Mainland can be a nice place to visit, but I have to speak Mainlander there, because they don’t understand my jargon. This has its annoying aspects, but also a tradeoff that’s sometimes useful: Mainlanders know nothing whatsoever about Island customs, history, or aspirations, nor do they care to. (“I work in computers.” “Oh! That must be interesting.”) The Isles can be fractious places, and from time to time it’s nice to evacuate away from the latest hostilities. Intra-Isle conflicts are the most exhausting, because they split a population into two or more groups, each with its own native shibboleths and taboos. Keeping track of norm proliferation takes a lot of attention.

Mainlanders, generally unaware that the Isles even exist, know nothing of Island norms either. And Mainlanders get really weirded out when Islanders try to explain them. Mainlanders don‘t want to understand Islander inside baseball. Hell, most Islanders don’t want to understand inside baseball from other islands, unless they’re like me and have a summer home on the Isle of Inside Baseball Dissectors. A Mainlander and someone from the Isle of Fen might both listen politely to an explanation of why the Burr-Feinstein encryption bill is a terrible idea, but neither one has any reason to be particularly interested, so if there’s something I need to get across to them, I’d better be entertaining first and informative second. Burr-Feinstein is inside baseball for someone from the Isle of Infosec, and technically so is the Wassenaar Arrangement, but international trade agreements are Isle of Policy inside baseball. δ_Wassenaar(A) is almost certainly 0 for any Mainlander A, but it could be 1 or 0 even on my native island.

Being able to estimate δ_x(A) for many different topics (x) and people (A) is an incredibly useful skill for anyone who does much Isle-hopping, or anyone who travels between the Isles and the Mainland with any frequency. Being bad at estimating δ_x(A) is a recipe for awkward interactions at best, and interpersonal conflict at worst. Mainlanders who can’t estimate Dirac measures for Islanders can still usually get by if they find themselves on an Isle by mistake, depending on how amenable the local Islanders are to humoring foreigners. (Do not expect to be humored on the Isle of Outlaw Bikers.) Islanders who can’t estimate Dirac measures for Mainlanders are kind of screwed on the Mainland, especially if they don’t realize how far away the Mainland filter bubble is from their own. Double especially if they expect the Mainland filter bubble to overlap one-to-one, or even significantly, with their own.

This is something I have to keep firmly in mind every single time an allistic person comes to me with well-meaning questions about autism.

In my first post on this blog, I criticized a piece Sumana Harihareswara wrote about eliminating “inessential weirdnesses” in technology because of its inherently ableist stance on inclusion. Proclaiming command line interfaces and non-face-to-face interaction to be of a universally “unwelcoming” nature erases the subset of people for whom those affordances make a space more welcoming, not less, and many of those people are on the autism spectrum. Harihareswara reached out to me and linked me to a revision of the piece that she gave as a talk at LibrePlanet, hoping to “do better in discussion of ableism and competing access needs.” Encouraged by the lure of dialogue, I opened the transcript and started reading …

… only to discover that Inessential Weirdness Number One, LibrePlanet Edition was still “the command line.” The new Number Two was “not using small talk.”

From the back, off the floor, through the hoop, nothing but headdesk.

In the OSCON version of the same talk, the “inessential weirdnesses” she digs into are contempt (which I can get behind, though frankly, contempt is a communication antipattern in all communities; tech isn’t special here), lack of consideration for religious observances (ditto) … and the command line. There’s also a laundry list of Inessential Weirdnesses Not Appearing In This Talk, and small talk is still one of them.

Look, I’ll be the first to admit that the Archipelago of the Spectrum is a place most people just don’t visit. Particularly people from the allistic Mainland, unless they work in special education or have a relative who lives there. But when someone claims a position of moral authority on how to be inclusive of people with wildly varying levels of ability — and telling people how they should or shouldn’t behave is inherently claiming a position of moral authority — it is ever so slightly incumbent on that person to do at least the equivalent of visiting Wikitravel beforehand. Here’s what you get when you Google “autism ‘small talk’.” With 138,000 results focused on how much of a challenge small talk presents for autistic people, it’s reasonable to assume there’s at least some there there.

When a dialogue reaches the point of “how could you possibly not know this,” I find it useful to apply Miller’s law:

In order to understand what another person is saying, you must assume that it is true and try to find out what it could be true of.

Harihareswara states her beliefs that not using small talk is “across-the-board inhospitable” and “off-putting to working-class people and women,” and that “some people get socialized” to “perform solo, freestanding utterances” while others don’t. I’m willing to assume that she believes these things. But what would have to be true in order for someone to form such beliefs?

Here is an incomplete list of possibilities, from least to most charitable:

• She believes that the preferences of allistic people are more important than those of autistic people, and that autistic people should subordinate their needs to the preferences of the allistic majority.
• She’s never met anyone for whom an expectation of small talk is itself inhospitable.
• She’s never met anyone for whom freestanding explanations are the most natural way to communicate.
• No one she’s met has been willing to admit having such uncanny-valley traits to her.
• The possibility that there are actual people for whom her beliefs about human behavior do not hold did not occur to her.

I have to reject the first hypothesis, because a genuine allistic-supremacist wouldn’t have asked for feedback in the first place. Number four gets a bonus to its likelihood estimation because of the prevalence of the attitude that “indistinguishable from neurotypical is the same as cured” (protip: it isn’t), the stigma that people who are open about their mental health issues face, and how severely underdiagnosed autism is in women. (Here’s a partial taxonomy: white girls are highly sensitive people, white boys are autistic, and black boys have oppositional defiant disorder.)

I don’t have any further data to use in evaluating the likelihood of the other possibilities. One way or another, though, Miller’s Law leads me to conclude that δ_autism(Sumana) = 0. Whether it’s because she hasn’t gone looking for information or because no one has told her about it, the details of autistic lived experience are outside her filter bubble.

Thus, I went to the Isles for counsel. There’s a little place I like to hang out where all the regulars are autistic women, and all of them except me are feminists. I linked them the transcript of the LibrePlanet talk, and here’s what one of them — who also has some physical disabilities, and from whom I hope to have a guest post soon — had to say.

Look, the tech community is already not very inclusive of me.

It’s not very welcoming to women.

It’s not very welcoming to physically disabled people.

It’s not very welcoming to people with dyscalculia that affects their perception of time.

It’s not very welcoming to people who work best by focusing on one thing to completion or exhaustion instead of interleaving several tasks that must be mentally unpacked, expanded, reabsorbed in both breadth and depth of detail, then switched away from and packed back up again. The type of thinking that gives me wide-ranging expertise — jack of all trades, master of none, better than master of one — makes me nearly unemployable despite my skills and extensive crosstraining.

It’s not very welcoming to people who prefer both breadth and depth of detail instead of only one or the other, in general, come to that.

It’s not very welcoming to people who have suffered short-term memory loss and require, require, a single, assigned, constantly-messy space in which to work by viewing frequent reminders of work-in-progress as it shifts to long-term.

IT IS STILL MORE WELCOMING TO ME THAN MOST PROFESSIONS.

And there are areas where these things are close-to-least true of all the fields I am suited for: I do not have the emotional-labor aptitude for most traditionally-feminine work, I do not have the physical capacity to walk from office to office or lift 30 pounds or sit in the chair assigned to me for hours or perform repetitive arm/hand movements reliably, I do not have the ability to clock in or out on time, I do not have the ability to switch tasks flexibly during the day without excessive mental fatigue and loss of productivity, and I do not have the ability to leave a question unanswered without emotional fatigue and a need for frequent breaks, stimming, or other self-soothing behaviors.

There are areas of tech where these are serious barriers to employment.

There are areas of tech where these things matter less than whether I can pick up a problem, research it in detail, keep it all in my head, keep myself comfortable, and produce a result in shorter order than most people could.

If we’re going to talk about making tech more inclusive, we could talk about physical disability inclusion. We could talk about chronic illness inclusion. We could talk about cognitive disability inclusion. We could talk about neurodiversity inclusion in general.

We could even talk about the self-satisfied but false proclamation that tech is already a meritocracy where none of these things matter so long as I produce a result in shorter order (or of better quality, or both) than most people could, a proclamation that makes any of the other problems more difficult to solve by insisting that they cannot possibly exist, because we have always been a meritocracy (and at war with Eastasia) that would not tolerate such barriers and any such complaints therefore must come from special-snowflake syndrome harbored by incompetents.

We could also talk about the stuff the tech industry does/is/accommodates that still leaves it the closest thing I can have to self-sufficiency and a feeling of personal accomplishment and general utility, and the people who have been not just willing but eager to steer me into the small tech-harbors I’m most compatible with because they get it.

But hey, no, sure, allistic person, let’s talk about this one place you feel locked out of and how we can make it even better for the majority, who already run so many other industries to the exclusion of people like me, first. Let’s make sure the already-privileged majority is comfortable in all places, at all times, before appreciating small pockets of minority safety and accommodation, and asking what they used to do right before they, too, were colonized by the tyranny of the narrowly-defined “default” human being in need of additional comfort while I try to survive. THAT FEELS FUCKING INCLUSIVE TO ME, HELL YEAH.

The tech community is a place where many spectrum-dwellers feel welcome because people don’t expect us to engage in small talk or other social rituals that privilege the desire for a perception of harmony of experience over information sharing. Not everyone whose behavior supports anti-small-talk norms is autistic. In fact, I expect most people whose behavior upholds those norms are not merely not autistic, but have no idea that they’re making the community more welcoming for us. (Thanks for that, by the way, to everyone who does it, whether you knew that was what you were doing or not.)

In other words, anti-small-talk norms are a Chesterton’s fence. Expecting autistic people to get better at small talk in order to make allistics feel more welcome is like expecting people in wheelchairs to get better at walking in order to make physically abled people feel more welcome. It puts all the onus for change on the marginalized population rather than the marginalizing one, and it drives away people for whom the expectation is too much to ask. The fence is there because it protects marginalized people whose presence the community values, and people seeking to tear down the fence have not taken the time to follow Wikipedia’s editorial recommendations for handling that urge:

If you’re considering nominating something for deletion because it doesn’t appear to have any use or purpose, research its history first. You may find out why it was created, and perhaps understand that it still serves a purpose. Or if you do feel the issue it addressed is no longer valid, frame your argument for deletion in such a way that acknowledges that.

Deleting the norm of “small talk isn’t necessary here” and replacing it with a norm of “you must engage in small talk in order to make newcomers feel welcome” excludes those whom the original norm exists to include. For them, the issue the old norm addresses will remain valid no matter how many allistic women get tech jobs. Wouldn’t the truly inclusive solution be to give allistic newcomers more insight about autistic modes of interaction compared to allistic ones, explain how the tech community has evolved over time to include autistic communication styles, and encourage them to set aside their bias against not being small-talked to?

I’d think it would, at least. But this is a lot harder when people don’t even realize they have biases they need to examine, and worse still when the same people consider themselves to be the rightful arbiters of which biases are even worth examining.

As I mentioned earlier, Pariser defined a filter bubble as a personal ecosystem of information catered by algorithms. What this definition obscures, however, is that algorithm is nothing more than a fancy term for process, derived from the name of a 9th-century Persian mathematician. In every single one of those handwringing articles you see about “Are Algorithms Running Our Lives?”, you can safely replace “algorithm” with “process.” Do processes run our lives? Consider how many processes you ran through today on your way to taking out your phone or settling in at your computer, and you tell me. Taking a shower is a process. Making coffee is a process. Riding the bus and driving a car are processes. For that matter, so are the interactions you have with other people, whether you recognize those interactions as processes or not. Other people curate the information that they present to you just as you curate the information you present to them. The only novel purpose that “algorithms” in the handwringing-article sense serve is to remove the constraint of physical distance from the problem of who can curate information for whom. Whether online or in meatspace, there is still some process that filters what information you receive. The only salient difference is the extent to which you can control that process.

“But don’t Facebook’s and Twitter’s algorithms limit what information I see?” Yes, and so do the choices you make in friends. The fact that your friends cater the information that ends up in your filter bubble means that your choices in who to listen to determine whether you’re a Mainlander or an Archipelagian, which inside baseball means something to you and which doesn’t. If your filter bubble contains no outliers, where do you expect to learn that outliers exist, much less what their lives are like? If your goal is to make existing spaces more welcoming to the mainstream, what effect do you think that has on outliers? Especially when only the mainstream gets a say? If your goal is, instead, inclusivity of both mainstream and outlier populations, what actions do you think you could take to learn more about outliers and the Chesterton’s fences they rely on?

I’m leaving those questions open-ended rather than answering them because, frankly, by the norms that Harihareswara claims to support, it’s not my job to answer them. They are, however, absolutely not rhetorical. Mainstream feminism has some serious catching-up to do when it comes to learning about the lives of people who aren’t nice normal middle- to upper-class ladies, not to mention a lot of earned distrust. When you tell people that a skill to which they are inherently maladapted is a new requirement for participating in some culture, you are telling those people that they are no longer welcome in that culture. Bluntly, that is not your decision to make, and people are right not to trust the motivations of anyone who behaves as if they think it is. Too many of us have been burned too many times by people who told us “we want to make this a great place for everyone!”, only to find out that in practice, “everyone” actually means “all the allistics.” If you really do intend to be inclusive, then it’s on you to find ways to discover what marginalizations people experience — including the ones they’ve been socialized into believing it’s shameful to tell you about — and what accommodations they need, because, as the saying goes, intent isn’t magic.

Nor is it lost on me that I am sitting here patiently spergsplaining theory of mind to people who supposedly have it when I supposedly don’t. Allistics can get away with developing a theory of one mind — their own — because they can expect most of the people they interact with to have knowledge, perspectives, and a sensorium not all that different from theirs. Autists don’t get that option. Reaching adulthood, for us, means first learning how to function through a distorted sensorium, then learning to develop a theory of minds, plural, starting with ones different from our own. (Even when it comes to other autists. As the saying goes, if you’ve met one autist, you’ve met one autist.) Developing good Dirac measures for other people’s filter bubbles is an autistic survival skill that allistics can go their entire lives without realizing other people have to acquire.

Except you, dear allistic reader. δ_{pro-small-talk norms marginalize autists}(you) = 1, now. What will you do with your newly expanded filter bubble? Think carefully, because your credibility now hinges on it. Granted, that’s credibility among autists and among people who expect others to behave in a manner consistent with the values those others claim to hold, which I’m starting to think may be the exact same set. If you want to prove me wrong, you can start by taking up Miller’s Law yourself: what are the things I’ve explained true of?

Finding an answer to that question is going to require a lot of traveling to unfamiliar islands where the natives may be hostile specifically to people with your neurotype. Understanding why they’re hostile, and why what’s already going right for them is going right, is going to require updating your theory of minds. If you’re not prepared to do that, it won’t be a pleasant trip. If you are, however, then please enjoy your stay in the Archipelago.