Looking at the same call stats as George, Pandora removed her headphones and lowered her laptop screen. It was a little weird to realize it—what with a therapist in the family—but she’d never watched her father practice. Sure, she’d recorded his sessions but didn’t watch them so much as fast-forward through them, listening to the clients’ sped-up voices, poised to alert on certain keywords. And now that “voice-recognition technology has really upped its game”—something she’d told her father about, by the way—Pandora hadn’t had to screen the sessions herself at all, sped-up or otherwise. Plus, when her dad wasn’t working, they talked about other things—meals, the weather, what to watch on TV, what interesting factoids she’d learned reading the Google News headlines for science and technology, but never politics. Pandora had found that by focusing on the first two and ignoring the third, she was able to remain hopeful about the future of the species without succumbing to despair and/or that session’s magic word: suicide.
Seriously?
That’s what the guy wanted to discuss? Pandora was about as far away from being suicidal as a person could be. She knew precisely what she wanted to do with her life: keep it. Pandora wanted to live—she felt embarrassed even thinking it, but it was in her heart and it was true—forever. The thing she wanted to be remembered for on her death bed? Not dying. And as far as her epitaph, Pandora thought “This Space for Rent” should do nicely.
But George Jedson wanted to talk about suicide—specifically, teen suicide. He’d said it was because of a work assignment, but if that was some sort of excuse . . .
Damn it.
The original plan had been to look up George’s contact info in his case file and reach out directly, say, “Hi,” or “Wassup?” Nothing too stalkery. He’d made the first move, what with the unsolicited cuteness comment, so . . .
But this suicide business was worrisome. And so, as with a new show on HBO or Netflix she was intrigued by but not quite committed to, she’d have to keep tuning in and hope it wouldn’t turn into another Young Pope. If only there were a way to binge this George Jedson or skip ahead to where she’d figured out whether he was worth it or not. Unfortunately, Pandora had not found the fast-forward button for the real world, which had been a blessing when it came to Gladys, but George? Not so much.
29
There was nothing in the language of George’s work assignment that stated—explicitly—that he was to develop that Holy Grail of AI: a fully conscious, general artificial intelligence. It wasn’t even identified as a stretch goal. But it was there, implicitly, between the lines. It was there in the difference in success rate that his predecessor had achieved (a respectable annual projected suicide rate within the target demographic of one to two percent per year) and George’s target, i.e., a nearly unthinkable zero suicides in the prescribed demographic on the platform moving forward. And that wasn’t a zero percent, or even a zero-point-zero percent. Zero meant zero. It was a whole number, not a rate.
George knew of one way to achieve this goal, though he doubted it would make any of the higher-ups happy: overpredict. Set the criteria for flagging a potential suicide so loosely it’d capture everybody. After that, suspend their accounts and declare victory and bankruptcy all at the same time.
Yeah, probably not. Ditto calling his bosses’ attention to recent studies showing that a reduction in social-media usage was directly correlated with lowered instances of depression in the target demographic. If his run-ins with Milo had taught him anything so far, it was this: social media was not a reality-based industry.
So: zero suicides among Quire users aged fifteen to twenty-five when his predecessor had been fired for missing—let’s be honest—one? There’d been other cases Vickers’s approach had missed—a handful of the nonfamous—but it was the Gunn suicide that constituted the tipping point of interest. And it was the smallness of the number that was so daunting because it was always the last few of anything—miles, votes, raked leaves, percentage points—that cost the most time, money, and other, subtler investments like creativity and/or emotional involvement. In the case of Quire’s would-be teenybop stay-alive bot, those last few bodies cost the last guy his job, making way for George, who was quite conscious of the hidden complexity of the task ahead of him. He’d already assumed he’d use the first guy’s algo for screening out the no-brainers but after that . . . ?
Well, let’s see: he was a computer hacker surrounded by computer systems with access to some of the most sophisticated neural nets available and his pick of machine-learning algorithms, suggesting, yeah, he should let the machine do the heavy lifting. He wouldn’t even have to code anything, because machine learning wasn’t about coding; it was about training. What mattered was the quality of the data you gave the machine to learn from—good old-fashioned GIGO: garbage in, garbage out. And vice versa. So George started feeding the beast pretty much everything he could think of related to his task. This included but was not limited to: case studies from psychology textbooks and journals; doomed love stories like Romeo and Juliet; treatises on ritualized suicide from kamikaze to jihadi suicide bombers, from Christian martyrs to self-immolating monks; the bulk data from past Quire members who had, in fact, killed themselves. Feeling the beast needed more, George proceeded to hack into police departments across the country and copied all the reports of teen suicides he could, along with death certificates and autopsy notes. He then moved on to downloading material from anorexia-encouraging websites and online suicide games like the Blue Whale Challenge and that creepy-looking Momo. He followed the you-may-also-likes these linked to down the deepest existential wells until all he wanted to do was sit in the corner of his corner office, sucking his thumb and rocking back and forth.
The algorithm, meanwhile, had begun populating word clouds from the raw information it consumed, with one word taking up the center of the cloud early, where it grew and shrank under the shifting statistics, beating like a heart: consciousness.
Victims and victim wannabes “lost consciousness,” “returned to consciousness,” “never regained consciousness.” The word seemed to have a certain statistical stickiness that resonated throughout the data set. And that was how a routine exercise in corporate soul-searching became an investigation into the nature of consciousness itself and—as a stretch goal—emulating it artificially.
Because that was the answer, reading between the lines and implicit in the numbers expected. George’s chatbot had to have skin—silicon—in the game. It had to appreciate what was being lost by the game’s other players.
Oh, and one other thing. Thanks to his predecessor’s ham-handed handing off of the intervention side to local law enforcement and their surfeit of SWAT equipment, an addendum had been added to George’s hand-delivered, hand-me-down assignment: “Once target subjects have been identified and reasonably confirmed, all interventions shall be kept strictly in-house and on-platform.” Meaning, basically, a chatbot that would engage and talk the subject out of suicide if, in fact, the subject was contemplating such.
So: talking—that would be the singular tool in George’s toolbox. And not even that. Text. Written words on a screen. On the user end, individuals might have customized their experience using voice-recognition apps, but the default was text and maybe emojis, especially given the target demographic. Now all George had to do was figure out what those lifesaving words should be and how to get his chatbot to type them.