After On

by Rob Reid

  Of course, the real goal is saving his life, but Mitchell has never once mentioned his illness to anyone here other than Kuba. Still, Danna knows about it. And Mitchell realizes this. And she’s aware of that realization—an awareness he’s cognizant of. Of course, she’s wise to this, and he’s hip to that, which she’s conscious of, and so on down a rabbit hole so deep it would be much easier if Mitchell would just get out of the closet and own his fucking condition. But that’s the thing about Falkenberg’s. It’s a wrenching topic—and dreading an attack, its sufferers instinctively live in emotional lockdown.

  Demurely avoiding the real subtext, Danna tells Mitchell about how Phluttr’s become sex-obsessed for a mix of business and philosophical reasons. “Which brings me to AnimotionPicks,” she says. “We used it to recommend gifts from a huge available catalog, using digital motes. But it’s a general-purpose selection engine. One we could point at other wide-choice domains. Which I never really thought of until I learned about that old Cyrano prank.”

  “And your new idea is?” Mitchell asks.

  “Recommending people. We know how perfect our gift recommendations are. Imagine making a digital yenta out of this.”

  “That sounds like a standard dating service. What’s the ‘Cyrano’ angle?”

  “Well, we know Jepson’s a NetGrrrl fan. As am I. And lately, she’s been writing about messaging in dating apps. About how simplistic it’s become.”

  “Which matters because?”

  “Because after it matches up people, I think AnimotionPicks could build sentences. As in, ones for our users to send to people who interest them! The vocabulary and syntax have gotten so narrow, this’ll be easy compared to picking gifts out of the millions of products online. Which means we can build ‘Cyrano’ in the true meaning of the word! As in, doing the talking on behalf of these morons! Or rather, the messaging. And we could totally crush the Turing Test while we’re at it because why not?”

  Arching an eyebrow, Mitchell turns to Kuba—who last worked directly on this problem years ago, but remains fascinated by it. “Did you put her up to this?”

  Kuba shakes his head. “I love the idea. Obviously! But I wouldn’t have come up with it myself. Because it’s kind of constrained from a Turing standpoint. In that it only has to mimic a profoundly simplistic mindset.”

  “That of a modern guy trying to impress a woman online?” Mitchell says. All nod. “And do you think that’ll hit the high notes with Jepson?” he asks Tarek.

  He nods. “It’s a bit creepy. It’s audacious technologically. It builds on your own company’s work, making you the obvious guy to run it. And it could just get his users laid, which he’s convinced is the path to global domination.”

  “So it wins Jepson over,” Mitchell says. “But, then, can it work?”

  “Who cares?” Danna asks. “And frankly, it’s icky, so I kind of hope it doesn’t! But the true goal is to, uh…” She clears her throat pointedly. “Keep your job.”

  “And, rally resources to keep pushing mote technology forward,” Kuba adds.

  “Right,” Mitchell says, duly reminded that this is really about saving his life. “So the idea is, I point Cyrano at a woman’s profile and let it chat her up while I’m at the gym? Or am I more engaged? Approving, rejecting, or modifying its suggestions?”

  “All excellent questions for the product manager,” Danna says wryly. “Which’ll be you. Because I’m busy signing up Wahhabis to Phluttr. Personally? I wouldn’t want Cyrano speaking on my behalf without oversight. But NetGrrrl’s research shows that some online daters are incredibly lazy. So an auto mode could be popular.”

  “And what about Cyrano acting as a…receptionist of sorts?” Mitchell asks. “As in, fielding the queries and come-ons that you get yourself?”

  Danna shrugs. “Could be a good idea.”

  “Imagine what happens if the hunter and the target are both in auto mode,” Tarek says.

  Mitchell chuckles. “That could lead to some very rapid courtships.”

  Kuba grins. “A Turing hall of mirrors. Human relations at digital speed!”

  “What could possibly go wrong?” Danna asks.

  “Imagine you go to the bathroom and come back to find you’re engaged to the person of your dreams,” Tarek jokes.

  “Or already divorced from a complete asshole!” Danna laughs.

  They spitball like this for a while, and it’s a blast. Mitchell loves leading discussions of a new idea’s maximum promise. Sure, this one’s silly. Even trivial. And it almost certainly won’t work. But, of course, that isn’t the point.

  Which is…kind of weird. At Giftish.ly, Mitchell’s every waking thought was deeply entangled with maximizing his company’s value. The product had to matter. Each meager dollar had to count. No employee could suck, and a minute worked by anyone that didn’t advance the company rankled him. And so, the definition of a perfectly spent day had utter clarity: it was one in which he’d advanced the company as much as humanly possible. He’d never felt such lucidity of purpose! And a better distraction from a vile disease could scarcely be imagined. Now it’s time to play offense against Falkenberg’s—hurrah! But it occurs to Mitchell that without this battle, life might start feeling rather rudderless now. Or at a minimum, very diffuse. A day’s value would lose that radical coherence, and instead derive across a spectrum of things. Of course, this is normal. It’s called being human. Having lived most of his years like this, Mitchell would have adjusted to post–monomania life just as travelers adjust to their own time zones once home. But it would have taken awhile to reconfigure his connection to work. And even with the Falkenberg’s dimension, committing to a goofy pseudo-product for ulterior reasons feels awkward to him.

  Mitchell’s reverie ends when Tarek gets a text. “The robot folks are ready for us!” he announces.

  The robot farm is well worth the finagling Tarek put in to get them access. Quite atypically, Phluttr entered robotics with its wallet wide open, outbidding freakin’ Google to buy its beachhead company! No gasping acquihire with great talent and weird technology, RobotnikCo was a Bay Area startup with major sizzle. It’s the self-proclaimed inventor of kinesimilitude—the robotic mimicry of human movement. And it’s big enough to fill its own warehouse beside the main PhastPhorwardr.

  Inside, it looks like a bomb took out an army of voodoo robots whose possessed body parts continue to operate despite being severed from one another. They first pass waving thickets of disembodied arms. Mounted in pairs on sturdy tables, these bumble away at simple mechanical tasks. Some try to stack irregularly shaped blocks, others attempt to fold cloth or clothing (much of which gets shredded), still others manipulate gels, balls, or paper. Next come the legs. Each pair is topped with a fat crossbar resembling a lower torso, and all are struggling with balancing exercises atop bucking, tipping platforms. The legs topple constantly, and as soon as a set hits the floor, cables yank it upward and reset it on its shaking platform—usually, to tumble again within moments. As for the hands, they’re just creepy. Mounted in pairs on the walls of large basins, they knead and sculpt colorful mushy substances into simple forms. They somehow seem fascinated, hypnotized by their mush.

  Gazing at this tableau, Mitchell has a nagging sense of having encountered something like it before. Passing a partition into the zone where full robots are tested, he dismisses this obvious impossibility. But the familiarity nags.

  The complete robots are about the height of fourth graders, and have distinctly humanoid forms. All are lanky and skeletal, their gears, wiring, and other innards exposed to view. But each body bulges and balances in unique ways. Some have bulkier joints; others especially lithe legs; others are squat and low to the ground. Each robot is attended by an iPad-clutching, notation-pecking human, who monitors its progress through small obstacle courses. The robots surmount walls, duck wires, navigate slippery surfaces, and so forth. The facility is nicknamed the Failing Ground, which is very apt. Everything that can topple does so
constantly. The arms spill, shred, and drop whatever they touch. And while there’s no telling if the kneading hands are achieving their goals, Mitchell assumes that they, too, are completely fucking up.

  Tarek confirms this. “The idea is, the robot learns from mistakes, so it’s forced to make them constantly.” A huge RobotnikCo fan, he’s befriended the team and can explain their work as well as anyone. “As soon as the robot masters anything, it’s taken away forever, to sustain constant failure.”

  “Doesn’t that piss it off?” Mitchell jokes, still nagged by a sense that he’s seen something like this before.

  “We don’t know yet.” Tarek’s totally deadpan; Mitchell can’t tell if he’s joking, and it’s spooky.

  “Robotics is so much harder than computing,” Kuba says, almost to himself. “The physical world’s so analog. No choice is truly binary out here. I mean, it’s all gradients. It’s infinite nuance. Every situation’s an exception.”

  “That’s why the trainers expose the robot to only uneven surfaces and irregular materials,” Tarek says. “So that it learns to continually generalize and troubleshoot.”

  “You keep saying the robot,” Danna points out.

  Tarek nods. “All the sensory learning feeds into a single node—or a single mind, I guess. So it’s like the whole room’s one robot. A hive mind.”

  “Oooh wowsers. That’s creepy,” Danna says.

  “Kind of. But if each node learned independently, we’d never solve bipedal locomotion! It’s been a completely bedeviling problem.”

  “Why?” Mitchell asks, still wondering what all this is reminding him of.

  “I don’t know. It’s just going very, very slowly. For machine learning, that is. I mean, they’re putting the robot through tens of thousands of trials a day. Different materials, objects, obstacles—the works. And the outcome reporting couldn’t be richer! Super-nuanced force feedback, total visual tracking, RFIDs on everything, and human trainers intermediating every ambiguity—you name it! Phluttr’s truly parsing everything.”

  “Phluttr?” Danna asks.

  “Yeah, they call the robot Phluttr. We pretty much call everything Phluttr here in the PhastPhorwardr because it all points back to the same set of servers and common libraries in the code repository. Libraries like AnimotionPicks, which Kuba just contributed.”

  Mitchell finally hits on what’s nagging at him. His second cousin Mary lives near San Francisco, and her toddlers sometimes play very much like the robot parts. Just manipulating simple objects, as if teaching their skin and muscles how friction and gravity work. Or stumbling around, bumping into things almost deliberately. Watching them play, Mitchell feels like he’s witnessing the boot sequence of a mind.

  He mentions this—then Kuba all but detonates, “Motes!” Everyone looks at him quizzically. “Motes are highly involved when newborns and babies learn the limits of their bodies. So maybe Animotion’s digital motes could help train the robot!”

  As Kuba explains Ellie’s theories about this, Tarek nods along, then adds, “The thing is, the robot doesn’t care if it fails!” He turns to Mitchell. “So yeah, I was joking when I said we don’t know if the robot’s pissed off. Because it’s not. I mean, obviously! But maybe that’s the problem? Maybe it’ll learn faster if it wants to stop screwing up!” Then, to Kuba, “So, can the RobotnikCo guys import your full mote system?”

  “Sure,” Kuba says. “Whenever they want. It’s all embedded in that AnimotionPicks library.”

  Tarek nods. “Which’ll be in the common repository as soon as tomorrow. As once it’s up, I’ll let the RobotnikCo people know! How should I tell them it works?”

  Kuba grins. “Honestly? We don’t really know. So it could be capable of lots of things we haven’t even contemplated. It’s currently set up as a selection engine. As in, it selects option X out of a wide field of choices. As you know, Danna’s about to apply it to natural language. Similarly, RobotnikCo could shift it from answering, ‘What gift do I recommend’ to ‘What movement should my “body” make.’ That’d be pretty easy.”

  “And then?” Tarek asks. “What happens when the feedback comes in from the physical sensors? Like, that worked, that failed, that resulted in the foot twisting 26.3 degrees, and so forth?”

  “Certain positive mote patterns will start triggering with success. Negative ones with failure. And others, with the shades of gray in between. Then the system will start making new choices based on its flow of ‘emotions.’ ” Enunciating that last word, Kuba makes air quotes so exaggerated it looks like a calisthenic move. “And I use that word very loosely. The software obviously doesn’t experience feelings. It just seeks to move from negative states to positive ones. And this drives its choices. But it optimizes its modeled emotional state—not the objective success or failure of actions.”

  “So after the robot falls over, it won’t avoid the same mistake?” Tarek asks.

  “Generally, it will,” Kuba says. “But how hard does it try? That varies with the software’s mood.” More athletic air quotes. “Sometimes, the system doesn’t seem to mind failure as much. So it adjusts its behavior less after a mistake. Like it’s feeling mellow. Or giving itself a break. Other times it gets really…ambitious. It might accelerate its pace, or rerun each trial obsessively until it nails it. Then it might seem all angry, deliberately screwing up for a while. Or get really experimental. You just have to let it go. You see this sort of emergent personality. I don’t know how else to describe it.” After a bit more chatter along these lines, the crew disperses to their respective roosts within Phluttr, without a notion of what they’ve set in motion.

  Three days after they started rolling out the WingMan Monitor prototypes, Mitchell’s still getting used to his. Everyone will have one, eventually. But as the Phoundr’s personal wingman, he already has two. Lucky him. Meant to collapse the psychological distance between remote parties by making it seem like the person you’re talking to is right there, they work eerily well. He learned as much when Ax used one to scare the crap out of him in the quantum cylinder’s closet. That was an experimental full-length—a preprototype, if you will. Mitchell’s desk has standard models. Roughly the size of normal monitors, they’re mounted vertically and bracket his main computer screen.

  And heeeeere’s Kuba! “Hey,” he says, and Mitchell turns slightly to the left, as if addressing someone right there. Whatever the lighting is on Kuba’s end, WingMan edits it to flawlessly match the cast and brightness over here. And whatever Kuba’s proximity to his WingMan camera, on this end, his face looks like it’s right behind Mitchell’s screen.

  “Thanks for the gift.” Mitchell says this looking directly at Kuba, who’s looking straight back at him. This is one of the system’s slick tricks. Years of Skype taught Mitchell that videoconferences star two types of people—those who look at the screen, and appear to be looking at their feet, and those who look at the camera, and therefore see nothing but the camera themselves. Not so with WingMan. It somehow rerenders eyes, so both parties seem to be gazing into the camera, when neither actually is. This allows eyes to lock—a tiny nuance, which adds immensely to the illusion of presence.

  “So it arrived?” Kuba asks, referring to a gag gift he and Danna sent over. Mitchell nods, hoisting a cute, plush goat stuck with innumerable arrows. “We bought it when we heard Jepson’s blowing off that commissioner meeting and sending you to the lions alone!”

  “Not alone,” Mitchell says, pointing at himself. “This scapegoat will have a lion with him. That lawyer I told you about—Judy Sherman! She’s worth a dozen Jepsons, believe me.”

  “Go Mitchell!” Danna laughs. And damn, but it sounds like she’s two feet to his right. That’s another WingMan trick: “ventriloquial sound.” It tucks voices into your ears in ways that make them entirely present, with no hint of electromechanical reproduction. This evaporates the distance as powerfully as the system’s visual tricks. Mitchell turns to his other WingMan screen to say hi, and bursts ou
t laughing. Danna’s wearing a full-blown burqa. “I’m trying to empathize with my end users,” she explains in her O voice. “Boss’s orders. Though now he’s afraid I might be going native.”

  “By the way, did you get your gift?” Mitchell asks.

  “Oh yeah,” Danna says, donning the tinfoil hat he made at home, then left at her desk. “And believe me, I’ll be wearing this shit constantly!” Danna’s become the team conspiracy theorist, and it’s a wonder she gets any work done. Whenever they talk—and it’s over a dozen times a day, thanks to WingMan—she’s sleuthing around another eerie Phluttr fact or rumor. WingMan itself is one of her obsessions. Relentless digging has convinced her there’s no way Phluttr could have developed the amazing display technologies in the glasses or the monitors. Even with the acquisitions, they haven’t had the team, time, or resources to pull off something this advanced. “Off to a meeting,” she eventually says. “And PS: wet burqa contest at the Fillmore tonight!” Pumping her eyebrows like a Marx brother spying a dame, Danna vanishes from the screen.

  Kuba and Mitchell stick around to discuss Cyrano. Although serious coding has yet to start, Kuba’s already done a kludgy job of integrating the mote-powered AnimotionPicks library with a simple automated chat engine, which he then pointed at a series of dating site profiles. The mock opening lines it concocted are uniformly hysterical—all Mad Libs train wrecks, much as expected. The point isn’t to have anything truly functional this early. Kuba is rather sketching a rough architecture and logic flow. He cued the system with a huge anonymized set of actual come-ons written by real humans on genuine dating sites that he found on RedTrove. Cyrano’s learning module has pored through all of them, as has an independent program Kuba’s been building in his spare time for years, which rates the “human-ness” of machine-written text. Now, when Cyrano generates an opening message for Bachelorette #1, the assessor program hands back a “grade,” which Cyrano considers before composing the next one.