Love in the Time of MapReduce02 Apr 2012
This piece was written for an internal Google fiction contest, for the 100th edition of the engineering newsletter. The call to arms arrived in my inbox like so:
For this special Eng Newsletter issue, we’re running a “google eng-y” short fiction contest. You can write about anything, but the story must begin with these two words: “The MapReduce”.
Please note that some meaning may be lost on non-Googler’s, notably the bits concerning company hierarchy. All the opinions expressed are my own and obviously do not constitute the workings of an actual Google plan, etc. Jeff Dean is a very nice man. This is a piece of fiction in almost every sense.
The MapReduce was a piece of technology whose existence its steward, Jeff Dean, sometimes begrudged. It was glamorous, in a way, to be the public face of the algorithm that had essentially rewritten interpersonal contact, but it was also draining and surreal.
In one of Jeff’s increasingly common attacks of perspective, he realized that his daughters, too, had been completely swept up by a thing that he himself had designed, built, and evangelized. They were, of course, perfectly happy with the product. Jeff noted this with a tinge of grim pride, remembering the long nights of trial runs. Victoria and Natalie were a bit too happy, Jeff mused, so completely satisfied with something they could never understand (indeed, that he himself no longer understood well), that he found their lack of doubt troubling. Why didn’t they care that it probably shouldn’t work, that time and computation could twist statistics in such a fundamentally disturbing way? It was probably due to both of them being so preoccupied with Natalie’s wedding, he concluded wearily.
Later, outside of his office, Irina was waiting for him.
“Jeff, you have a visitor waiting for you in your office,” she said. Something in her tone gave away the urgency of the situation, and Jeff nodded, having long grown used to trusting Irina to manage his calendar more deftly than he could tie his shoes.
His suited visitor was a trim man of about sixty, which was unusual enough for the Googleplex in terms of both age and dress. He wore his graying hair swept back and neatly cropped. With a start, Jeff realized that his visitor was none other than a senator of Iowa.
“I’m Robert Graves, and sorry about showing up so unannounced, Mr. Dean,” said the man, with a smile. Jeff paused for a moment to admire how finely countenanced the man was, and to feel a small thrill at being so delightfully underdressed, himself.
“You’re the senator pushing for patent reform. I don’t watch TV much, but I’ve seen you on when my wife watches the news.” Jeff shook his visitor’s hand and seated himself behind his desk.
“The very same. Look, I’ll spare you the pleasantries and get right to why I’ve come. I’m told that engineers prize truth and directness.” Jeff lifted an eyebrow at this, having found that lately he valued being left well enough alone better than both of those things. “As you well know, MapReduce is proving problematic, socially. FOX is filming a reality TV show at this very moment about an engaged couple who are convinced that after trying out their MapReduce partners, they’ll still want to get married.”
“Jesus. How’s it looking for the couple?”
“Not good. Even worse, they’re filming it in my hometown.” Graves massaged his temples.
Jeff was not surprised. MapReduce rarely erred. Though it had begun as a general purpose framework for parallelizing search index updates, it eventually lent itself to analyzing the massive amounts of user generated social data Google+ collected. In time, this would become all that MapReduce was known for (at least externally of Google), in a queer reversal of how the words escalator and aspirin came to describe all such contrivances, though they were once only brands.
“Basically the right is getting as much fuel as it wants for its eternal fire of shouting about our perpetual moral decay. On top of that, MapReduce is having a powerful economic impact, which doesn’t help. We’re having an employment problem, as you’ve doubtless inferred by now, since you must have all the numbers on how many people are using MapReduce to pair up.”
The first Jeff had heard about the phenomenon the media had dubbed as the “honeymoon effect,” had been from the news itself, but he nodded anyway. “My citizens are up and leaving jobs they’ve worked at for a decade to meet their dreamboat on the other side of the world. I mean, great for them, but our coffers weren’t in great shape before, and your invention is a drain we can’t possibly afford right now, never mind the bad press. As much as I am for the future, I desperately need you to stop operating in my state.”
Jeff suppressed the urge to tell the man to just contact firstname.lastname@example.org, and instead reluctantly launched into a narrative he had delivered many times before. “I’m sure you’ve seen and read all the press releases about this. What we do isn’t terribly new. We provide a service that users want. In a sense, we provide nothing more than what eHarmony and Match.com have been providing for years, just with much less uncertainty and a bigger candidate pool.”
Robert snorted. “I’d hardly call ‘every Google user on Earth’ a bigger pool. Your operation is different, too. You know all the things that people have searched for, and all the things they’re too ashamed to search for. You know why some actresses draw men to them, and which men women will wait hours to receive texts from. Those bankrupt dating sites had only the constructed personas of the desperate to work with. There’s a case that could be made here for unlawful invasion of privacy and monopolistic abuse of information.”
A Googler rode past Jeff’s window on a small yellow bicycle. Jeff focused on the bright colors to briefly escape his current uncomfortable tension.
If Graves was right about anything, it was that MapReduce was uncannily effective. Through what some people might call sorcery, or what Jeff’s team leads described as “massively parallel Bayesian-adapted machine learning plus deep social mining,” it was able to identify, with nearly 97% confidence, a lifetime romantic partner for any given user. The algorithm could even supply just the right amount of shared interests as conversation starters, while leaving enough unsaid for the nascent couple to discover independently, leaving them feeling as if they had come to know each other intimately of their own volition. Some people found this deeply unnerving.
Even those people commonly derided by society could find love in this way, though MapReduce might take weeks instead of seconds to produce a suitable pairing. People of every sexual deviancy and every personal vice were being matched up, to the horror of the many people alienated by the brutal efficiency of MapReduce’s perfect lack of bias.
In short, romantic fulfillment was, for most people, little more work than clicking “I’m feeling lucky” and buying a plane ticket. This is what the people wanted more than anything else. Graves knew it, and Jeff knew that Graves knew it. Furthermore, Jeff knew that Graves was powerless to do anything about it, so strongly did the public crave MapReduce’s presence in the world. Yet Jeff felt sympathy for Grave’s willingness to shoulder the impossible task of squaring the budget against falling revenues and changing social tides.
“Mr. Graves, I understand your dilemma. The last thing you need right now is the income rug pulled out from under you. But look at it this way: about half of those people who have gone and paired off will probably come back to their hometown, bride or husband in tow, so your population will probably end up about even. After these couples outgrow their honeymoon period, they’ll settle down, work, have kids, and spend with an intensity that only the truly content can bring to bear. In the coming decade, your books might even make it into the black.”
Robert was not easily placated. “Can you say for certain that this is the way it’s going to play out? The world has never seen this kind of mass social movement. What if the people become complacent instead of motivated? What if your algorithms can’t guarantee long term stability?”
Jeff had an inner conflict. As usual, the side favoring the least amount of social friction won out. “We’re the ones who managed to pair everyone up so well in the first place, aren’t we? The models say the population will eventually converge on a higher level of stable productivity. I can’t promise you it’s going to happen, of course, but here at Google we have pretty high hopes for the future.”
The two men talked in this way for some time. The elder statesman pushed and the younger (but not exactly young) engineer deflected until the senator grew weary or satisfied enough to defer discussion to a later date. Jeff had managed to end the meeting with only vague promises, a surprising talent that had earned him his relative autonomy from Larry Page’s inner circle. Later he would have to file a report, naturally, detailing the intricacies of his conversation with the senator, but for now Page trusted him to keep third parties at arm’s length on his own.
Later that evening, after a quiet supper with Heidi, Jeff lay in bed thinking. The models actually didn’t say much about the economic reality of the future. The social data that allowed his team to pair people so effectively seemed to shrug mutely at the problem of what the future might be like. He had assured Graves that everything would be fine, but by the time Jeff could be proved wrong, he would be long retired.
Sleep took him. He dreamt, which was not unusual (though he didn’t know it), but he also remembered his dreams from that night, which was. He dreamt of a young man smashing a perfect chalice in a decrepit hallway, and of women who laughed while they danced away from their homes.
When he woke, Jeff knew what he had to do.
Sanjay probably could be trusted, but Jeff couldn’t take the chance. He would split his change into pieces, and sneak them into other, tangentially related changelists. The other developers on his team would probably rubber stamp these, anyway, since Jeff was one of the most prolific programmers there was. Who would look at yet another Jeff Dean code review too closely?
What did it mean to adhere to Google’s famed “Don’t be Evil” policy, when it came to arranging marriages? The standard Google answer would be to make the user as happy as possible without violating their trust. But what trust was there to violate if users themselves didn’t know what they wanted in relationships, or what would truly make them happy? Marriages are long lived beasts, Jeff reasoned, subject to slowly building changes in the macroeconomic climate. If marriages affect the economy, and the economy affects all marriages, what should you optimize for, and how?
Jeff’s changelists were approved, as a matter of course. Years later, he retired. The day he first started noticing what might have been the fruits of his subversion ripen, he remembered a thing that his old mentor Urs loved to say, before Urs had left him in charge.
“It is better to ask forgiveness than permission,” Hölzle would often chuckle, in a particularly German way. Jeff chuckled now, too.
Wired was doing a bio piece on a recently minted tech millionaire. The man was one of the few people for whom MapReduce’s pairing hadn’t worked out in the long term. When asked what had motivated him to start the company he had just sold, the man somewhat abashedly said that he wanted to prove to his ex-wife that dumping him was a mistake.
Ambition and talent sometimes survive contact with love, Jeff mused, but are more often dulled by it. MapReduce could identify those individuals who are defined by intelligence, drive, and pride. In other words, the archetypal entrepreneur. A few modified terms in a complex linear algebra equation could yield surprising results, Jeff had discovered, like optimizing for romantic partners that would net the largest increase in a person’s ambition, rather than happiness. A lot of the unfortunate people of talent singled out by Jeff’s modification would probably yield little value, but one, he hoped, would build the next Google. Jeff longed to see that day.