Greg Lindsay's Blog

January 02, 2019  |  permalink

WarGames, But For Real


(In 2017, my colleagues at the Atlantic Council’s Foresight, Strategy, and Risks Initiative housed within the Scowcroft Center for Strategy and Security asked me to write a fictional memo set in 2032 outlining how artificial intelligence was brought to bear on the National Security Council. My memo was a response to a short story by “Ghost Fleet” co-author August Cole envisioning an all-seeing, all-knowing national security AI named METIS. Rather than host presidential debates, future contenders would wargame worst case scenarios live on the Internet. I was challenged to backfill how Cole’s future came to be. Here is my memo to President Ocasio-Cortez as she prepares to win her second term…)


Subject: Faster than the speed of thought
From: GLindsay, White House Press Secretary
September 7, 2032

Ma’am, with the IVN news network’s “Presidential Leadership Live” simcast airing tomorrow, you asked for a summary of how we got here, where we’re headed, and how we intend to communicate your vision to the American people.

As you know better than anyone, METIS – the Multidomain Enhanced Thinking Insight System – represents the culmination of an AI revolution twenty years in the making–one that was not without its missteps and tragic, unintended consequences. But the bottom line is that we succeeded in radically streamlining your predecessor’s bloated NSC, scrapped the fatally flawed interagency process–exposed once and for all during its response to the Bus Flu pandemic–and transformed the NIC into the Artificial Intelligence Council (AIC) to prevent intentional or unintentional human bias from polluting METIS’ recommendations.

The result is a national security apparatus capable of operating at, as you like to say, “at the speed of thought”–which is still barely fast enough to keep up with today’s AI-enhanced threats. It required a wrenching shift from deliberative policymaking to massively predictive analysis by machines, with ultimate responsibility concentrated in your hands at the very top.

The strategic implications of bigger and bigger data coupled with ubiquitous machine intelligence were apparent as early as the Trump administration. A precipitous drop in the cost of predictions led to constant probabilistic analysis being brought to bear on previously unexplored problems, including national security. In turn, as the capabilities of machine prediction started to outstrip human performance, they also began absorbing the role of analysts. By the time Zuckerberg took office in 2021, the IC-as-we-knew-it was hollowing out.

A new consensus emerged that three functions were critical in the era of the National Security AI: split-second decisiveness at the top, massive raw intelligence gathering at the bottom, and the correct predictive models in between. The last piece was the most important, because otherwise we’d be gambling nearly a trillion dollars on a machine sucking garbage in and spewing garbage out. It was the Germans who cracked it first.

In July 2020, the German Foreign Ministry unveiled its Vorhersagemaschine, the first modern NSAI system capable of building–and then learning from, in an endless iterative loop–thousands, or even millions of plausible scenarios for any given policy dilemma. Built primarily by the consulting group SAT, the prediction machine was expressly designed to tackle “wicked problems” of deep uncertainty–so deep that policymakers and planners can’t even agree on how to structure the problem. So, the Vorhersagemaschine does it for them a million different ways using what SAT’s consultants called “exploratory modeling and analysis.” After the machine generates its scenarios, the ministry’s team of data scientists, visualization specialists, and policy experts drill down and explore variations on their decisions.

It was thanks to this approach that the ministry convinced Chancellor Merkel to abandon her stance of “Wir schiffen das” at the outbreak of the Iraqi Civil War, instead choosing to close the borders and assist in creating safe zones in Turkey after the Vorhersagemaschine suggested a 90 percent probability of the Alternative for Germany party (AfD) winning a majority of seats in the next Bundestag elections.

But massively predictive approaches contained their own hazards, as the Zuckerberg administration discovered to their lasting regret. As the NSAI’s predictive accuracy rapidly improved through unsupervised machine learning–which the president had endorsed given his successes at Facebook–his national security team couldn’t resist the temptation to act pre-emptively on the basis of its prediction, rather than wait for events to unfold. After winning their internecine struggle with members of the “reality-based community, they began rubber-stamping its recommendations to forward deploy assets – effectively thinking several moves ahead, or so they thought.

While it’s one thing for Amazon to ship our packages before we’ve even thought to order them, it’s another to dispatch carrier battle groups to contested waters without knowing how the algorithmic black box on the other side might interpret them. The ensuing Spratleys Crisis touched off the current Sino-American Cold War–starting with the crippling economic sanctions that sank Zuckerberg despite owning the greatest propaganda machine in history.

It was a similar “mistake” that opened the door for you, Ma’am. The first Bus Flu pandemic in 2026 was the world’s worst since the Spanish Flu more than a century before. After tracing it back to China’s Unit 21, your predecessor’s re-election seemed assured. That was before The New York Times noticed vaccine shortages in specific majority-minority communities–all in toss-up Congressional districts.

Ugly terms like “ethnic cleansing” were thrown around until the press settled on “genemandering” (i.e. genetic gerrymandering). Upon investigation, it became clear that someone – we may never know at who’s direction – deliberately introduced bias into the joint DHS/HHS AI handling the crisis. Besides the obvious electoral implications, the scandal permanently soured the public on black box algorithms and underscored the dangers of what a singularly motivated actor could do.

One of your first acts as president, of course, was signing into law the bi-partisan AI Transparency Act of 2029, which required federal agencies to use “clear-box” models that publish their rules as they go along. This doesn’t apply to classified systems, obviously, which is why we added a Federal Reserve-style board of governors to the NIC (now the AIC) to vet the rules generated by METIS and its predecessors. (If you’ve noticed METIS is acting more hawkish than usual, it’s because they voted last week to loosen the reins a bit.)

The Bus Flu pandemic also exposed a literally fatal lack of data for sensing and thwarting epidemiological weapons of extreme consequence (WECs). Just as 9/11 prompted the Patriot Act, the DHS and DNI, and “Total Information Awareness,” one of the many provisions of the “Shane Doctrine” declared by the prior administration required that the NSA’s backdoors into the Big Five and other tech companies became front doors.

Controversial to say the least (and heavily resisted by Silicon Valley’s lobbyists), these proposals were held up in court for years until Director Rhodes hit upon the idea that we should simply pay Americans for the privilege of surveilling them. Starting from MIT professor Alexander Pentland’s “New Deal on Data” proposal for the World Economic Forum, we won over a deeply skeptical electorate by restoring personal ownership of their data exhaust. Then we offered a blanket license to every man, woman, and child living in America, along with e-citizens overseas.

Making the American people and their digital simulacra a line item in the defense budget was one of your most inspired ideas–which we sold to the public as a “peace dividend” at the official end of the War on Terror. From there, it was a fairly simple matter to build on work by Pentland, West Point’s Network Science Center, and many others to build the underpinnings of METIS.

The final piece was the simsuit. DARPA had been toying with exoskeletons for decades, and the first gaming suits hit the market in the late 20-teens But just as the U.S. Navy researched heads-up displays to lessen the mental burden on pilots during combat, our challenge was to somehow increase your cognitive bandwidth before METIS resorted to using PowerPoint slide decks.

One solution was haptic, using WearWorks’ touch lexicon to communicate feedback from METIS But the real breakthrough was the biolink–the science of which I don’t truly understand beyond what the techs call the “Dobelle Effect.” When you put on the helmet, you don’t actually “see” anything–it sends signals straight to your brain to trigger phospenes at levels that never appear in nature. That’s how METIS gives you a God’s-eye view.

So that’s where we are. METIS is the logical response to a world in which wars are now fought with drone swarms directed by repurposed high-frequency trading algorithms–replacing NSC meetings with fast-twitch gaming. Speaking of which, Steam and Twitch have agreed to host the next round of simcasts on 21 September.

Break a leg, Madame President. I hear Arkin was a decent motocross racer back in his day. But he’s never driven anything like this.


Posted by Greg Lindsay  |  Categories:  |  Comments

About Greg Lindsay

» Folllow me on Twitter.
» Email me.
» See upcoming events.

Greg Lindsay is a generalist, urbanist, futurist, and speaker. He is a 2022-2023 urban tech fellow at Cornell Tech’s Jacobs Institute, where he leads The Metaverse Metropolis — a new initiative exploring the implications of augmented reality at urban scale. He is also a senior fellow of MIT’s Future Urban Collectives Lab, a senior advisor to Climate Alpha, and a non-resident senior fellow of the Atlantic Council’s Scowcroft Strategy Initiative.

» More about Greg Lindsay


December 01, 2023

“The Age of Principled AI” (Video)

November 28, 2023

Fast Company & Curbed: Cars broke Los Angeles. Could a new form of transit fix it?

November 14, 2023

Have Deck, Will Travel: Fall 2023

November 13, 2023

Welcome to the Age of “Principled AI”

» More blog posts

Articles by Greg Lindsay

-----  |  August 3, 2023

Microtargeting Unmasked

-----  |  July 1, 2023

2023 Speaking Topics

CityLab  |  June 12, 2023

Augmented Reality Is Coming for Cities

CityLab  |  April 25, 2023

The Line Is Blurring Between Remote Workers and Tourists

CityLab  |  December 7, 2021

The Dark Side of 15-Minute Grocery Delivery

Fast Company  |  June 2021

Why the Great Lakes need to be the center of our climate strategy

Fast Company  |  March 2020

How to design a smart city that’s built on empowerment–not corporate surveillance

URBAN-X  |  December 2019


CityLab  |  December 10, 2018

The State of Play: Connected Mobility in San Francisco, Boston, and Detroit

Harvard Business Review  |  September 24, 2018

Why Companies Are Creating Their Own Coworking Spaces

CityLab  |  July 2018

The State of Play: Connected Mobility + U.S. Cities

Medium  |  May 1, 2017

The Engine Room

Fast Company  |  January 19, 2017

The Collaboration Software That’s Rejuvenating The Young Global Leaders Of Davos

The Guardian  |  January 13, 2017

What If Uber Kills Public Transport Instead of Cars

Backchannel  |  January 4, 2017

The Office of the Future Is… an Office

New Cities Foundation  |  October 2016

Now Arriving: A Connected Mobility Roadmap for Public Transport

Inc.  |  October 2016

Why Every Business Should Start in a Co-Working Space

Popular Mechanics  |  May 11, 2016

Can the World’s Worst Traffic Problem Be Solved?

The New Republic  |  January/February 2016

Hacking The City

Fast Company  |  September 22, 2015

We Spent Two Weeks Wearing Employee Trackers: Here’s What We Learned

» See all articles