In the coming two decades the challenge and opportunity is to harness filtering technologies to cultivate higher quality attention at scale. Today, the bulk of the internet economy is fueled by trillions of hours of low-grade commodity attention. A single hour by itself is not worth much, but en masse it can move mountains. Commodity attention is like a wind or an ocean tide: a diffuse force that must be captured with large instruments.
• • •
The brilliance behind Google, Facebook, and other internet platforms’ immense prosperity is a massive infrastructure that filters this commodity attention. Platforms use serious computational power to match the expanding universe of advertisers to the expanding universe of consumers. Their AIs seek the optimal ad at the optimal time in the optimal place and the optimal frequency with the optimal way to respond. While this is sometimes termed personalized advertising, it is in fact far more complex than just targeting ads to individuals. It represents an ecosystem of filterings, which have consequences beyond just advertising.
Anyone can sign up to be an advertiser on Google by filling out an online form. (Most of the ads are text, like a classified ad.) That means the number of potential advertisers might be in the billions. You could be a small-time businessperson advertising a cookbook for vegan backpackers or a new baseball glove you invented. On the other side of the equation, anyone running a web page for any reason can allow an advertiser to place an ad on their page and potentially earn income from this advertising. The web page could be a personal blog or a company home page. For about eight years I ran Google AdSense ads on my own personal blogs. The hundred dollars or so I earned each month for showing ads was small potatoes for a billion-dollar company, but the tiny size of these transactions didn’t matter to Google because it was all automated, and the tiny sums would add up. The AdSense network embraces all comers no matter how small, so the potential places an ad could run swells to the billions. To mathematically match these billions of possibilities—of billions of people wanting to advertise and billions of places willing to run ads—an astronomical number of potential solutions are needed. In addition, the optimal solutions can shift by time of day or geographical location—and so Google (and other search companies like Microsoft and Yahoo!) need their gigantic cloud computers to sort through them.
To match advertiser with reader, Google’s computers roam the web 24 hours a day and collect all the content on every one of the 60 trillion pages on the web and store that information in its huge database. That’s how Google delivers you an instant answer whenever you query it. It has already indexed the location of every word, phrase, and fact on the web. So when a web owner wants to allow a small AdSense ad to run on their blog page, Google summons up its record of what material is on that page and then uses its superbrain to find someone—right that minute—who wants to place an ad related to that material. When the match consummates, the ad on the web page will reflect the editorial content of the page. Suppose the website belongs to a small-town softball team; the ads for an innovative baseball mitt would be very appropriate for that context. Readers are much more likely to click on it than an ad for snorkeling gear. So Google, guided by the context of the material, will place mitt ads on softball websites.
But that’s just the start of the complexity, because Google will try to make it a three-way match. Ideally, the ads not only match the context of the web page, but also the interest of the reader visiting the page. If you arrive at a general news site—say, CNN—and it knows you play in a softball league, you might see more ads for sports equipment than for furniture. How does it know about you? Unbeknownst to most people, when you arrive at a website you arrive with a bunch of invisible signs hanging around your neck that display where you just came from. These signs (technically called cookies) can be read not just by the website you have arrived at, but by many of the large platforms—like Google—who have their fingers all over the web. Since almost every commercial website uses a Google product, Google is able to track your journey as you visit one page after another all across the web. And of course if you google anything, it can follow you from there as well. Google does not know your name, address, or email (yet), but it does remember your web behavior. So if you arrive at a news site after visiting a softball team page, or after googling “softball mitt,” it can make some assumptions. It takes these guesses and adds them to the calculation of figuring out what ads to place on a web page that you’ve just arrived at. It’s almost magical, but the ads you see on a website today are not added until the moment after you land there. So in real time Google and the news site will select the ad that you see, so that you see a different ad than I would. If the whole ecosystem of filters is working, the ad you see will reflect your recent web visit history and will incline more to your interests.
But wait—there’s more! Google itself becomes a fourth party in this multisided market. In addition to satisfying the advertisers, the web page publisher, and the reader, Google is also trying to optimize its own score. Some audiences’ attention is worth more to advertisers than others. Readers of health-related websites are valuable because they may potentially spend a lot of money on pills and treatments over a long period of time, whereas readers of a walking club forum buy shoes only once in a while. So behind each placement is a very complicated auction that matches the value of key context words (“asthma” will cost a lot more than “walking”) with the price an advertiser is willing to pay along with the performance level of readers who actually click on the ad. The advertiser pays a few cents to the web page owner (and to Google) if someone clicks on the ad, so the algorithms try to optimize the placement of the ads, the rates that are charged, and the rate they are engaged. A 5-cent ad for a softball glove that gets clicked 12 times is worth more than a 65-cent ad for an asthma inhaler that gets clicked once. But then the next day the softball team blog posts a warning about the heavy pollen count this spring, and suddenly advertising inhalers on the softball blog is worth 85 cents. Google may have to juggle hundreds of millions of factors all at once, in real time, in order to settle on the optimal arrangement for that hour. When everything works in this very fluid four-part match, Google’s income is also optimized. In 2014, 21 percent of Google’s total revenue, or $14 billion, came through this system of AdSense ads.
This complicated zoo of different types of interacting attention was nearly unthinkable before the year 2000. The degree of cognification and computation required to track, sort, and filter each vector was beyond practical. But as systems of tracking and cognifying and filtering keep growing, ever more possible ways to arrange attention—both giving and receiving—are made feasible. This period is analogous to the Cambrian era of evolution, when life was newly multicellular. In a very brief period (geologically speaking), life incarnated many previously untried possibilities. It racked up so many new, and sometimes strange, living arrangements so fast that we call this historical period of biological innovation the Cambrian explosion. We are at a threshold of a Cambrian explosion in attention technology, as novel and outlandish versions of attention and filtering are given a try.