Interoperability is the ability to use elements designed for one system in another system the designers may never have heard of, and in ways that they did not anticipate. In slide deck after slide deck, the concept of interoperability is introduced with a photo of Lego pieces. But that’s misleading. The Lego Group fiercely kept other companies from producing interoperable toy brick sets for as long as its patent held out. On the other hand, Lego pieces are a great example of things that intraoperate—they work beautifully with other pieces from the same system. A 3.5 mm audio plug is a better example of interoperability because it can be plugged into so many different types of devices from many, many different manufacturers.
John Palfrey and Urs Gasser give another type of example in their book Interop: the standardization of the vocabulary used by air traffic controllers.2 Pilots must speak English or at least learn Aviation English: three hundred words relevant to flying a plane.3 Likewise, standardized barcodes on commercial items enable products to interoperate with checkout systems in stores around the world. Credit cards enable people’s wallets to interoperate across different national currencies. Standardized export formats for spreadsheet files, such as CSV (comma-separated values)—invented in 1972—let you load a Lotus 1-2-3 spreadsheet you created in 1995 into a modern spreadsheet app, as well as into most database software and graphic charting applications.4
Indeed, interoperability is at the heart of many of the examples we have looked at so far:
Platforms with public APIs crack open a site’s internal services so that applications written by others can interoperate with those services.
Open data lets information be put to use by any app or system that needs it.
With agile programming, developers create interoperable functional modules that can work with any other module that knows how to provide it with the expected inputs.
A user can mod a computer game if the game accepts new objects, graphics, and rules of interaction that work with its core mechanics.
The ability of computers to interoperate with other hardware devices enables on-demand printing, embedded insulin pumps, household thermostats, and the Internet of Things.
Machine learning works by bringing together massive amounts of data generated by many different systems in forms that the system can ingest and analyze. Now vendors are finding ways to enable their machine learning systems to interoperate with other machine learning systems. For example, six giant technology corporations—Amazon, Google, IBM, Microsoft, Oracle, and Salesforce—have taken the Cloud Healthcare Pledge to develop open standards by which health data can be shared among their systems.5 Judea Pearl in The Book of Why talks about his own work in making data “transportable”—that is, usable across machine learning systems.6
The internet itself makes local networks interoperable so they can pass along information packets using standard data formats and transmission protocols.
We’ve gone with interoperable solutions in these cases because interoperability makes systems more efficient, flexible, sustainable, and expandable. But the effect that is most directly altering our understanding of how things happen is interoperability’s math: when the resources and services designed for one system are interoperable with other systems, unexpected value can, and probably will, emerge.
The Useful Unpredictability of Standards
You can see this in an important joint effort by Google, Bing, Yahoo!, and Yandex (a Russian search site) called Schema.org, which specifies a vocabulary of terms that website owners can invisibly embed in their web pages so that information on those pages can be better understood by the search engines.
For example, imagine the word bridges is somewhere on a web page. To respond to user queries with relevant results, a search engine would first want to know whether the page is about architecture, travel routes, movies, or orthodontics. Schema.org provides a standard way for the page creator to embed a hidden tag that says explicitly that the page is about, say, movies. Then the search engine wants to know what sort of thing “bridges” is referring to on that page. Is it a word in a title (The Bridges of Madison County), the name of an actor (Jeff Bridges), or part of the marketing description (“bridges the gap between love and desire”)? Schema.org provides a vocabulary to express just this sort of information, all hidden from the people visiting the page but visible to the search engines. Likewise, with Schema.org, a site about the book Sense and Sensibility could indicate that Jane Austen is the name of the author while Elinor Dashwood and Colonel Brandon are names of characters in that book. With information like this, when someone searches for Jane Austen’s Sense and Sensibility, the search engines can confidently offer a list of characters in the book, as you often see in the information boxes Google provides on the right-hand side of many search results pages.
Schema.org thereby makes the search engines smarter and their users happier. But, not coincidentally, it also makes the information on web pages far more interoperable.
Schema.org does this by providing sets of tags for scores of different types of things that pages talk about, including organizations, people, events, recipes, medical conditions, news articles, and local businesses.7 These vocabularies are created in collaboration with the relevant communities of practice in an open fashion that has more in common with how the internet was created than with how Henry Ford designed the Model T.
Once Schema.org’s tags have made it clear that a page is about a book, and has tagged the book’s title, the name of the author, the names of characters, the date of publication, its geographic settings, and so forth, Google can cluster that information with what it has learned from other pages. It can connect all that it has learned about Jane Austen with other works of fiction, with geographical data, with weather information, with historical information about the role of marriage, with the structure of British royalty, and with other Oscar-winning movies starring Emma Thompson. By making information on billions of web pages intraoperable, these webs of information—“graphs,” as they are known technically—have made search engines the most knowledgeable institutions in the history of the planet.
Because Schema.org’s vocabulary of tags has to be public so web page creators can insert them, it also increases interoperability by enabling any application on the web to locate, extract, and reuse the information on web pages just the way the search engines do. Evan Sandhaus, who was executive director of technology at the New York Times when we spoke, said that Schema.org is “probably even a bigger deal in terms of how news organizations get their data out there than APIs.” He explained that sites such as Facebook and Twitter automatically scan pages linked to by users, looking for Schema.org’s hidden tags in them. This lets those sites do things like identify and format the headline appropriately for their site, display the date in a way consistent with their users’ preferences, and embed topic tags so that the site can suggest related links. What Facebook and Twitter do, literally any developer with access to the web can do for her or his own purposes. “The page itself becomes kind of like an API,” Sandhaus said, in that it can be queried for information about itself. “That’s huge!”8
So what started as a simple way for website creators to identify the nature of their contents to all the major search engines has become an open-ended resource for any app that can think of a use for that information. For example, Yext began as a service that provides the search engines with trustworthy information about the locations, hours, and so forth of its clients. Then in 2018 it began a new initiative, called Yext Brain, that builds knowledge graphs for businesses—connecting a wide range of information about, and relevant to, a company. “Why should you have to be a Facebook or Google to have that resource for your company?” asks Marc Ferrentino, Yext’s chief strategy officer. As a primary way to represent information in those graphs, Yext has repurposed Schema.org, with some extensions. By turning this publishing standard into an internal data standard, Yext is not only benefiting from a well-thought-out data structure, but i
s also making it easier to publish information developed by the Brain to the web by embedding it in Schema.org tags—a virtuous circle of interoperability.9
Or, as another example, Microsoft Cortana—Microsoft’s Siri, as it undoubtedly hates to be described—uses hidden Schema.org tags about flight information to show users a flight’s status.10 That makes Windows a slightly more valuable product for Microsoft, but the real value of making all this data interoperable is what the next developer will do with it. Perhaps she’ll mash it up with astronomical information so hobbyist plane-spotters can identify the flights overhead. Perhaps she’ll start a lost luggage recovery business based on that data. Perhaps she’ll look at data about the geographic clustering of illnesses to disprove the contrail conspiracy theory that planes are releasing harmful chemicals when they fly. “But that’s not what it was designed for!” is not a criticism of Schema.org but a statement of its value.
The adoption of this system is driven by the economic and attentional power of search sites. But exercising that power most effectively required the search sites to give up some of their control. Rather than specifying their own standards from on high and shoving them down the throats of website owners, the search engines wisely adopted the microformats approach that had begun in the mid-2000s. Its early advocates were impatient with the typical methods industries had used when they wanted to make information in their documentation interoperable. For example, a standard called SGML (Standard Generalized Markup Language) was created in the 1960s to enable entire industries to make the documents they share with their supply chains interoperable by agreeing on those documents’ structures and their tags for common elements. But SGML standards were complex and prone to becoming mired in years of contentious arguments about trivialities, such as whether a table in a document is a set of rows intersected by columns or a set of boxes aligned into rows and columns. Microformats, on the other hand, are developed by small groups of knowledgeable people who skip over the contentious parts so they can come to quick agreement on the 80 percent that everyone agrees on. The result is quick and dirty standards that are simpler to implement, and from which companies can quickly benefit. The more companies that adopt a particular microformat, the greater its attractive force for getting yet more companies to adopt it.
Standards, whether a niche microformat or the more encompassing Schema.org, shape the space around them and enable further interoperability. For example, currently, if you gathered up college course syllabi from around the world to see how patterns of assigned readings vary geographically, or how the same sources are used in units on different topics, you would face a difficult computing problem, for there is no standard way in which syllabi express the information they contain and no Schema.org-like way of tagging that information. So a computer program trying to identify the information in a syllabus can’t easily distinguish a book title from the title of a study unit, or the topic of an article from the topic of the final paper. As a standard emerges for how the elements of syllabi are to be tagged, it will become far easier to extract, aggregate, and interrelate those elements. With interoperable syllabi, professors could learn from how their colleagues around the world are organizing courses and readings.11 This would encourage yet more professors to make their own syllabi interoperable. Perhaps apps will be written to let the information about required books interoperate with used-book services to help drive down the cost to students. The uniformity of the data format would also make quantitative analysis far easier and machine learning more accurate, which is likely to unearth relationships that could unveil hidden currents in our educational systems and culture. Educational platforms could use that information to create new learning services and to guide students to interests they might never have otherwise discovered.
It will also … who knows? And that’s the point. In the causal clockwork universe, we look for the dials to turn and the levers to pull. The essence of levers is that they have predictable outcomes. The essence of interoperability is not simply that the information in syllabi can be more effectively shared, or that an audio cable fits many different devices, or that search engines can show us local times for the movies we ask about or list the cast and the characters they play. Rather, the essence of interoperability is that it specifies what pieces do and how they fit together so they can be used in unpredictable ways, sometimes for projects the original creators couldn’t have imagined and will never hear about.
Interoperability Is the New Causality
If you grasp a pool cue, your hand is interoperating with it. When you slide the cue forward into a ball, the stick interoperates with it. When the ball hits another ball, it interoperates with it. When that second ball hits the side of the pool table, they interoperate.
These interactions within the material world are causal: an action brings about a determinate effect. Newton did a superb job enabling us to predict the outcome of colliding objects, and we’ve obviously made advances in understanding the causal relations among everything from microbes and diseases to windshields and pebbles.
Now our new technology is leading us to think about causality as just one—obviously crucial—type of interoperability. The internet has shown us how fluidly information-based systems can affect one another if we want them to, and occasionally when we don’t. Machine learning has made abundantly clear how inadequate most conceptual models of causal events—A causes B—are, for in a machine learning system, B may be brought about by the interrelationship of innumerable variables. Together, these two technologies are getting us used to the idea that causality is just one way things may interact.
This is changing how we think things happen in several crucial ways:
Working across kinds
A clockwork’s gears work together causally because they are of the same kind: solid, scaled to fit with one another, shaped with the connecting gears in mind. Interoperability, on the other hand, allows things that are different in kind to interact.
That got much easier as we moved business and culture onto digital media, for, like metal gears, digital bits have much in common with one another: they are binary and can be manipulated by digital machines. But unlike clockwork gears, bits can do more than go around in circles. Bits can represent just about anything we want. We can then tell those bits to interoperate in particular ways simply by writing the rules—the code—governing their engagement. For example, a developer can often just write a couple of lines of code for an app to display digital images, whether the app is a word processor embedding a photo, a music player displaying an album cover, or an online game allowing the user to choose an avatar.
Interoperability’s ability to work across different systems is important for more than technical reasons. As Palfrey and Gasser say in Interop, “There is an essential difference between making complex systems interoperable and simply making everything the same.”12 Making things the same works well when the units are relatively simple and neutral, such as mapping systems that use longitude and latitude to precisely position items on the map.13 It gets much harder when the units carry human meaning, as when geographic systems disagree about whether Palestine is a country or what the exact boundaries of “the Wine District” are.
Interoperability, on the other hand, can enable interaction while preserving differences. For example, when you use your credit card to pay for a meal in a foreign land, the transaction is enabled by uniform standards for the transmission of financial data, connecting banks in different countries, each of which has its own laws, customs, and currencies. Similarly, you can share the photos you’ve tagged “Pat and Ari’s Amazing Vacation 2019,” even though the friends and relatives who receive them may view them on a different device and relabel them “Pat and Ari Go Somewhere Boring.” Libraries and archives that disagree about which information about items is worth collecting can share what they have, transposing it into common categorization standards while maintaining their own, local way of thinking about things. A little bit of shared in
formation can lead people to works that disagree about everything else. And a lot of shared information can enable machine learning systems to help us discover meaning in those differences and similarities.
Indeed, for the past twenty-five years, we’ve been getting accustomed to interoperability working not just across banking systems, image programs, and archives but across radically different sorts of devices—although anyone who has ever tried to connect this year’s television to last year’s laptop can attest that it’s not always as easy as we would like. Nevertheless, the integration of different devices has been a part of our experience of the web at least since 1993 when scientists at Cambridge University hooked up a video camera to the web so they could use their browsers to see whether there was a fresh pot of coffee in the community kitchen.14 That coffee pot became world famous, setting the tone for how we imagine the web’s future: more and more disparate items and information connected across their differences. The Internet of Things is making this real. Your thermostat can interact with your smartphone, your smartphone can interact with your baby monitor, your baby monitor can send JPGs to your smartwatch, and all of them can tweet at you angrily behind your back. (Be sure to check your toaster’s privacy statement.)
Classic causality lets things interoperate only if they are very much alike: two metal gears, nine billiard balls bounded by a pool table, a comet and two planets. Interoperability, on the other hand, bridges differences.
Everyday Chaos Page 12