Adjustable rules
If you throw a rock against a wall, the consequences will occur according to the same laws of physics that govern what happens when an asteroid hits a planet. But if an email message contains a date, it may trigger a notice that automatically creates an event in your online calendar, it might ask if you want to add an item to a to-do list, and it might translate the event’s time into your local time zone. It all depends on the rules of interoperability that you have set up—and there is no predetermined limit on those rules.
In the interoperable world, we get to decide the rules of engagement. Even something as straightforward as copying information from one app into another can have different rules. The designers of Twitter made it easy to embed a tweet into another page. If you paste text copied from Amazon Kindle, it automatically includes the bibliographic information about the source, as well as a note that it came from a Kindle. The German site Angst & Panik has created different rules of engagement for itself, which work out to rules of disengagement: the site does not let users even select text, much less copy and paste it.15 Flickr has an API that lets users retrieve photos based on tags and comments, while the Google Photos API does not. Before long, an email app that lets you click on a date to add it to your calendar will provide a button to let you book an autonomous car to take you to any appointment not within walking distance.
Newton discovered a handful of fundamental laws that rule nature. Imagine if he had been able to invent new ones.
Action at any distance
In Newton’s universe, if you bring two things close enough, they causally interact: an asteroid can collide with a planet or a cloud of dust; water vapor can condense on a cold surface and the surface can absorb some of the condensation; the moon can pull the oceans, wearing away at the coastlines that help shape the tides. In fact, as others have said, Newton’s genius wasn’t in figuring out that there is a gravitational force pulling the apple down to the earth but in realizing that the apple was also pulling up on the earth … and, infinitesimally, on every star.16
In our new world, interoperability does the job of gravity, connecting every object across every length. But the gravity of interoperability does not diminish over distance, the way Newton’s gravity so rapidly does. The basic hyperlink—the technology that started the web era—is perhaps the best example of this: no matter how far away a page is in terms of real-world geography or topic, a hyperlink keeps it as close as a click.
Responses out of proportion
In Newton’s world, effects are proportional to causes. How far the ball goes depends on how hard you’ve kicked it. There are, of course, effects that seem disproportional to causes: a yodel causes a landslide, or a single stone melts and unleashes an earthquake. But in Newton’s world, we’re confident that a closer look will reveal pent-up forces at play; the accounts always balance. Such seemingly out-of-scale events are the exceptions in Newton’s world, where we far more commonly turn to clockworks ticking, planets orbiting, or billiard balls clacking as our models of how things happen.17
The interoperable universe has gotten us more used to small events triggering huge ones: the invention of hashtags turns Twitter into a new type of news medium, a video from a mobile phone triggers weeks of demonstrations, a software program written by a college student ends up connecting billions of people. Small causes can trigger huge events because interoperability enables more pieces to interact with more pieces more easily in a universe that is already ineffably complex.
Everything affects everything
F = Ma
That’s “Force equals mass times acceleration,” one of Newton’s most important and well-known equations. But why is it expressed that way by every high school textbook when the basic rules of algebra tell us that it’s exactly equivalent to M = F/a: mass equals force divided by acceleration, or a = F/M? W. Daniel Hillis, a polymath computer scientist, thinks he knows why: F = Ma fits into a scientific framework—causality—that Hillis argues is ready for retirement, or at least needs a rest.18
A cause is something that changes something else, so in the relationship of force, mass, and acceleration, we informally think of the cue ball as the force, the eight ball as the mass, and the change in the eight ball’s position and speed as the main effect. Hillis thinks this is because most often we think of force as the thing we get to choose to do: we can choose to aim the cue ball at the eight ball, but we can’t easily choose to change the eight ball’s mass. Cause and effect thereby fit nicely with thinking of ourselves as active agents of change who get to determine our fates by intervening in a universe of passive objects.
Hillis calls cause and effect an “illusion” and a “convenient creation of our minds.” Causal explanations “do not exist in nature” and are “just our feeble attempts to force a storytelling framework onto systems that do not work like stories.”
This is different from the critique provided by the eighteenth-century Scottish philosopher David Hume, who said that cause and effect is nothing but our mental association of one event with another that regularly, in our experience, follows it; Hillis supplies a motive for the narrative we add to these associations. But Hillis’s account doesn’t deny that there is more than a mere psychological association at work: if causality is “just a framework that we use to manipulate the world,” we only use that framework because it works. In this way Hillis is not contradicting Judea Pearl’s important recent argument for giving machine learning systems causal models so that they can make more useful predictions. Pearl argues against traditional statistics’ unwillingness to acknowledge that in some correlations, one side is the cause and the other is an effect. For example, where I live, the appearance of the constellation Orion correlates with the coming of winter but does not cause winter to arrive. On the other hand, the appearance of the sun each morning not only correlates with a temperature rise but causes it. Pearl argues that machine learning systems need to be given models of causality, and not just left to suss out statistical correlations, if they are to become true engines of science. With an understanding of causality, we can ask counterfactual questions, which are the cornerstone of scientific understanding: Mosquitoes and tropical flowers both correlate with the presence of malaria, but if we got rid of the tropical flowers, would there still be malaria? If not, then flowers don’t directly cause malaria.19
In his essay, Hillis does not dispute that some correlations are causal and some are not. He is instead pointing to our insistence on the simplification that causality can allow us. Hillis points to examples in which causality fails as a framework for our “storytelling”: weird quantum effects and complex, dynamic systems “like the biochemical pathways of a living organism, the transactions of an economy, or the operation of the human mind.” We have a motive for seeing cause and effect even in those systems, Hillis tells us: we get to continue to believe that we can control events by finding the single lever that will bring about the single result we want. Pearl wants to add a causal model to the purely correlative world of machine learning, and Hillis wants to limit simple causality as the primary model we bring to understanding every aspect of the real world.
Machine learning could benefit from having Pearl’s sort of causal model available to it, even while machine learning and the chaos of the internet have begun reducing our reliance on the sort of simplified causality that Hillis is arguing against.
Our framing of what happens in terms of causality works for us, and it is difficult—or, if you’re Immanuel Kant, impossible—to imagine experiencing the world as nothing but a series of events that have no connection beyond a probability of correlation. But I think Hillis’s point is good, for simple causality can keep out of view the cascades and inverted cascades we step through every day. Cascades include inhaling a microorganism that puts us in bed for a week, which then disturbs the schedules of those with whom we live and work, causing more effects to sprawl outward. An inverted cascade occurs when an enormous amount of energ
y results in a relatively small effect. For example, the bus that arrives on time makes it seem as if the universe’s clockwork is functioning well, but the bus driver could only count on the vehicle to move forward when she stepped on the accelerator because the tank was filled with gasoline that resulted from massive investments in extraction tools, pipelines, refineries, ocean shipping, and regulatory and taxation regimes.20
If we think in terms of cause and effect, we tend to narrow down what happens to the most immediate and tangible events. If we think about interoperability, the entire world is present in the most routine of our everyday acts.
* * *
Every day, we experience the ways in which the interoperable world is different from the causal one: interactions among things of different sorts, rules that vary by domain, the ability to create new rules, an indifference to distance, and an apparent lack of proportionality between cause and effect. These daily demonstrations are tacitly leading us to a different idea about how things happen. Whether or not we have ever heard of Pierre-Simon Laplace, we have been living in his theoretically predictable universe because change, we’ve thought, generally ticks like a clock: a gear shoves the one to which it is connected, and so on, one gear tooth at a time.
When we instead think about causality as one facet of the interoperable universe, we experience what happens as the consequence of a wild network of causes in which everything affects everything everywhere, all at once but not all in the same way, across all distances, in ways that might upset our every prediction.
A Fruitful Unpredictability
Metaphors count.
When we think about the future—not our particular future, but how the future operates—it’s not uncommon for us to envision a landscape that we are moving through. It is filled with possibilities. We choose our destination and move toward it. As we do, the possibilities fall away in the same way that when we drive through a wooded landscape, the distant trees approach us, trail off to the edges of our vision, and then vanish behind us. At the moment of the present—the Now—all that remains are the possibilities that have managed to survive. They are the ones that somehow become real. That’s how things happen, at least according to the metaphors that have shaped our experience.
Being enabled to create new ways for parts and systems to interoperate reverses that flow. Rather than possibilities narrowing as they approach, when we take the world as interoperable, we create more possibility.
The internet does this by allowing networks to interoperate.
The web allows pages on the internet to interoperate with other pages through links.
Schema.org and microformats use the web to enable systems to recognize and reuse the elements of pages. So does the Semantic Web, a form of markup created by the inventor of the World Wide Web, Tim Berners-Lee, to enable websites to make the information on them available to the apps and services anyone might build.
Open APIs from social-networking sites, government agencies, libraries, media sites, and many more provide standard, documented ways for a new application to put data to use in unanticipated ways.
Applications, including games, databases, spreadsheets, and enterprise operations systems, often let us change the rules about how they integrate with other apps and data. Apps like IFTTT—If This Then That—let users specify triggers for cross-app integration.
Machine learning systems ingest piles of data, possibly from systems that might seem unrelated, and find hidden relationships that let us predict in domains that once seemed too random to do so.
Protocol by protocol, standard by standard, app by app, system by system, and network by network, we have created a richly layered ecosystem of interoperability, each layer enabling new types of interactions. The result can be diagrammed neatly in an abstract way: the internet rules for transporting data from one site to another go at the bottom, data interchange formats go above them, and the customer-facing applications and services go at the top.
But it is far messier than that. For example, programmers create libraries of functions that can be reused by other programmers to create services—which may themselves be interoperable—at various layers. It is not uncommon for gamers to create a mod that is not itself a game but is designed to enable other gamers to make new variations on the original game. Then there are the feedback loops that arise from this multilayer interoperability: Schema.org’s standardization of airplane flight information already enables machine learning systems to find gaps in airline schedules, and that information can then be used to fill those gaps. The rise of interoperable banking data and services is enabling the world’s poor to participate in the global economy, which is likely to require new banking services.21 The introduction of the hashtag let tweets interoperate in terms of their human meaning, but it also changed the role of Twitter in the ecosystem of news and businesses, which has resulted in calls to change Twitter, the news media, politics, and even the core internet protocols in order to prevent the “fake news” that the hashtag ecosystem has supercharged.
Our daily engagement with the internet has brought us face to face with chaos, in all its ugliness and awesomeness. The success of AI’s algorithms reveals complexity that we wrote off as not worth paying attention to because there was nothing we could do about it. Our experience with these two technologies is revealing interoperability as the basic enabling condition for the next moment not being exactly the same as the prior moment—the very definition of what it means for something to happen.
But the secret of interoperability is that even if it’s created with a narrow purpose, people will find a broad and unpredictable range of things to do with it: a standard way to connect printers to a computer becomes a way to attach a braille reader or a sewing machine that embroiders hats with custom slogans. The more unexpected uses interoperability engenders, the more valuable it is.
Reducing the world to what we could understand and predict made sense when knowing our world was our best way of controlling it. Now machine learning is letting us manage more and better by not insisting that we understand exactly how it works. Likewise, our new online ecosystem brings us significant benefits by letting others build whatever they want using the resources openly available to them online, and generally without having to ask anyone for permission.
Yet we are living through an aching and demanding contradiction. We are accustomed to reading the past twenty years as a period in which governments have increased surveillance and businesses have extracted every scrap of personal data they need to micromanipulate our behavior. Indeed, a surprising number of us are quantifying and recording our own heartbeats, steps, sleep periods, and grams of food, always looking for clues for a better life. This boiling of chaos down into a controllable residue is happening at the same time as we are purposefully increasing unpredictability. This contradiction is part of the definition of our new age. We may not be able to resolve it any more than we were able to resolve the contradictions of reason and faith, free will and determinism, individualism and communitarianism, altruism and selfishness.
If I had to guess, I’d say that the ideas of connection, of collaborating while preserving differences, of openness, of enablement, of play, and of hope are going to be dominant in the long term. That’s my preference, which undoubtedly affects my judgment.
I understand that my argument for this hope is far from ironclad because it is based on some abstract assumptions: control is isolating, but interoperability is connective. Control is fragile; interoperability is resilient. Control is the narrow path a flashlight shows. Interoperability is the way light illuminates, feeds, warms, and liberates, all depending on what it touches.
And the forest through which the lit path runs is the world that happens.
Coda: Signs and Causes
When a culture looks at a bird’s entrails to predict the fate of a king, we snicker: bird guts have no causal relationship with whether the king lives or dies. But these cultures are not looking for causal relationship
s. For them, and for much of our own culture’s history before Newton, the universe is not a clockwork of causes but a web of meaning.22
For example, we used to assume that plants that look like parts of the human body can cure diseases of those parts. We now know that that’s wrong. But it is not as without sense as it at first sounds. In his book The Order of Things, the philosopher Michel Foucault quotes the fifteenth-century medical genius Paracelsus: “It is not God’s will that what he creates for man’s benefit and what he has given us should remain hidden.… And even though he has hidden certain things, he has allowed nothing to remain without exterior and visible signs in the form of special marks—just as a man who has buried a hoard of treasure marks the spot that he may find it again.”23
And here’s the twist ending: machine learning is making us comfortable again with relying on signs.
Despite its ethical shortcomings, Cambridge Analytica’s promise about the 2016 US presidential election was not patently ridiculous: by analyzing Facebook data, its machine learning algorithms might have been able to predict which political ads would best work on users clustered by personality type. That analysis need not focus on, or even consider, overtly political information from Facebook. For example, in 2013, two psychologists at Cambridge University gave fifty-eight thousand volunteers a personality test and then correlated those psychological profiles with what the volunteers “liked” on Facebook. (“Liking” in this case means pressing the Like button.) For example, it turned out that being extroverted correlated strongly with liking Nicki Minaj, while openness correlated with liking Hello Kitty.24
We can perhaps make up stories about why that’s so, but we can also imagine correlations that defy such attempts at explanation. For example, Cambridge Analytica may well have had access to more than what people liked on Facebook. Applying machine learning to all that data might reveal—hypothetically—that writing long posts on weekdays, responding quickly to posts by people whose pages the user infrequently visits, and using the word etc. in more than 12 percent of one’s posts all correlate with being a moderate Republican. Maybe posting photos that often show a city skyline in the background and double-clicking on buttons that only need a single click correlate with liking cats over dogs and supporting the gold standard. It’s also conceivable that very small changes might result in very different predictions. Clicking the Like button for Nicki Minaj might make it much more likely that you’re an extrovert, but a tiny bit more likely that you overtip. Put those correlations into a web in which another thousand data points each make it slightly more likely that you’re an overtipper, and the system might make a probabilistic prediction that you’re 86 percent likely to tip your Starbucks server two dollars when fifty cents would be enough and zero would have been acceptable.
Everyday Chaos Page 13