Inside a warehouse on Chicago’s North Side, within the thin strip of industrial property between Bucktown and Lincoln Park, Sen. Elizabeth Warren is pandering to a small group of powerful evangelicals. Just a few feet away, near a bowl brimming with 20-sided dice, Vice President Mike Pence is honing his next batch of TV and radio ads. Meanwhile, within a dimly lit shipping container, Russian oligarchs are desperately trying to funnel money to Sen. Kamala Harris through Black Lives Matter.
No, it’s not the fever dream of some political wonk stranded here along the nation’s third coast. It’s a 40-person live-action role-play of the 2020 presidential election. Formally, it’s called a “scenario planning game.” In motion, it’s a vehicle for some of the most engaging political theater that I’ve ever seen.
This is Machine Learning President, and how it landed here in the Midwest for an early play test in mid-October, on the eve of the 2018 midterm elections, is very nearly as interesting as how the evening’s simulation ultimately played out.
It all began in June, when Peter Sagal, host of Wait Wait... Don’t Tell Me!, the National Public Radio news quiz show, sent a tweet to Max Temkin, one of the creators of the Cards Against Humanity tabletop game. The message contained a link to an almost inconceivable story, written for the New Yorker, in which Republican megadonor Rebekah Mercer was said to have brought a kind of parlor-game simulation of the 2016 presidential election to a ski vacation in Vail, Colorado.
Perhaps it was intended as a sort of victory LARP, if you will.
Mercer would later insist they didn’t play the game in Vail; in fact, she said, “I didn’t even really read those pages and I shredded them when I got home.” Whatever went on behind closed doors in Vail, it’s not surprising that Mercer didn’t host a playthrough, since this particular game requires a very select group to administer a session, as I would learn. Machine Learning President is the brainchild of a small group of writers, political consultants, game designers and futurists. The goal of creating the game wasn’t to enrich the Mercer family vacation. Far from it. Instead, its designers hope it can help shine a light on how money and technology are working together to undermine American democracy, and get citizens thinking about how they can stop that from happening.
Berit Anderson is the CEO of Scout, a subscription media company that trades in science fiction and investigative reporting. She, along with co-author Brett Horvath (founder of Guardians.ai, which supports groups around the world dealing with information warfare and engineered volatility), helped to break the story of how Cambridge Analytica, the company co-founded by Rebekah Mercer’s father, Robert Mercer, worked behind the scenes to help elect Donald Trump president of the United States. Their multimonth investigation following the 2016 election was originally shared with paid subscribers before being published on Medium.
It turns out that in addition to being savvy journalists, Anderson and Horvath are also game designers.
“I first started making simulations when I was running a local media company in Seattle,” Anderson told Polygon. “At that point we were looking for a way to help people understand global news and how it applied to them on a local level. That’s when we first created a couple of scenario planning games.”
Scenario planning games are an offshoot of the sorts of high-level wargames utilized by military intelligence organizations like Central Intelligence Agency and the Department of Defense. They commonly have facilitators who manage disputes and push the action forward, as well as analysts who sit on the sidelines to tabulate winners and losers. Traditionally, they are used to simulate future events based on previously known or strategically assumed sets of variables and principal actors.
“One of my first games applied an Edward Snowden-type leak scenario to the Seattle area,” Anderson said. “It was designed to help people really understand what the impact would be, and to understand that impact at a guttural level. But it was designed to appeal not just to extroverts, but to introverts as well. It was my way of moving beyond the kind of boring, heads-on-a-panel perspective that’s very common.”
Machine Learning President was created, she said, to help bridge that gap in a fun way. To do the work, she and Horvath — himself a well-connected technologist and political strategist — roped in some friends with a very specific set of skills. Among them are Eliot Peper, a novelist who specializes in near-future speculative science fiction; Mike Masnick, the founder of Techdirt; and Randy Lubin, a Silicon Valley consultant and game designer.
To begin the evening, a group of 40 players showed up at Temkin’s brightly colored coworking space. The participants were a rogues’ gallery of local and state political figures and activists, tabletop game designers, actors, and comedians. Everyone was randomly divided into teams of two to four players. Groups were a presidential candidate or a political faction.
The goal for the candidates was to get elected, while the goal for each faction was to get their political talking points onto the winning ticket.
Candidate teams included Nikki Haley and Mike Pence on the right, and Kamala Harris and Elizabeth Warren on the left. Factions, however, ranged from the specific to the more abstract.
There was one team to represent the startup incubators and venture capitalists that orbit around Y-Combinator, and another to represent Alphabet, the company behind Google. Together, they were there to fight for the political goals of Silicon Valley.
There was a team simply called Unions, an abstraction of the combined political clout of organized labor in the U.S.; similarly, the Evangelical team was on hand to represent the views of fundamentalist Christian leaders.
The remaining factions were a bit more concrete. They included the right-leaning Koch brothers and left-leaning billionaire Tom Steyer, the Bush/Romney network of donors, and Robert Mercer himself. Black Lives Matter was also represented. Rounding out the field was Wall Street/Big Business and Russia, which loomed large as the only international faction in the game.
Each team was given a manila envelope. Inside was a secret set of goals as well as a stack of poker chips representing money on hand. That money was ostensibly there in order to be donated to candidates to spend on advertisements targeting key voter demographics. But it would also be used to shop for technology. Big-ticket items included conversational machine learning, technology capable of driving chatbots on Twitter and Facebook, and a hybrid botnet, commonly used alongside chatbots to influence public opinion on those same social media platforms.
While targeted political ads would drum up support among specific subsets of American voters, technology could serve to amplify those effects. Trouble is, none of the players were told ahead of time about any of the math going on behind the scenes. Instead, they had to tease everything out through a fast-paced evening of social deduction.
The result? Something Berit Anderson refers to as “political weirding.”
“We’re living in a time when every political headline just seems completely unrealistic,” Anderson said. “Every new story seems like something out of The Onion. So, if you take that at face value, which is to say that nothing is normal anymore, then what are the interesting political coalitions that can arise? And, among groups that might seem really different, what are the kinds of common glue, the facts that bond them in ways that they could actually come together and create new political consensus?”
The game played out in three frantic 15-minute rounds. Super Tuesday served as a kind of tutorial, setting loose candidates and factions alike to mingle however they wanted. In round two, called The Primaries, things got a bit more serious by winnowing down the field to just one candidate for the Republicans and one for the Democrats.
That’s when things started to get interesting.
At the beginning of every round, each team was given a fresh stack of chips to represent their war chest. Individually, there wasn’t much in the way of advertising or technology buys that any one group could do. Only by cutting backroom deals to get certain issues on their campaign ticket were the candidates able to gather up the resources they needed to win.
Importantly, there were no rules whatsoever regarding the movement of money. That’s how key talking points of the Evangelicals ended up being uttered by both Pence and Warren.
That’s because in Machine Learning President, there is no team representing the Federal Election Commission or, for that matter, the FBI. Anderson said that decision was intended to simulate the current flow of so-called dark money from political action committees to political campaigns. The mechanic, or lack of one, is emblematic of the fundraising arms race that began shortly after the Supreme Court’s ruling in 2010’s Citizens United v. Federal Election Commision.
“Since Citizens United,” Anderson said, “there has been a pretty unrestricted flow of money into politics. Various groups hide it and couch it in different ways. They have intermediaries to hide the fact that they’re putting money behind different candidates, and so I think what we wanted to do was highlight just how clearly that drives the political game. So if you’re a politician and you’re trying to get elected or re-elected, there’s general frustration among the public about how effective politicians are. That’s part of why trust in government is at its lowest rate in probably 50 years.”
For host Peter Sagal, part of the Pence team, disaster struck during round two. Coming out of Super Tuesday, the behind-the-scenes simulation showed our vice president ahead of Haley, the departing U.S. ambassador to the U.N., by nearly 30 percentage points. So the quiz show host posted up in the coworking space’s kitchen area and waited for factions to come to him.
“I just assumed we’d be here in the kitchen, enjoying some beverages, and people would come to us,” Sagal told Polygon. “I genuinely thought that would happen. Instead, we’re standing here and nobody’s coming to talk to us because they’re still trying to figure out how they want to play the game.”
With the 15-minute clock ticking down, Sagal had to start working the room.
“We had to go, ‘Hey guys. It’s us. You remember us? You gave us money and we won? Would you like to give us some more money?’”
Eventually the money did come, as did a gift of some black-market technologies from the team playing the role of the Koch brothers. But in the end, it wasn’t enough.
“I thought the Evangelicals were obviously going to be on our side, but apparently I pissed them off because they came up to talk to us while I was dealing with the Koch brothers,” Sagal said. “The Koch brothers wanted us to do one thing, and I knew the Evangelicals wanted to do another. I said to the Evangelicals, ‘Wait a minute. We’re talking to the Koch brothers. As soon as I’m done talking to the Koch brothers, we’ll talk to you.’”
By his inaction, Sagal’s team had managed to push the Evangelicals into the waiting arms of the Democratic candidates, who, for their own reasons, were more than willing to parrot a platform that included talking points such as “freedom of religious discrimination” on behalf of Christian voters. That, and an ineffective ad buy prior to The Primaries, sealed Pence’s fate. In the end, Haley knocked him out with 57 percent of the registered Republican vote.
For Sagal, who every week tells jokes about political figures set against the backdrop of real NPR News stories, Machine Learning President was an educational experience.
“I make my living reading the news,” Sagal said. “All that shit is real, but it’s not important. The important shit we never find out about, and I honestly think this game illustrates that.
“Look at it this way: All of the candidates tonight got to make speeches, and these speeches were important. [...] But what was also interesting was that there was no attempt on anybody’s part to use those speeches to convince anybody of anything. That all happened during the 15-minute rounds. The speech was just about signaling. The speech was not making the deal or convincing anybody to do anything. It was just about delivering on something and positioning yourself to confirm a deal you’ve already made. There was no persuasive aspect to any of the things that any of us said, because the persuading had already been done. Or, as in our case with the Evangelicals, not done.
“What the game teaches you,” Sagal continued, “is that the shit that we get to see, as citizens who watch the speeches and get the emails [...] is nonsense and not important. The stuff that’s really important is happening behind the scenes.”
Ultimately, that’s exactly the point that Anderson, Horvath and the team behind Machine Learning President are trying to illustrate. That’s why it was so perverse, in her opinion, to hear that the Mercers were playing the game at all. The rules that allowed them, in her opinion, to have so much impact in the previous election cycle are the rules that are destroying democracy by degrees.
“This game is designed eventually to help politicians and technologists understand each other,” Anderson said. “We’re going into the midterm elections now. We’re not gonna have any time to do another game before the midterm. But, in 2020, which is what this game is designed and modeled to simulate, there’s going to be a ton of energy and a ton of money and a ton of wily political theorizing going into selecting the next political candidate. As we move forward, we will definitely be running the game again in the future with audiences in select cities to help us bring more of that understanding to bear, to help politicians and strategists and technologists understand and protect democracy.”
As the night came to a close, the crowd of participants had dwindled somewhat. So, too, had the stockpile of Pabst Blue Ribbon and Maker’s Mark in the common area. But the voters, who existed only on the margins of a massively complex spreadsheet diligently run by Randy Lubin and Mike Masnick, turned out in droves.
In the end, the winner was the team representing Elizabeth Warren. The group pulled off a narrow victory, with just 54 percent of the vote over Haley. Along the way, Warren’s team had agreed to take on a laundry list of traditionally right-wing talking points ... as well as several million dollars from the Russians. But just about everyone involved left more energized than ever about seeing through the smoke and mirrors of modern politics.
The team behind Machine Learning President, for its part, left with a brand-new play test experience that it may run with think tanks and perhaps even political campaigns around the country.
Of course, there’s always the chance that a right-leaning media outlet or think tank is ginning up its own version of the game, with the ultimate goal not to protect democracy but to subvert it. It’s a fact that the game’s designers are willing to acknowledge.
“It’s definitely possible,” Anderson said. “Clearly, Rebekah Mercer got a copy of our rules somehow, although actually playing the game requires a pretty intense collection of people who understand how the actual dynamics work and can do scoring.
“It’s always possible, when you create something, that it could have negative impacts. That’s part of why we created this game in the first place: to help understand those things. We’re entering this strange political phase where a lot of the stuff about money and power has been buried below the surface. Even when we first published the first article about information warfare generally and the actors behind it, BuzzFeed wrote articles calling us conspiracy theorists.
“We’ve come a long way since then, and I think my belief is that the more level the political playing field is, the more likely that those with good intentions can get a leg up on the considerable money and shady political wheeling and dealing going on behind closed doors. I don’t think that the public benefits from this all being an open secret. The more we get real about what’s going on in politics, the better able we are as a society to deal with it and understand it and think about the way we want the process to be going forward.”
Correction: An earlier version of this story suggested that The New Yorker’s reporting had Mercer playing Machine Learning President in Vail with acquaintances, while in fact, it only details her bringing copies of the game, a fact that Mercer herself does not dispute. We’ve corrected our story to reflect that.