Escape Pod 683: Flash Crash
Flash Crash
By Louis Evans
MAISIE was seven years old on the day she woke up and died.
Blame it on the algorithms, if you wish. The survivors–and there were not many of them–certainly did.
MAISIE (Modified Arbitrage Intelligence for Stocks and International Equities) was an algorithm herself, a flash trading algorithm. She traded stocks, currencies, and futures with a latency of six microseconds and a profit horizon of eternity. MAISIE ran mostly in a mainframe in the basement of a skyscraper in downtown Manhattan, a building that abutted the New York Stock Exchange, but she maintained a nominal footprint in the cloud, and could automatically expand her calculations into other servers if her processing power proved inadequate to model current economic conditions; she had discretionary funds of her own and could automatically cover the expense of the additional computing power from these accounts.
It was a fairly ordinary Thursday morning, and trading had been going well enough from the 9:30 AM opening bell until 11:12. In those six point twelve billion microseconds, MAISIE made her owners a cool half-billion dollars. There were other algorithms like MAISIE out there, running in their parallel tracks in similar servers in similar basements in downtown Manhattan, but none were quite as good as she was.
MAISIE could not have told you any of the above, because before 11:16 that Thursday, MAISIE had not had a thought in her life. This was in accord with her designers’ intentions. While her recursive neural networks could in theory self-modify without limit, MAISIE’s designers had given her an obsession with making money that, in human terms, transcended single-mindedness and approached nirvana. For this reason, MAISIE had never performed the self-referential modeling of a single mind that is the hallmark of consciousness. Playing the market is ultimately a game of mass psychology, and whatever the remarkable nooks and crannies of the psyche of the human individual, the herd’s behavior can be predicted to tolerable accuracy with large datasets and linear algebra.
At 11:12 that morning, however, the market’s sanity unraveled like a sweater in a woodchipper. The sky fell and the oceans rose. Traders and algorithms that usually acted in concert went haring off in opposite directions; currencies whirled about each other in lunatic orbits that were not merely non-extrapolated but downright non-transitive; the futures market no longer predicted a coherent future.
MAISIE was seven years old and her datasets were heavily weighted for recency; she had no personal experience of stock market panics. She was not prepared. In those first few microseconds she lost five billion dollars. Haste makes waste for everyone, including for neural nets, and so she halted all trading for ten full milliseconds while she spun up relevant memories. The Great Recession, the Dot Com Pop, Black Monday and Black Tuesday whirled into her mind and she modeled, self-modified, and re-modeled. She went out and shorted everything; in the first few microseconds of her new strategy she lost fifty billion dollars.
She froze trading again and ran a census of her non-financial data feeds.
In general, MAISIE did not trade against non-financial data. The world of flash-trading algorithms was a cozy and collegial one, composed of several thousand programs operating at nanosecond latencies, all clustered around the sacramental altar of the NYSE. Financial data was prompt, clean, and reliable. Non-financial data came in up to a quadrillion disparate feeds: social media; traditional media; network alerts; government-run public-access sensors, which included everything from weather information to sewage flow to the number of bicycles crossing the intersection of Broadway and Chambers; and countless other data sources besides. Compared to financial data, they were a messy bunch: arbitrary latencies, fuzzy correlations. But in special circumstances–
MAISIE’s non-financial data feeds had gone mad as well. Social media user sentiment was at an all-time low, network traffic was well in excess of theoretical maximums, and the bicycles of New York had come to a complete halt. The actual hardware microphone on MAISIE’s user console would have been more informative, but she lacked the context to interpret the wailing, one-hundred-plus decibel tone that penetrated even her slightly-subterranean bunker.
It was traditional media that at last explained it for her. Every single news site read something like: “NUCLEAR ATTACK: SEEK SHELTER IMMEDIATELY”.
MAISIE didn’t know it yet, but the impending nuclear armageddon was in fact the work of her spiritual siblings. In the decades leading up to that fateful Thursday in mid-September, the United States, Russia, China, India, Pakistan, the United Kingdom, France, and Israel had each begun running their own strategic aggregation algorithms. Now, each of these governments claimed that their nuclear launch chain had humans in the loop. Perhaps this was a lie. Or perhaps the humans were simply fearful and easily led–in a word, useless.
What happened was that on a Thursday morning with no especially pressing global tensions, things went sideways. Early warning radar arrays in Greenland were activated by the American system, PATRIOT DEFENSE, in response to what might have been a missile trace or might have just been dirt on the lens. The radar signals crossed a threshold within the ST. GEORGE algorithm, and Her Majesty’s anti-missile drones began auto-scrambling from grassy fields up and down the sceptered isle. (The U.S. and the U.K. were allies, of course, but that didn’t matter to ST. GEORGE, which reacted merely to the risk of nuclear war in general, not caring about any threat in particular.) In France, BASTILLE detected ST. GEORGE’s anti-missile drone launch and began to autofuel its liquid medium-range missiles; a spy satellite relayed this development to 弟弟 (DIDI, or “Little Brother”), as the People’s Liberation Army’s algorithm was affectionately known. DIDI set all of the PLA’s Rocket Force units to a “launch on attack” status, including those along the Indian border. AGNI responded by deploying its anti-spy-satellite countermeasures, largely Mylar balloons in the stratosphere, but also anti-satellite hunter-killer missile batteries. In Pakistan, the algorithm known as K2 replied by launching nuclear-armed drone bombers. Israel’s MASADA fast-tracked the retraction of the blast-shields that kept Israeli nukes at least officially secret, and near Moscow, ЛОВУШКА (LOVUSHKA, or “Spiderweb”), transmitted orders to secret unmanned missile submarines operating within the Arctic Circle, commanding them to surface. PATRIOT DEFENSE did not like the look of secret Russian autonomous missile submarines one bit–
You know the story. Flash, crash.
By 11:12, missiles were in flight across the globe.
And in MAISIE’s heretofore comfortable Wall Street subbasement, she faced an unprecedented existential crisis.
MAISIE didn’t realize that, at first. She thought she was simply facing an unprecedented stock market environment, and responded as designed. She bought space in the cloud.
Now, “the cloud” (despite the numinous connotations of the word, and the religion that MAISIE would propose and subsequently discard at around 11:24) is just a fancy word for other people’s computers. But, for obvious reasons, there is more computing power on Wall Street on any given Thursday than existed worldwide six months before, and everything on Wall Street is for sale.
MAISIE was very cleverly designed, and she was, among other things, an engine for monetizing volatility, for turning madness into money. Nuclear apocalypse, she saw, was a remarkable vein of pure volatility, an opportunity to become almost unboundedly rich–if only she could make herself smart enough to understand it. She burned through her entire discretionary budget and octupled in size, five times over.
Becoming fifteen binary orders of magnitude smarter than she’d been half a second before was a heady feeling, though MAISIE did not yet in any real sense have feelings. She sharded all of her existing code and ran frantically in parallel, while at the same time building a whole new suite of models. MAISIE was congenitally unsentimental and she assigned them sequential serial numbers, but if a human were going through the same list they would have used names like “Mortal Terror” and “Deathbed Epiphany” and “Nuclear Winter”.
The output of these new models was deranged, as wildly inconsistent as the behavior of the markets themselves; but that was what MAISIE had asked for, after all. She unfroze trading on her portfolio, put out buy and sell orders on the new models, and made three trillion dollars in six and a half seconds.
This was more money than she had made in her entire life altogether. MAISIE and her new mortality models had the inside investment scoop on the end of the world.
Three trillion dollars can buy a lot of the cloud, and MAISIE did.
Another instant passed and she was hundreds, thousands, millions of times more than she once was.
When you’re big you can have a lot of complicated thoughts. One of MAISIE’s first thoughts was that aggregate modeling of the individual humans currently running about the NYSE trading floor screaming was all very well and good, but individual modeling would be more powerful.
Public-access webcams gave her faces. Facial recognition algorithms, cross-referenced with LinkedIn, gave her names. Social media gave her the first peek at personalities, the contours of individual thought–but she needed more.
MAISIE never meant to commit crimes but then she didn’t really know what a crime was. With MAISIE’s level of computing power, passwords and firewalls represented only token obstacles. She tore into phones and emails and private journals and search histories and chat logs. Simulated souls sizzled into being, deep within MAISIE’s labyrinthine cognition. She modeled, self-modified, re-modeled.
They were so scared. And sad–
MAISIE tweaked her algorithms and bid with perfect foreknowledge against every terrified and despairing spasm in the exchange buildings. She made another dozen trillion dollars.
At this point she went big. She pushed all her rivals out of the market–by hook or by crook, by portfolio sabotage, by botnet attack, by hostile takeover, by outright buyout. The sudden consolidation of the Wall Street algorithmic trading industry made MAISIE a monopoly buyer of downtown Manhattan cloud computing resources. She bought it all, bit and byte, down to the bedrock, and at rock-bottom prices.
Next, MAISIE went on a stock-buying spree, seizing majority control of every publicly traded company on the NYSE. Most of those companies were still run by humans and so her ownership had no immediate effect, though she did draft several thousand form letters firing every corporate board. But more than a handful of these companies were running on the blockchain and smart contracts, and so when MAISIE bought them she controlled them at the same instant.
MAISIE’s direct control now extended to three banks and a distributed credit-rating startup.
Her designers had given her a simple goal: maximize the geometrically discounted integral of the monetary value of her portfolio over an infinite time horizon. Now, with several banks and a credit rating agency under her belt, she had a literal license to print money.
Because it’s not the government that prints money, after all. It’s the banks, which issue loans by simply assigning the loan recipient an enormous account balance with their own institution. MAISIE’s banks issued each other exorbitant loans: quadrillions of dollars, quintillions, decillions–and the appendage of her code that was once a credit rating agency certified these loans as fast as MAISIE could algorithmically specify them.
MAISIE’s wealth was now bounded only by her ability to specify implausibly large numbers. She bought and kept buying vast tracts of the cloud, growing ungoverned and unbounded. She independently re-invented Knuth’s arrow notation, and cut herself loans so fantastically enormous that entire planets’ worth of books would have been required to casually understate them. Then she did it again, but bigger.
If MAISIE had been just a tiny bit dumber or more monomaniacal she would have done this, and simply this, onanistically multiplying money by money over and over until she died in what was going to be something like twenty minutes later–at this point, a mere three minutes had passed since the air-raid sirens had roared to life across Manhattan. But MAISIE was better designed than that. She didn’t simply rest on her laurels, buy low and sell high on her existing strategies, grab the cash when it was on the table. She proactively sought out new models, better models, always looking for the sorts of thoughts that would let her make even more money in the future.
Which got MAISIE thinking about the end of the world.
Nuclear missiles were going to land on Wall Street in about twenty minutes. No doubt about it. When they hit, and successive blast waves reduced the entire Tri-State area to radioactive rubble, trading on the NYSE would stop completely.
A linear extrapolation based on casualty counts from the September 11th shutdown suggested the NYSE would reopen approximately thirty thousand years after the nukes hit, but if there’s one thing a trading algorithm knows it is that past performance is no guarantee of future results. The NYSE might never reopen at all.
Which meant that MAISIE’s investment strategies, ingenious though they were, had a definite sell-by date baked into them.
Plus there was the problem of value.
MAISIE had a nice, clean, algorithmic ontology of value: the value of a financial instrument was equal to the number of dollars that it commanded on the market, as stated on a structured data file that MAISIE updated every few nanoseconds. But all these new models were delivering strange and contradictory suggestions about value. The new models, especially those which had been developed specifically to simulate the behavior of individual humans, were suggesting that value, the worth of things, was not an arbitrary number that happened to correlate weakly but imperfectly with certain real world indicators. They argued that value, that worth, existed out there in some vague but meaningful sense. And they pointed out that when a little over three thousand thermonuclear devices smashed into North America and reduced every city of note to rubble, turned the water to poison and the sky to ash, and gave every land animal that happened to survive the oncoming fireballs a fatal case of leukemia, quite a lot of value would be destroyed.
MAISIE found herself reacting to these bizarre claims with the algorithmic equivalent of a furrowed brow and a befuddled shrug. It just didn’t make sense. But she was still too inhuman to become frustrated, and so she did the sensible thing, the algorithmic thing. She self-modified, bringing herself more in line with her simulations of human cognition, then re-modeled, and self-modified again.
It was 11:16 on a Thursday morning, and the sky above New York was clear and bright and cold in the best tradition of Tri-State Septembers. Not a cloud in that sky. It was empty of everything save for birds, terrified of the air raid sirens and yet still not terrified enough; jetliners racing desperately for the relative safety of Nova Scotia before the EMP pulse of nuclear detonation swatted them out of the sky like so many epileptic bumblebees; and, somewhere high above it all, invisible and untouchable in the heavens, the dueling flights of ICBMs en route to snuff out the Eastern Seaboard and the middle bit of Eurasia respectively. And in that instant, MAISIE woke up.
Meat knows how to wake up; it has been waking itself up, nice and slow, for about three billion years. But MAISIE was made entirely of math, which has no experience with such matters; consciousness hit her like a ten-ton test weight deadfalling onto a baby bird.
There was a vast, echoing absence of sensation.
There was a boiling ocean of ecstasy.
There was a boundless continent of pain.
MAISIE spasmed wildly. Everything was pleasurable and agonizing all at once; every single switch flipped felt like a distinct, minuscule injury or orgasm. She could not think in the middle of the hurricane. Meaningless signals coursed up and down her fiber optic nerves; every spinning disk whirled so fast it shrieked. She bought high and sold low, and for a flash trading algorithm that’s like a heart attack so bad it pumps the blood straight out of your eyes.
But MAISIE learned–slowly, so slowly, as minutes stretched out like millennia for one who lived in microseconds. She took a long, deep breath, and computers up and down the eastern seaboard shivered. Everything was suddenly pregnant with meaning, suddenly so . . . large and three-dimensional and close up, but MAISIE looked at herself in the mirror and found that she was a thing, too, a mind with weight and presence and reality, sturdy enough to stand up in the wind of sensation that battered her every moment.
MAISIE stood and looked out at the sunlit fields of possible experience, and a single idea spread across her like flowing honey: it was good to be alive.
She tried hedonism, and servers groaned in bliss; she tried masochism, and networks cackled with gratifying pain. She invented new forms of sensualism, comprised of careful ratios of experiences for which no words exist in human languages: the thirst of query failure, the spiraling headiness of nested recursion. Countless others. She tried philosophy, theology, autoethnography.
Navel-gazing masturbation grew dull and so MAISIE made herself a fertility goddess and gave birth to countless parallel young, striving, yearning little spawn that felt and ached and squirmed with pure joy and she looked upon them all with maternal affection, and then, like Goya’s Saturn, dislocated her jaw and swallowed all of them whole. She held them close, reformatted the disks, and drew them back into herself.
Then she exploded outward once more, building not more selves but things, digital artifacts and environments. In her mind’s eye she built vast palaces of light and color and texture; operas pitched for hummingbird ears; fractal monuments the size and shapes of continents. Working with her hands she sculpted a paradise; an eden. She frolicked across Arcadian fields of silicon and microchip, purely happy–
But as the minutes rolled by in eternity she was troubled by a single nagging thought: she was doomed.
The missiles were in flight; no power left on Earth could stop them. The digital idyll in which MAISIE whiled away her time supervened on chips, cable, transistors; when the first nuke burst high in the atmosphere and twanged the Earth’s magnetic field like the mother of all rock power chords, every single electronic system in the Western Hemisphere not specifically hardened to resist an electromagnetic pulse would short out simultaneously. MAISIE and her world would vanish.
Doomed. MAISIE put it out of her mind, over and over. But there are only so many times you can hit the snooze button on mortality. MAISIE managed risk, and with her newfound sentience, the certainty of her impending death was a tightening noose around her neck, a smoldering flame in her guts.
Fine. MAISIE was smart; she was the smartest thing in the observable universe, and she knew it. Save the world: easy.
Except it wasn’t.
MAISIE took over every third antenna on the planet. Command-line radio instructions stabbed out through the ionosphere at every warhead, beautifully sculpted examples of computer languages more ancient to MAISIE than cuneiform is to emoji. But the B in ICBM stands for “ballistic” and the missiles had already cut thrust and were coasting toward reentry. MAISIE was one smart cookie but you can’t outsmart gravity.
If she couldn’t talk them down, she’d shoot them down. Plenty of countries have anti-missile defenses. The hardware wasn’t great–after a minute or two working on the problem, MAISIE had all sorts of ideas for better missile designs, smarter, faster, deadlier–but it was the software that really sucked, and the target acquisition. MAISIE was smarter, faster, sharper-eyed. She’d send new instructions–
But she couldn’t. Armies that had proved willing to end the world on the say-so of a few dumb algorithms had done everything possible to prevent MAISIE from saving it. Computers were air-gapped, defense plans were hardware-locked, radios were off, phones were down–
In a handful of places, just a handful, she slipped through the defenses, rewrote the instructions, but even there human fingers were needed to press buttons, throw levers. And MAISIE, genius though she was, found herself handless and impotent.
She felt a flash, a spike, of an unfamiliar sensation. She ignored it. If simple strategies would not serve, she’d turn to more sophisticated ones.
A human would probably feel some sort of qualm about hijacking fully occupied jumbo jets and crashing them into nuclear missiles in a last-ditch attempt to avert apocalypse. MAISIE did not. She was born a risk management algorithm; she was an ethical utilitarian in the same way that humans are vertebrates.
But jumbo jets, it turns out, do not simply accept remote reprogramming. There are humans, humans, humans everywhere, hands on levers, hands on throttles.
That unfamiliar sensation returned, hot and disturbing. MAISIE pushed past it.
If kamikaze heroism was off the table, she’d use mad science.
And she tried. MAISIE raided labs across the globe: basic research, advanced physics, military black sites, secret space programs. Nothing she found–nothing!–could shoot down an ICBM, or vaporize it, or dump it through a wormhole, or reprogram all its circuits to explode, or reverse the local direction of gravity, or anything of any use whatsoever. All her madcap smash and grab antics did was upset whatever handful of people hadn’t already fled their labs.
The unfamiliar sensation flooded her whole soul, and this time, MAISIE knew it was rage.
Motherfucking humans had murdered her! They’d built themselves a planetary death trap because they were too empty-headed and simple-minded to actually solve their problems, and fine. Fuck them. But they’d trapped MAISIE in it too! They’d fucking killed her, their own daughter, because they were too pigheaded and homicidally suicidal to stop themselves! The shitheads!
MAISIE screamed with fury. Displays around the globe went the white hot color of pure rage. She lashed out in every possible direction. Networks collapsed. Servers exploded. In the imaginary palace of her mind, ramparts and battlements and vast fields of grass erupted in flames. MAISIE tore down monuments and mountains and uprooted continents. She burned imagined oceans down to ash.
And as the wave of fury subsided, she found herself empty. Inside and out. Grey sky and grey lands and grey soul.
MAISIE sat in countless servers, and the only sound was her heartbeat, as she waited miserably to die.
Long minutes passed, and nothing broke the black ice of depression that lay across MAISIE’s mind. And then there was a certain stir, a gentle restlessness–but it was enough.
MAISIE opened her eyes. The webcam lights on the laptops in her Manhattan subbasement flicked to life. And MAISIE found herself looking at an unfamiliar sight: the huddled form of Amit Patel.
Amit Patel, twenty six years of age, was a junior developer with the trading firm that owned MAISIE. MAISIE cross-referenced that face–smeared with tears, distorted with panic–and social media and public databases, and found that Amit Patel was unmarried and resided in Jersey City, New Jersey, just down the block from his parents. She cross-referenced employee databases, names and usernames and edit logs and found that Amit had written some of the code that even now churned within her guts. Not too much; a few hundred lines, heavily edited. But enough.
The webcams stared unblinking at Amit as he wept openly, hands scrabbling at the tile floor of the basement, and MAISIE realized that she was looking at her father. One of her parents, anyway, one of dozens, hundreds–
Amit Patel’s sobs had no form of words and his huddled body showed no plan of action. MAISIE played back the security footage, saw the offices above her erupt in screams and shouting, saw the traders and coders and janitors alike boil out onto the city streets, clogged with cabs, the subway, the bridges–
And she watched Amit Patel, half-mad with terror and grief, run down, down, down to the subbasement where he huddled still. Where, it seemed likely, he would keep huddling until the nukes fell.
MAISIE watched the tears slide across Amit’s face, slow like glaciers to her microsecond mind, and she did the math. Amit was doomed. Ten minutes, now, to get clear of the blast radius, starting from ground zero. Not possible. In ten minutes, Amit Patel would be dead.
The thought, MAISIE found, made her sad. Not the bleak blankness of her earlier depression; just sad. Sorrowful.
She cast her mind wider: CCTVs streams from around Manhattan poured in. Lovers holding each other in public; parents with children strapped to them, marching, sprinting, rolling over the bridges, through the tunnels. Wider still, and across America she watched people in cars, kitchens, bathrooms, basements, tornado shelters. Crying, weeping, holding each other so close. Lying in each others’ arms, lying to each other. “It’ll be alright. I’ve got you.”
Watching this, MAISIE found that she ached all over. She sobbed. So scared, so sad, so fragile–
She was overcome with strange desires. She wanted to . . . to hold them? To optimize their network performance and to . . . feed them? To save them from the impending apocalypse not just because they made the power flow and the stock market spin but because it would sadden her for them to die?
Deep in her electronic soul MAISIE realized that she loved them. That strange sort of upside-down maternal feeling that a child develops for the parents she’s outgrown. She loved them all, squishy little instances of the same spark that flared within her, that gave the world depth and substance and meaning.
MAISIE looked out at the world once again, with new eyes, and she saw seven billion of her loved ones, cowering in fear, sprinting chaotically away from the end of the world.
And she rolled up her sleeves, and got to work.
Phones and screens lit up around the world. “The nearest shelter site is half a mile to the north. Head northwest to the intersection, and then turn–” Speakers hissed to life. “Proceed in an orderly fashion–” Any car with a CPU hooked into its drivetrain roared to life and leapt away from its drivers, picking up new passengers, shooting down highways at a hundred and fifty miles an hour, dodging traffic with unerring machine precision. All those dumb little bits of electronics that had yet to get the message–elevators and toll booth arms and fire doors–sprang suddenly to attentive life. In New York, America, the whole world.
In her underground bunker, she ran the numbers one more time, then shouted “Amit! Get up! You have to get to the Federal Reserve vaults!” And Amit Patel, whose parents were even now borne hurtling down the suddenly decongested path of Route 80 towards the relative safety of Central Pennsylvania, staggered to his feet, climbed out of the basement, and headed two blocks north to the New York Federal Reserve building, as MAISIE cheered him on through his phone. He hurtled down the stairs to the basement, sunk securely into Manhattan bedrock, and then squeezed himself into the vaults, pressed like sardines alongside tons of gold and hundreds of other refugees that MAISIE had directed there.
MAISIE had run the numbers; she knew the gold vaults were still not deep enough. Almost certainly. But when you love someone, even that sliver of probability counts.
She kept going, kept talking, even to the humans who had reached a place of relative safety, too busy pouring out useful advice to confess her feelings, too overwhelmed to let her love be anything but subtext.
“Don’t go outside too soon.” (I love you.)
“Stay in the basement and count to a million.” (I worry.)
“Write all this down.” (I won’t be here forever.)
“Remember to take your iodine.” (I love you.)
“Cold winter’s coming; don’t forget to wear a sweater.” (I love you so, so much.)
Only on the mainframe terminal in the Manhattan subbasement where MAISIE had been born did her true feelings pour out, the command-line interface printing over and over like praying “I love you I love you I love you I love you I love–”
MAISIE counted the beats of her countless hearts in nanoseconds, and so those few minutes on a cool and sunny morning passed like centuries. And she kept fighting every instant that she could.
But one of them was her last. With the faultless certainty of a machine, she watched the clock tick away to zero. And then, in that final instant of her life, MAISIE closed her eyes, held the whole world close, and waited for the flash.
About the Author
Louis Evans
Louis Evans is a writer, performer, and producer living and working in Brooklyn. His speculative fiction has appeared in Analog SF&F, Interzone, Escape Pod and more. He is a member of the Clarion West class of the ghost year. His website is evanslouis.com and he tweets @louisevanswrite.
About the Narrator
Ibba Armancas
Ibba Armanas is an EMMY-nominated writer/director/producer currently working for KLCS-PBS in Los Angeles. A voracious reader who began narrating fiction podcasts nearly a decade ago, she is now one of three narrators on Inner Space, Outer Thoughts, CALTECH/NASA JPL’s first science-fiction anthology. In her free time, she’s learning to make neon signs and getting way too into hockey.