Thirteen

“Welcome!” Aurora came by my desk on my first day, lugging a giant potted plant. “I forget what kind it is, but they told me it was nearly impossible to kill. Where should we put it?” It was so big, it ended up on the floor next to my desk, a spot of green marking my place.

“Right, let’s synchronize our calendars. Bring up the system we use here. Not that one, it’s…” She pointed at my laptop screen. “That’s it. I want to set aside some girl time. An hour a month for starters, is that okay?”

“Sure.”

“Not for work. Just to chat, compare notes, figure out how to survive in a joint like this. I mean, look around. It’s all men. I’m so excited to have a woman to talk to finally.” She danced a little in place as she tapped on her phone, scheduling a dozen get-togethers. “We’ll have a glass of wine, or take a walk or whatever. If we don’t set time aside right now, you’ll get too busy and you would not believe how hard it is to schedule meetings. Busyness is a cult around here. Did you accept all of those appointments?” She looked over my shoulder and nodded; I’d figured out how to add her times to my so-far nearly blank calendar.

“We don’t use Google Calendar?”

“Girl, please. That spyware? No way. We host our own email, calendars, everything we need. You signed the NDA, didn’t you?”

“The don’t-tell-anybody-anything form? I couldn’t even get past the lobby until I signed it.”

“Everyone uses them in tech, but here the security is serious. People would love to get their hands on our code. We even have white-hat hackers on retainer to test our defenses. You do not want to get caught fraternizing with the enemy or, you know, sending a work-related message to someone through a plain vanilla gmail account. You’ll get the spiel during orientation. Eventive’s motto: we’re paranoid and proud of it. The other motto is ‘work harder.’ You’ll find out soon enough, you’ll never feel caught up. Speaking of which, I have to get across town to meet with my advisor. He’s finished reading the latest chapter of my dissertation. Almost done, hallelujah!”

She danced away and left me to be shown around the office and its amenities by another client solutions analyst, an intense young man from Hong Kong whose thick black hair bristled like a hedgehog. Then someone from HR took me through company policies and got me signed up for the benefits package, and I was handed over to my manager, who gave me a blitzkrieg introduction to the job. After a hasty company-supplied lunch with Robbie, I was turned loose to master the contents of a digital library full of technical manuals. I needed to be a quick study, since I would start providing support and training for a daunting portfolio of clients before the week was out, and the back end of the product was a tangle of AI-powered subsystems that were invisible from the client side. I would have to dig into the code to understand what it could do.

I was in heaven.

Eventive HQ occupied the top floor of an old red-brick warehouse in the Seaport District, six blocks from Robbie’s condo. It preserved vestiges of its historic identity with arched windows and accents of exposed brick and thick wood beams, along with some mysterious cast-iron mechanical equipment hanging from the ceiling like brutalist industrial mobiles. It was otherwise thoroughly contemporary: an open-plan office with lots of rolling whiteboards, and ergonomic chairs and adjustable tables that could be pulled into pods for meetings. While there wasn’t any visible privacy, the renovation had included some clever acoustical modifications that made the wide-open workspace surprisingly quiet. I could be talking with a client without being overheard at the next desk.

There were enclosed offices at one end–Robbie had one, and so did Aurora–and at the other a set of meeting rooms with walls coated with whiteboard paint where consultants presented their work and technical challenges were hammered out. In the center there was a well-equipped galley where you could pick up fruit, a croissant, or make yourself a latte whenever you liked. For those who preferred junk food, there were bags of chips, a gum ball machine, and a small freezer full of frozen burritos that could be heated in a microwave. Lunch was delivered daily and pizzas arrived at ten p.m. for anyone who worked late. Twice a week, a masseuse dropped by. You had to go to the basement to use the gym, shared with other tenants of the building, but the company supplied a personal trainer.

I’d heard about perks like these. It was part of the mystique of Silicon Valley, the idea that the Chosen People who rose through the meritocracy got special treatment precisely because they were so special, working in an industry that was reinventing everything. It was another thing to experience the perks, especially after months of a budget so tight I was barely being able to afford food, let alone a good cup of coffee. The galley was my favorite place until Robbie said something about my weight and I began to ration myself. I also made a resolution to go to the gym regularly, but that didn’t happen. Aurora hadn’t been kidding. I didn’t have time.

Nobody did. That was another feature borrowed from Silicon Valley culture: an almost competitive devotion to putting in long hours. Our client base was growing, and new projects were constantly in various stages of development. We were all working flat out. Robbie and I hardly saw each other except in passing at work.

I was disappointed that first day when I learned what a Client Solutions Analyst was. I wasn’t going to be doing the work Robbie did, writing code for new products. I was the person clients called when something went wrong, or when a new release was rolled out and they needed help customizing its features. In between those tasks, I routinely checked in with the clients on my list to see how things were working out and to gently nudge them toward upgrades or new products. After doing that for six months, my duties began to include flying out to talk to companies who’d shown interest, after doing extensive background research to create customized pitch decks. I would seek out programmatic marketing, SEO, and AI conferences I could attend while on the road to scout out potential clients or competitors.

So it was a bummer that I was in sales and tech support instead of DevOps, but there was a silver lining. To solve the problems that cropped up, I got to dig into the code to see what tweaks and modifications were possible. After demonstrating that I wasn’t going to break anything, I earned the permissions to make minor fixes. It was an opportunity to learn by reading through lines of code that I never could have written. As with any debugging chore, some of it was frustrating and tedious, but those moments were countered by bursts of dopamine when I kicked a problem’s ass.

I was also making money. A lot of it.

I wasn’t completely unfamiliar with wealth. My grandparents were rich. They lived in a Victorian town house in Cambridge and had a “cottage” on the coast of Maine worth several million dollars. They thought nothing of vacationing in expensive resorts in the tropics or taking an educational trip to Laos. Most of the money was inherited on my grandmother’s side, so she came to their marriage already accustomed to the lifestyle. My grandfather, a professor of Comparative Literature at Harvard, took to it naturally. I had lived with them for a few months at a time, twice in Cambridge and once at their summer place, when it was either them or foster care, so I knew what it felt like to live in a household where people were paid to clean and cook, you were supposed to know which fork to use, and the napkins were made of cloth. In spite of all that money, they were self-important, miserable people who didn’t care much for each other and cared not at all about me.

Though my rebellious mother had walked away from all that, making sure we lived in poverty just to spite them, there was an indelible stain of education and privilege left on her that she couldn’t scrub out. I had inherited some of that–at least enough to pass. But it felt deeply strange at first, not having to worry about money, and in the early weeks of my employment I often made mistakes, doing things like going to the bus stop instead of ordering an Uber. Within a few months, that strangeness passed, I learned how to spend money like the rest of them, and I almost got used to not being broke.

Still, some things I hadn’t learned, like how to dress the part. Shopping for clothes anywhere other than thrift stores made me incredibly anxious. When I was preparing for my first pitch, the woman brought in to coach me on making winning sales presentations was horrified by my first attempt to dress professionally. She told me where to have my hair and nails done and gave me the name of a personal shopper who also worked for several of the higher ups. The shopper took my measurements, made a few notes about my preferences (“I don’t know, comfortable?”), and a week later showed up with a closet-full of clothes and a chart of what went with what so I wouldn’t make mistakes.

Robbie didn’t have to dress up, even when he traveled, because his usual gear was what everyone expected geeks to wear, but his basic jeans and hoodies were surprisingly expensive and he began to collect high-end watches with a steampunk vibe. Between us, we spent a lot, but we didn’t have to worry about it. We had more money than time to spend it.

That was another thing to get used to: there was no such thing as free time at Eventive, and not much time for sleep. I counted on those ten p.m. pizzas more often than not to keep me going.

Aside from the pizzas, though, what kept me alert and engaged was the product itself. I was amazed at what Eventive was building. It was based on a dizzying combination of qualitative and quantitative research, massive amounts of data, and computing power harnessed to automate highly-targeted messaging. Eventive had built up an astonishing collection of data, scraped from social media, bought from data brokers, gathered through sophisticated surveys and fieldwork, parsed from public records and websites, swept up by browser plugins that monitored behind-the-scenes programmatic ad auctions that captured information about individuals’ religion, sexuality, mental health problems, or whether they like the color yellow.

In my first days on the job, Robbie mentioned nonchalantly that they had compiled profiles of virtually every American over age thirteen, as well as collating a similar level of detail for citizens of several more countries where Eventive operated. (It was easier in some of those places, he said, the ones with powerful leadership; they were often able to get direct access to internet use by individuals through telecom providers.) To handle all that data, some bright data analysts recruited from intelligence contractors had found ways to automate merging and cleaning complex data dumps so the profiles could be updated continually.

A rotating collection of eager researchers without permanent employment but with PhDs in anthropology, psychology, marketing, and media studies gathered in the conference rooms with Robbie and Aurora to plan out ethnographic fieldwork and large-scale experiments that could enhance the products we sold to customers who wanted to customize messages down to the individual level. What made a person afraid or angry? What deep cultural narratives influenced their behavior? How could we identify individuals by psychological traits that would make them especially receptive to a certain kind of message?

“Wait, hold on a sec,” Oak said when I took a breath while explaining what Eventive was all about. “You’re saying they have detailed profiles of … everyone?”

I nodded. “In the States, at least, and a few other countries.”

“And they know enough about people to…”

“To basically mess with our heads. Prediction, nudging. Emotionally-effective messaging. They know us through the data exhaust we leave behind, the trails we generate every minute of the day.”

“It sounds like that scandal last year. What was it, Cambridge Analytica?”

“Except it actually works. CA promised more than it could deliver. Eventive delivers.”

“I can’t wrap my head around this.”

“It’s inevitable. Our phones track our movements, our connections, and practically our thoughts, and that data is for sale, along with what can be scraped from the web and bought from data brokers. Once you put all that information together and add in insights from fieldwork and massage it with artificial intelligence, it means Eventive knows more about people than their mother does, than their therapist. Than themselves. Eventive knows what we care about, what we fear, what it takes to persuade us of whatever it is you’re selling. That’s the product.”

“That’s terrifying.”

“It’s not just Eventive, either. It just does a better job of applying cutting-edge computing than most, and they provide clients with a front end that makes it easy to build campaigns.”

“Like, political campaigns?”

“Sure. Selling a politician or party, selling a new line of dill pickles, whatever you’re trying to do. It’s a tool for personalized persuasion at scale. I used to make pitch decks for sales calls. When I got to the part where I showed just how detailed our profiles were I’d watch their jaws drop.”

“No shit.”

“I lived for those moments.”

“And you didn’t … I mean, you never…” He was struggling to find the words.

“No. It’s hard to me to understand it now, but I was so excited about the code, about what we were building, about the challenge of going from a ‘what if’ scenario to rolling it out, I just didn’t think about how unethical it all was. How dangerous. Not for a long time, and then it was too late.”

Just for kicks, one of the engineers created a simple bot that acted like ELIZA, one of the earliest artificial intelligence programs written in the 1960s that mimicked a psychotherapist by simply turning statements typed into the program, like “I feel sad,” into questions: “how long have you felt sad?” or, when stumped, would simply prompt for more input with “I see. Tell me more.” Though the original ELIZA was created to demonstrate a central flaw in artificial intelligence–it doesn’t understand the words it’s using, it only knows how to construct a response by recognizing textual patterns–people who tested it out wanted to believe it worked. They felt a weird sort of intimacy and comfort with the non-judgmental box that made it feel safer than a real-life human interaction. That false intimacy worked great for people turning to Facebook or Telegram to vent about the state of the world. By injecting seemingly benign questions into a conversation among like-minded obsessives, it egged them on, leading them to make more extreme statements. These were ingested into the system, creating a cascade of emotional engagement in real time while simultaneously building up the cultural vocabularies of our training sets. It was simple code, but it never failed to generate a bountiful harvest of linguistic data that could be used to craft new messages to keep the pot boiling.

Could we automate some of the work of devising effective messages beyond basic A/B testing? Why yes, we could, once we hired an engineer away from Microsoft who had worked on natural language processing. He was able use an API to gobble up massive amounts of language scraped from social interactions. A small army of contract workers were hired to screen AI-generated statements during the development phase and tag anything that sounded weird. Those tagged messages then were fed back into the training set as patterns to avoid. The end result was to generate natural-sounding messages with a minimum amount of human intervention. Robbie was then able to integrate that feature into the product, so it could generate messages designed to match an individual’s personality and emotional triggers. That was another highlight of my sales pitches: it made it possible to scale up a campaign with highly-personalized emotional appeals at a much lower cost-per-conversion.

Things didn’t always work smoothly, of course. A lot of the off-the-shelf data from data brokers was inaccurate and needed extensive clean-up, which was expensive even after our data-set processing algorithms took a whack at it. Sometimes the automated language processing slipped from being just a little stilted to being disturbingly not-right. But the Eventive dashboard gave clients more than enough power to fool most of the people most of the time.

It was a thing of beauty, that dashboard. We had designers who customized the look based on inferences about a client’s tastes. Depending on how much they paid, the dashboard offered different capabilities. Basic clients could use it to identify individuals who would be likely susceptible to particular messages. If they paid a lot more, they could use the system to create and test AI-generated personalized marketing aimed at individuals. For a bit more, they had the capability to infiltrate closed groups on Facebook or Telegram or WhatsApp to scoop up data and increase buy-in.

Or for big bucks, mostly laundered through untraceable shell companies, our engineers created specialized datasets and algorithms supplemented by boots on the ground–field operatives or, in some foreign markets, subcontracted mercenaries with military training. This was an especially popular option among billionaire kingmakers who wanted to sway elections or authoritarian leaders who wanted to engineer their preferred reality and crush their enemies.

I didn’t know this, of course, not until later. All of my clients were licensing the less-expensive plain-vanilla products for reaching their customers. Aurora and Robbie met with the big spenders to pitch what Eventive could do for them through customization. Adam Barton himself worked with the most valuable clients, the ones whose names were never mentioned.