One of the novels happens to be on a government watch list; they want to keep tabs on who’s reading such “radical” ideas. Wonder if she talked about the novels with her sister? Maybe. Maybe her sister talked to her boyfriend. Did it go further? Unlikely. It’s even more unlikely that Antonio talked about it with his mom. Who would? But unlikely doesn’t mean impossible. He might have discussed it with his mom, who then talked to your mom, who then talked about it with you in some passing comment. Unlikely—yes. Impossible—no.
If you have enough of these types of connections, eventually the probability that someone will talk to you about something they heard second, third, or fourth-hand approaches inevitability. You’re connected.
Don’t talk to people much? No problem. Your connections, and everyone else’s, aren’t just to people, they’re to web sites, to products, to places. You don’t need to talk to people, you just need to interact with something, anything, that Ubatoo tracks.
No man is an island. That’s why Stephen’s approach would work, and that’s why the technology behind the WatchList project was never in question. It consisted of a large graph with almost 200 million black dots representing the people online in the U.S., and 5,000 red circles, the people who were being watched. Ubatoo’s knowledge was represented succinctly in the connections. All he had to do was analyze the graph. Just radiate the “bad” influence along the connections, and see which of the black dots start turning red. The further away you were from a red circle, the safer you were. Connected to too many people who were connected to too many reds? Be careful.
The system wasn’t perfect. It really would benefit from using the NSA’s entire watch list as a seed set. If he knew all the people who were being watched (what did Sebastin say again, a million people?), then imagine a million red circles to initialize the system. Having the data would make all of this much more reliable. But for now, any seed set would do. He could still find people with too many questionable connections—too many to be considered just a coincidence.
-COLLIDE-
August 5, 2009.
A graph is a graph—a bunch of dots and a bunch of connections between them. How hard could this be? Stephen was days behind schedule. Creating the graph and finding the data to populate it was easy. Ubatoo had the infrastructure ready and the profiles and the right connections were easily extracted from its databases. But the problem was that Stephen wasn’t versed in the latest algorithms to tackle such massive graphs. There were so many parameters, or virtual “knobs,” to tweak: connectivity constraints, damping factors, evidence cutoff thresholds . . . and the list went on. With graphs of the size he envisioned, the effects of improperly setting any of the knobs would be horribly magnified, making the resulting assessments of “red” versus “black” labels garbage. In fact, until two days ago, when he had suffered a mild panic attack, fueled in equal parts by desperation, sleep deprivation, and caffeine, absolutely nothing was working.
There were probably dozens of people at Ubatoo who could have helped, but given Kohan’s reaction to the project, he hesitated asking. Fortunately, he was far from alone in trying to understand the impact of connections and evidence in graphs. As far back as the early 1900s, demographers like Alfred Lotka and the husband and wife team of P. L. and E. M. Gross started exploring the basics of a field that became known as bibliometrics. Bibliometrics, the study of texts of information, included studying the impact of information by looking at how often it was referenced by others—how often it was found useful, how often connections were made to it. Within the last decade, the latest research coming out of the computer science schools at Cornell and Stanford took bibliometrics a step further, using it to measure the influence of web sites by looking at how many other sites linked to them. From this, ranking in all major search engines took their latest inspiration.
Within the past few years, a new round of research was emerging—applying the principles of graphs and connectivity to ranking and importance assessment in specialized domains. In the countless academic papers Stephen had hurriedly examined to find any hints on how to set the “knobs” for his algorithms, he found the research paper of Aore and Mikens from Georgia Tech particularly helpful. Though they were working in an entirely different domain than Stephen (they sought to personalize video and music recommendations), they were the most pedagogical in their writing. They spelled out the intuitions they used for setting the “knobs” for their domain. Building from these insights, it was only a few more days of non-stop trial-and-error before Stephen got things back on track.
Between the fourth and fifth coffees of the day, Stephen was ready to run his first trial experiment with WatchList. It was still too soon to try his approach on the full graph with 200 million people; instead, only a small fraction would be used initially. Less than 250,000 people were included in this experiment. It would take only a few minutes to start seeing the first results.
It was 10:55 a.m. The data and programs were loaded, ready to begin. Leaning back in his chair, he typed the command that would put the gears in motion, then let his finger hover over the Enter key, gliding back and forth across its surface, but not pressing hard enough for it to register. He wanted a moment to review all the tweaks he had just made—to reconsider whether he had forgotten anything. No, he didn’t think so. “Please, please, please work this time,” he pleaded under his breath.
He let his finger drop, and with that, dozens of computers somewhere in Ubatoo’s cloud instantaneously sprang into action. The graph was realized, the points were colored black and red, and the influence of the red nodes traversed the millions of links connecting the points in the graph. Before he could finish his cappuccino, the progress bar closed, and the hopeful “Run Completed” message popped up.
Now it was time to check the results and make sure everything worked correctly. The task of examining 250,000 results was not an easy one. He randomly picked a few names and manually traced their connections to see how far they were from the red nodes. So far, so good; the results seemed reasonable. Things were finally looking up. This was further than he had gotten in the last eight days.
Next, he searched for a few specific names—he had added a number of people he knew from high school, SteelXchange, and even Molly, Kohan, and Yuri to the list of 250,000 people used in this experiment. His high school classmates checked out just like he thought they would; no major surprises there—no one was a terrorist. Same with SteelXchange.
Knowing the task at hand at both an intellectual and intuitive level, understanding thoroughly the work he had done for ACCL, and playing with the data for the past few weeks, why was it not evident to Stephen what was about to happen next?
When he found Molly’s name, it was bright red. His first error, he thought. What was wrong now? Even if her name were a little red, that would be understandable, but why was it such a bright red? He checked a few of the people she was connected to—they were red, but nowhere nearly as bright. The further you looked from Molly, the less red the graph became. How could this be? From where, then, was she accumulating the red influence? Something didn’t make sense. The only way her name could be that red was if she was actually on the original list herself, and so she was making her surrounding connections red—being the cause rather than the effect.
Before this, no reason had existed for him to look over the full list of 5,000 names. He was an intern; this was one of several dozen projects he had worked on over the summer, who could have expected him to look at each and every name in every attachment? But now he found himself hurriedly scrutinizing the list, scrolling through the pages, looking for Molly’s name. It didn’t take long to find. He stopped his scrolling at line #4,793: “Molly Byrne.” His Molly.
There she was, along with all the rest of those who were being watched. She had made it onto the list that he had handed to ACCL. She was part of the red point set. She, along with 4,999 others, was the source of the troubles that propagated throughout the graph. She was part of the
set that was casting doubt and suspicion on all those who knew her, causing even those with no reason other than just talking to her, to be suspicious, to be red.
It was now 11:30 a.m. He left his computer and bounded to his car. Never once did it occur to him that if the graph were ever completed, Molly and he would be connected with thick dark strokes, not small thin lines. Her influence, whether drawn or not, was more on him than on anyone else.
“Molly’s been telling me what you’ve been up to since you left, Stephen,” Allison said as the three of them waited in line for a lunch of hotdogs and popcorn at GreeneSmart’s café. Allison waited for some acknowledgment, but it didn’t come.
“Stephen, you with us?” Molly said, as she tapped him for a second time, having failed to get his attention the first time.
“Oh. Sorry. I’m just lost in a problem I’m trying to figure out. The usual Ubatoo stuff. They keep you thinking about work all the time,” Stephen lied.
“Figures. You spaced out when you worked here, too. Speaking of which, are they working you too hard to come by and visit us sometime?” Allison affectionately chided. It had been a long time since he’d heard her motherly tone.
“Allison, I’m really sorry. Today’s just not been a great day. I’ll invite you to lunch at our cafés before the end of the summer. That way you can see what all the fuss is about for yourself. It’ll be worth it,” he said, motioning his eyes over to the hotdogs on their trays. “I guarantee it.”
“Sure, Stephen. Just don’t forget that invitation.” With that, the conversation thankfully ended. Stephen and Molly walked in silence to a table by themselves.
When they were seated in GreeneSmart’s swivel chairs, Molly was the first to speak, “Okay, so what’s going on with you? What’s so urgent that we had to talk right now? Is everything okay?”
Stephen didn’t waste any time, “Your name may be on a list it shouldn’t be on.”
“What are you talking about?”
“I’ve never really told you about what I’ve been doing at Ubatoo. I’ve tried a couple of times, but our schedules just haven’t worked out that well. I’ve been helping out this organization called ACCL. Do you remember me ever talking about them?”
“I’ve heard about them. They’re all over the news. And a bunch of people have been posting about them on EasternDiscussions. I don’t know if we ever talked about them, though.”
“The head guy there, Sebastin Munthe, and I have been working together. I mean, I’ve been doing some data mining for him to find out who’s next to make it on the official watch lists.”
“You’ve been doing this? Ubatoo does that?”
“We have so much data that it makes sense for us to help out ACCL. Anyway, I’m the person who’s been doing it.”
“Wow. That’s really, really, cool, Stephen. All the people on EasternDiscussions would be really big fans of yours. They’re really enamored with ACCL—well, at least the ones who aren’t cynical about everything,” she looked at Stephen almost adoringly.
There wasn’t time to bask in the glow right now. “But there’s a problem. The point is, Molly, I think you’re on one of the lists, too. You . . . you fit the profile too well. You spend all your time posting messages to the right sites, talking to the right people, buying the right books, and just think about your e-mail conversations, who they’re with and what they say. And I’m sure you’ve entered hundreds of search queries to Ubatoo with the exact right keywords. If I’m right about who’s most likely going to make it onto a watch list, I mean if the ACCL is right, you’re probably already on some watch list, or will be soon.”
Molly thought for a few seconds before responding. “So, what’s going to happen now?” There was no worry in her voice.
“I don’t know. Sebastin told me a while ago that anyone who is on their list will get an e-mail or a phone call from the ACCL, and they’ll tell you what you can do. Do you know if you’ve already been contacted by them?”
“No, I haven’t. I don’t recall having seen anything like that in my e-mail. No phone calls either. I would have told you.”
“Are you sure about the e-mails? You should have received it a few days ago. You haven’t seen anything out of the ordinary?”
“I get about a dozen out-of-the-ordinary e-mails a day. You know the stuff I write about on my message board; people are writing me all the time. Sometimes to my Sahim Galab account, or sometimes to my Zakim account, or any of the others, too. I haven’t seen anything unusual.”
“No, but anything besides that? Addressed to you, in particular, not to Galab or Zakim.”
“No. Not that I can think of.”
Stephen waited a second before continuing. He wasn’t quite sure how to proceed at this point. “Yes, you have,” he said quietly.
“What?”
“I checked your e-mail account before I came here. You’ve already gotten the e-mail. It’s been sitting in your inbox for a few days now. There’s a meeting tomorrow night. A few other people were also on the e-mail.”
“What? How do you know all this? I never said you could do that. Are you always checking my e-mail accounts?” She pulled herself back as far away from Stephen as she could in one abrupt motion. “Is that what you do all day?”
“No, no. Remember, you gave me your password?” He leaned in closer toward her and whispered, “‘Mollycoddle.’ Remember, you told me that the first time I was at your apartment?” Of course, he didn’t bother to explain that he wouldn’t have needed the password, and that her guess was right: This was exactly the type of thing he did all day. There would be time enough for that later.
“I didn’t even have this e-mail account then.”
“Everyone uses the same password for all their accounts, Molly. It’s not much of a secret,” Stephen said.
She was teetering between two options: storming out now or trying her best to wait the anger out. “You still read it without my permission.” She exhaled a long restrained breath but said nothing more. The scornful look on her face was enough. “This better be damn good,” it screamed.
Stephen didn’t wait. He needed any opportunity to continue. “The e-mail from ACCL said there was a meeting tomorrow night. Someone from ACCL will be coming to talk to you and, from what I could tell, a few other people in the area who made it onto the list as well. I’ve never been to any of these meetings, but I’m going to guess he’ll review what’s going on and tell you what you should do next. That’s all I know. I would ask Sebastin for details, but he’s out of town.”
“This is a giant misunderstanding. I don’t have to explain it to you, you understand that. I’m just studying the people who post on my web site. I’m not planning to join any group and I don’t believe any of the stuff posted. You still think I should go?”
“I think that’s why ACCL is around. I would imagine these mistaken cases are exactly what they want to resolve.” Stephen waited a few seconds to let it sink in, “I don’t think you have much of a choice. You need to find out as much as you can. I don’t know enough about ACCL to tell you what they would say, but if everyone is as passionate about helping as Sebastin is, you’ll be glad you went.”
“You think I’m in some kind of trouble?”
“I don’t think so. I’m sure it’s going to be fine. But I’d rather be too cautious than not enough in this case.”
Now Molly was angry with herself. “I can’t believe I missed that e-mail. What if I hadn’t seen it in time?”
“It happens. Just make sure you go.”
“Thanks for finding me,” she said with a very clear hesitation in her voice. “Sometime you should tell me more about what you do with ACCL.”
Molly understood enough only to know that she didn’t fully understand what was happening. She couldn’t predict how dire her situation was. Even online shopping companies who supposedly watched her and her shopping habits couldn’t recommend the music or books she cared about. If they couldn’t do that, how much should
she trust Stephen’s analysis? Every tech person she had ever known was overly confident about his own work. And there was enough empirical evidence from all of her studies to indicate that all the computer models, no matter who developed them, never captured the full picture. After all, that was what Stephen had created—another computer model of what it takes to be on a watch list. Maybe his analysis was faulty. For that matter, how much should she trust ACCL? They couldn’t have direct knowledge of what it took to get on any of these lists either. Moreover, what did it really mean to be on “a list”? If she believed the posts on EasternDiscussions, most of the members of the forums were likely to be on any number of such lists, but they went on posting and living their lives just fine.
Stephen, on the other hand, had reason to be more pessimistic. He couldn’t help but worry that it was his analysis that would eventually put Molly on a watch list if she hadn’t been on one already. The line between cause and effect was blurry. Rationally, he wasn’t guilty; he had meant only to try to replicate the signals ACCL said were being used by the NSA. He wasn’t the cause of Molly being on the list, was he? No, that wouldn’t make any sense. Now, there was no going back. He couldn’t just delete his e-mail and ask Sebastin to delete it as well and start over. What if Sebastin had forwarded it already? Besides, there was always a chance that somewhere in Ubatoo’s network there was a backup of that file in which her name was listed. It was only a matter of time before someone stumbled across the list again with a fresh pair of eyes and figured out what the list was for.
The Silicon Jungle Page 23