My first time selling was a shitshow. My sales manager and the head of the company constantly bickered over the best times to cold call, what to say, what not to say, who to call and everything else. They’d pull up stories about a call a decade ago, a sale the decade before, what happened to their cousin or their friend two towns over to make their points. One of my tasks was to look through our call records to figure out what actually worked. It was long arduous work, but I eventually I found patterns and ways to improve. They were ignored, unless the points were convenient in the process of bickering.
When I went to Google, one of the first things I noticed was how differently conflicts were resolved. Rather than disagreements devolving into shouting matches and personal insults, they led to the articulation of uncertainty, a plan to address that uncertainty, and agreement on next steps for each possible result. That came from a trust that the entire team was working towards the same goal, and that you’d both try to improve the team. Googlers, more than any other people I’ve met, were data-driven and saw the best in everyone.
Those were the qualities that made me love Google. They also made us blind to the damage done by our platforms..
A mantra at Google was Right Person, Right Message, Right Time. I can’t count how many quarterly business reviews I ended with that — how a new product, new tweak, new analysis, would let us find the people our client was likeliest to persuade, with the ad likeliest to persuade them, in the conditions likeliest to persuade them.
The best way to find that is to try to find what similar messages persuaded similar people in similar conditions. Instead of a human going through a few hundred sales records over the course of a month to find those patterns, a machine could go through trillions of records a day. It boggles the mind how much better those machines are than any human. They are so good that Google got people to click on ads 28 billion times in Q3 of 2017.
BUT the people likeliest to be persuaded by an ad are by definition the most susceptible people. Right user, right message, right time finds people most susceptible to your message, delivers the message they’re most susceptible to, in their most susceptible condition.
Machines optimize towards the most susceptible, because the most susceptible are the likeliest to click or buy. The non-susceptible don’t click, their actions aren’t recorded, and they’re omitted from the machine’s optimization, or the machine optimizes away from them.
A human may have qualms selling snake oil to a single mother of four with a love for psychics and the occult, but a machine only knows it needs to maximize clicks, and visitors to website wcFMA6mtLo from cluster xTAV3we and demographic At8vsVG26A6 are 4X likelier to click on ad Nnl9iTMMi.
By the time humans come in, they’re too far removed to notice. Google recorded 28 billion clicks in Q3 of 2017. That’s over a million clicks for each person working in ads, in just three months. Only a machine can keep track or make sense of the number of sites required to reach that number. A human never sees the occult websites the ads serve on and, because data is anonymized and aggregated, can’t see the single mother of four with a love for psychics and the occult.
All a human sees are the top level stats: the percent of ads shown that get clicked on, the percent of clicked ads that result in sales, and the value of those sales over the cost of the ads. And humans at Google get promoted on improving those numbers. That means getting better at finding the most susceptible.
— — — — -
To claim Google’s ads optimize toward susceptibility runs afoul of Google’s deep-seated belief in human potential. That belief imbues every part of the company. One of the ten foundational beliefs of Google is “Democracy on the web works.” The values of Google subsidiary YouTube include that “all people should have a chance to be discovered, build a business and succeed on their own terms, and that people — not gatekeepers — decide what’s popular” and “all people should be able to find communities of support, break down barriers, transcend borders and come together around shared interests and passions.”
YouTube’s done a great job fulfilling that vision. By letting machines find what makes people click on a recommended video and what makes them stick around to watch, YouTube creates the world’s best AI to deliver creators their most engaged possible audience. By hosting gigabytes worth of video, accessible instantly throughout the world, for free, YouTube breaks down barriers and transcends borders. Check out Google’s YouTube channel to understand why those goals are so worthy. With Google products, Mexican bakers find global customers, the ambitious but downtrodden turn into CEOs, and orphans find their home.
When you’re at Google, the belief that human potential is a limitless and unambiguous good is infectious. One of the most common interview question is, “What’s your favorite Google product”. Good answers show how Google products helped you improve, so Google only gets people who know how to use Google products to better themselves. Working at Google makes it easy to believe the world is full of the well-intentioned, the intellectually curious, and the positive. Google is full of great people and Google helps great people outside be better. It’s easy to think that’s the real world.
But this is the real world:
For every American with an advanced degree, one believes Obama is the antichrist. For every American who can name the three branches of government, one says planes spray mind control gas. For every American who knows the cause of the Cold War, one can’t find the Pacific Ocean on a map. For every American who learned a foreign language outside the home, twenty five think the sun revolves around the earth.
These stats don’t shake my belief in the potential of all people. Anyone can learn the three branches of government. Anyone can learn the earth rotates around the sun. Anyone can learn the chemical process behind contrails. And YouTube deserves credit for giving a platform to some of the world’s most engaging and accessible scientists, academics, and historians. But access to their videos is not enough. With 300 hours of video uploaded every minute, the gatekeeper to those videos is whatever lets users find their desired content. And, in the platform’s own words, “finding relevant content is at the heart of YouTube”. The machines behind YouTube’s search and recommendations are the gatekeepers, and that those machines look for the most susceptible users in their most susceptible conditions has dire consequences.
The chances a human gatekeeper makes decisions in the best interest of society falls far short of 100%, and probably even 50%, but that’s far better than the 0% chance machines will do the same. FOX aired a moon landing conspiracy theory film in 2001 and was panned for doing so. The network never aired the film again. But every minute of every day YouTube lets tens of millions of moon landing hoaxers “find communities of support, break down barriers, transcend borders and come together around [a] shared interest” in proving science is a scam. A human at Google may never recommend someone who watched NASA ADMITS WE NEVER WENT TO THE MOON also watch 5 Things NASA Doesn’t Want You To Know!, but a machine only knows viewers of video DpPMoIv1lxI are likeliest to watch yJ6MaYI3xtU.
Google, more than anywhere I worked before, was a data-driven company. No sin was graver than putting a viewpoint formed off of base instincts on the same footing as one backed by data and rigorous experimentation. But YouTube does that millions of times a day. If you’re a flat earther, an anti-vaxxer, a white supremacist, or a Russian propagandist, YouTube handpicks the people in the world most susceptible to your message and uses the world’s best AI to keep them coming back. The programmer responsible knows that a tweak to clustering in recommendation yielded a 8% QoQ growth in CTR on a test of 1% of traffic. For the eight million viewers of This is why we never returned to moon! [sic], they now know we nuked moon cities. Pity the gatekeepers at NASA, and their top video’s four million viewers.
