The narrator continued as this episode faded, but the words were lost in the light.
Andrew motioned to Yuri. Yuri tossed the camera to Andrew.
“One more picture, just for me,” Andrew explained. Flash. A room filled with harsh white light. The blanket did nothing to hide the body underneath.
Stephen fell into a quiet, unadorned, sleep. He held on a bit too tightly to Molly, so that when it came time for her to get up and press reload, she didn’t bother to move. It could wait.
His breathing had finally calmed down, and she didn’t want to wake him. They both could use the sleep. Just like every other night, she would face her own set of dreams in a few minutes, and it would be good to know that someone was there with her. They slept for a full five hours, the longest they had been with each other in days, holding on to each other through a pall of altogether too vivid dreams, nightmares, and complete and utter exhaustion.
-A FIVE-STEP PROGRAM-
Hallucinations and Archetypes
July 16, 2009.
Stephen woke to an empty apartment dappled with morning light. A note was waiting next to him, instead of Molly.
Some really nice results just this morning! Looks like traffic is finally picking up. Thanks for all your help. Whatever Andrew did worked! More traffic and people posting on my site than I could have imagined. Let’s try to talk tonight. There’s so much to tell you about. We haven’t talked in days.
Love, Molly
He held the note in his hand for a few seconds. If his imagined television audience could have seen the intensity of expression in those moments, they would have suspected nothing short of a man on the brink of foretelling the fate of humanity.
The reality, though, was less dramatic. Molly had never written a real physical note to him. Come to think of it, he couldn’t recall the last time anyone had handwritten a note to him. Molly had written plenty of e-mails, but they were short and to the point. She had never signed it with “love,” or anything else for that matter.
But here was a physical piece of paper in his hand, more than half of which talked about work. What was wrong with him? Just like the first night in this apartment, setting up Molly’s computer, this was too much thought about too little. For all he knew, it was probably how she signed all her notes. This was unquestionably the first and surest sign he was growing older, and becoming feeble-minded. But even more probable, he suspected, was that this overly long, and entirely unwarranted, pause for introspection was due to his overwhelming exhaustion. Likely, he may well have found a handwritten breakfast menu equally intriguing in his current state.
But then again, things were going well. After a four-year-long interruption, he was finally enraptured with his research again. All of this was heightened by Molly—someone who felt the same way about her own work. Who else would not only understand, but also experience, this kind of consuming drive? Perhaps, as he reconsidered it again, the breakfast menu really wouldn’t have been an adequate replacement.
Despite all of these thoughts and Molly’s request to talk that night, there was still a good chance he wouldn’t make it back in time. He had an enormous amount to do today before his meeting with Sebastin (it had been seven days since he had last spoken to him), and there was a company-wide meeting that he should attend as well. Last but not least, his curiosity about Kohan’s night was hard to contain.
Stephen had promised nothing specific to Sebastin. Nevertheless, he wanted to impress. He had always lived by the motto of under- promising and over-delivering. Plus, there was the chance that Sebastin might remember to put in a good word with Atiq. And the fact that Atiq clearly thought highly of Sebastin wasn’t far from his thoughts.
With a hot cappuccino on the desk, and two handmade chocolate croissants from the pastry cart ready to go, he opened the attachments in the e-mail that Sebastin had sent a week ago. He quickly scrolled through the list of 960 books on his screen. Just as Sebastin had indicated on the phone, there was no discernible pattern. The list contained cookbooks, textbooks, fiction, non-fiction, history books, agriculture, literature, philosophy, engineering, religion, mathematics, and more.
Normally, this in itself would be an interesting puzzle: How did all these books make it onto the same list? Maybe if he hadn’t delayed this project until the last minute by working so much on JENNY, there would have been time to figure it out.
Time for work.
Like their public-facing search engine that brought Ubatoo to such fame, there was an equally powerful, but never publicized, search engine for everyone’s e-mail that was available only to Ubatoo’s data-mining group. People outside Ubatoo were able to search their own e-mail, of course, and the data-mining group searched everyone’s e-mail. He typed the first book’s title, “World Survey on the Role of Women in Economic Development,” and submitted the query.
Instantaneously, several hundred e-mails came up that contained the full phrase. He had to choose “Sort by sender,” “Sort by recipient,” or “Sort by date.” He sorted by the sender—only a dozen or so unique senders had composed e-mails containing that book’s title in the past year. He clicked over to the “instant message” tab to see if anyone had sent any instant messages containing the title of the book. Only two people had, but that was just for this one book.
Step 1: Set up a program to send all the titles of the books through the internal e-mail and instant message search engine and gather the names of the people who had ever used the titles in any of their correspondences, and also note how often they had used them.
The time to write the program was far longer than the time it took to run it. After writing the program in about twenty minutes, within a few seconds of starting, it ended. A bit anticlimactic. But now any e-mail or instant message that contained the titles of any of the 960 books was accounted for. There were 73,291 people who had mentioned the books at least once in their correspondence.
Step 2: Let’s figure out who bought these books. The purchasing records of stores that used Ubatoo’s credit card and transaction processing systems were the easy ones to access. There were only some tens of thousands of those stores to look at. Those would take only a few minutes to scan through once started.
Within a few minutes, at least several hundred machines somewhere in Ubatoo’s cloud were dutifully scanning records from the past year trying to uncover all the people who bought any of the 960 books in the last year. Only 29,084 people bought any of the books listed, a terribly small number for that many books. These were clearly not bestsellers. That analysis was far too easy. “Okay, Sebastin, watch what we can do now,” he said aloud.
The problem with a number like 73,291, or 29,084 for that matter, was that, although it was good to know, it was way too large to be useful. Hopefully, ACCL wouldn’t send out a warning e-mail to all of these people and scare them unnecessarily.
He needed to narrow it down. In quick succession, he tried a few guesses to prune the number down to something more manageable. The one that worked well was to just look at the people who had purchased more than one book from the list. That made sense, since if they had bought only one, it may be just a coincidence. There were only 2,602 people who had bought two or more books. That’s a better number.
So 2,602 people bought at least two different books from the list. How many of them had actually sent an e-mail or instant message about them, too? That list had been created in step 1 and was waiting to be used.
Step 3: Find the people who appeared both on list-1 and on list-2 and merge the lists. These are the people who wrote about the books AND bought at least two. That’s probably the set of people who are most into the books.
The set of 2,602 that he found earlier only reduced to 2,423 after step 3 was completed. Not much of a reduction, which meant that if someone bought two of the books on the list, he was likely to talk about them, too. Nonetheless, this was definitely the set of people to concentrate on—buyers who cared enough about the content to discuss it. I
f Sebastin was going to contact anyone, this was a good set of people to start with.
But books? Why books?
It felt stupid. Who looks at book reading patterns and thinks they know everything there is to know about someone? No, that doesn’t make sense. The fact they bought these books was an indication of something else. The U.S. government, no matter how paranoid or ridiculous they seemed, couldn’t really care that someone just bought a book, could they? What they cared about was that if the person was the type to buy one of these books, then that person was the type to do something more dangerous, something actually worthy of being watched. But why not just find out what that is—Ubatoo had the data. Let’s figure out what type of people these people actually were.
To do this, he needed to look beyond books and see what else the 2,423 had in common. If these people really participated in activities that put them on a watch list, there were probably a myriad of other patterns to discover from the things they did online. Back to Ubatoo’s repositories to find all of their recorded actions for years past.
There wasn’t enough time to do this nearly as comprehensively as Stephen would have liked. He selected just four from the hundreds of things Ubatoo knew about its users, and looked for patterns in them. For each of the 2,423 people, he examined:
1. What web sites they had visited in the last two years.
2. What they searched for on Ubatoo.
3. What products they most often bought using Ubatoo’s credit cards.
4. Where they had traveled to, as inferred by Ubatoo’s records of 1–3 above.
All of these were already conveniently stored in the profiles of the 2,423 people that had been diligently created over years. This information was ready, patiently waiting to be used.
What he found was that, of the 2,423 people, many consistently went to Mideast news and entertainment sites, but also to political and religious discussion boards, sites about lectures given by people he had never heard of who spoke of religious points he had never encountered, and even some private sites that he would have liked to look at in depth, but just didn’t have the time. That’s just attribute #1. The other attributes, #2–#4, held patterns, too—there were dozens of common searches performed by the 2,423, many out-of-the-ordinary products they had all purchased, and many places in common to which they had traveled. These patterns were just as important, if not more important, than the books they read.
He would have liked to have systematically matched all of Ubatoo’s 200 million users in the U.S. to each of the profiles of the 2,423 people. This would have uncovered whether other users existed out there like one from the set of 2,423. But time—time was running out . . . He had promised Sebastin something today, and doing this would take far too much time.
Instead of comparing all of Ubatoo’s users to 2,423 profiles, he created a new hypothetical profile—a single archetype from the 2,423: Stephen called her Lucy. Lucy would be constructed through synthesizing all the patterns he had just uncovered. If most of the 2,423 had visited Aljazeera.net many times, then Lucy had too. If most of them had traveled to countries in the Mideast, then Lucy had as well. Whatever the 2,423 most commonly had searched for—that’s what Lucy had searched for. Lucy was the epitome of the patterns Stephen had found. He made this computer hallucination in the likeness of what the 2,423 would be, if they were just one person.
Step 4 (for fun) [Controlled Hallucinations]: Create a new profile that is the synthesis of the common patterns in the 2,423. Label this profile “Lucy.”
With only 2,423 people to examine, Lucy was created, and she developed all her web-surfing and traveling habits faster than the time it took for him to type his name.
Lucy wasn’t some outlandish artificial-intelligence dream come to life in a futuristic movie. She wasn’t a sentient being in the making, hell-bent on eliminating the human race. No, far from it. She was just a list of a few words and numbers in the form of a profile, like all the other profiles that sat in Ubatoo’s repositories. But Lucy had a bigger job to do than just exist as a computer hallucination that Stephen had fashioned.
Though Stephen didn’t have time to check which of Ubatoo’s users matched each of the 2,423 people individually, he did have time to see if they matched one profile: Lucy’s profile.
Step 5 (for even more fun): Find other people who may or may not be on the original list of 2,423 who match Lucy’s profile. These are the people who should probably be on the ACCL list, too, but may not have been uncovered by only looking at their book buying and reading patterns.
Thanks to Jaan’s system, the last question, which would normally take days to answer, would be completed in just a few hours. He set a high priority for the job—over 6,000 computers somewhere in Ubatoo’s cloud obediently aborted whatever project they were working on, and immediately started on Stephen‘s.
Stephen helplessly watched as the progress bar made its excruciatingly slow crawl to 1% . . . 2% . . . 3% . . . 4% . . . 5% . . . 6% . . . Finally, at 6% he convinced himself that it would be okay to leave the computer—at least long enough to grab some lunch and bring it back to his desk to watch its climb to 100%. He returned in time to see 28% . . . 29% . . .
Hopefully it would be done soon. There was so much more he could do.
-OVER-DELIVER-
July 16, 2009.
“Stephen! How are you? Give me a second. Let me clear out my office. I was just finishing up a meeting,” Sebastin said enthusiastically as soon as he recognized the caller.
“I can call back later . . . ,” Stephen said. As he was accustomed to at this point, nobody stuck around to hear his last words. He was talking to himself again.
About a minute later, Sebastin returned to the phone after ushering the others out of the room. “Alright, the room’s empty. I have the files you e-mailed me open on my screen. So what have you found?”
“First things first,” Stephen replied. “Let me tell you what I did. I started by doing exactly what we talked about, finding everyone who bought one of the books. The procedure I used was to . . .” and he continued his monologue for three minutes. He knew it was unlikely Sebastin was interested, but he had also learned from working with all the other advertisers in the past few weeks that his results never seemed as impressive to them as when they were derived in some mostly unintelligible manner that suitably confused them. “. . . so, in the first file I sent you, are all the people who purchased a book from your list within the last year.”
“That file looks huge. How many people did you find?”
“29,084. I also included their e-mail addresses. I thought you would probably want those.”
“Wonderful, this is truly wonderful,” Sebastin said.
“That’s just the beginning,” Stephen said happily. “That number was too big, of course. So, first, I found all the people who bought at least two books. Then I went ahead and checked which ones of those wrote about the books in their e-mails or instant messages. These people are the high-priority ones. Warn them first. They’re the most likely to be talking about the books in public, writing messages about them, and so on.”
“You scanned through all of their e-mails? Really?”
“Just the ones who used our own e-mail service. Fortunately, from what I remember, that’s most of them. Anyway, I found 2,423 people. The entire processing took less than two minutes.” It was always fun to boast about Ubatoo’s massive computing resources. A couple years ago, nobody wanted to hear about how many machines were used or how long the processing took. It was just in the past year that the number of machines grew so astronomically large that even the average advertising client, or charity client in this case, cared.
“Amazing. Just amazing.”
“Yeah. It really is. But, wait, there’s more!” he said. He always used this line in his presentations to advertisers. It was a bit cheesy, a bit too much like he was selling a set of knives in an infomercial, but if the advertiser liked the results already, this only endeared them f
urther to the presentation.
“Alright. I’m all ears.”
“Now, this is something I didn’t quite understand, but I checked it a bunch of times to make sure it was right. In the process of doing this analysis, I had to cluster the buyers and books.” Without Sebastin’s expression to judge, Stephen couldn’t be sure whether Sebastin had any idea what clustering meant. “I mean, I grouped the results together so I could find correlations between the books by using the people who bought them as signals. Basically, I created a bipartite graph and propagated the signals originating from each node . . .” Too much again. He stopped talking there, realizing all this was probably far more gibberish than Sebastin had bargained for.
“What I’m trying to say is that if I look just at the people who bought more than one book on your list, almost all of them only bought books from a tiny set of sixty books. Doesn’t that seem strange?” Stephen waited for Sebastin to digest this.
Sebastin didn’t respond, so Stephen tried to explain it one last time, “Out of the 960 books, 900 of them were random—they had nothing to do with that tiny set of 60. In other words, 900 books were just a distraction. It was almost like someone in your group just stuck them in there to make this task more challenging for me. Does that sound plausible?”
More silence on the other end of the phone. Stephen waited patiently, but when a minute had passed without a sound, he had to say something. “Sebastin? You still there?”
“Yes, yes. I’m here. Sorry, just thinking. That is very strange. I have no explanation for why that is.” Some typing started in the background. “That is the small list of books in your third attachment? I’ll look into that. Let me think . . .”
The Silicon Jungle Page 16
