Echoes of Sherlock Holmes: Stories Inspired by the Holmes Canon(115)

I gave him a moment and then prompted, “But you found something?”

He returned from whatever distant mental plain he’d been slogging over. “They wanted a big corpus to do information cascade analysis on. Part of a research project with one of the big unis, I won’t say which, but you can guess, I’m sure. They’d done a new rev on the stream analysis, they were able to detect a single user across multiple streams and signals from the upstream intercepts—I mean to say, they could tell which clicks and messages on the fiber-taps came from a given user, even if he was switching computers or IP addresses—they had a new tool for linking mobile data-streams to intercepts from laptops, which gives us location. They were marking it for long-term retention, indefinite retention, really.

“I—”

Here the fellow had to stop and look away again, and it was plain that he was reliving some difficult issue that he’d wrestled with his conscience over. “I was in charge of reviewing the truthed social graphs, sanity-checking the way that the algorithm believed their chain of command went against what I could see in the intercepts. But the reality is that those intercepts came from teenagers in a chat room. They didn’t have a chain of command—what the algorithm fingered as a command structure was really just the fact that some of them were better at arguing than others. One supposed lieutenant in the bunch was really the best comedian, the one who told the jokes they all repeated. To the algorithm, though, it looked like a command structure: subject emits a comm, timing shows that the comm cascades through an inner circle—his mates—to a wider circle. To a half-smart computer, this teenager in Leeds looked like Osama Junior.

“I told them, of course. These were children with some bad ideas and too much braggadocio. Wannabes. If they were guilty of something, it was of being idiots. But for the researchers, this was even more exciting. The fact that their algorithm had detected an information cascade where there was no actual command structure meant that it had found a latent structure. It was like they set out knowing what they were going to find, and then whatever they found, they twisted until it fit their expectations.

“Once we have the command structures all mapped out, everything becomes maths. You have a chart, neat circles and arrows pointing at each other, showing the information cascade. Who can argue with math? Numbers don’t lie. Having figured out their command structures from their chat rooms, we were able to map them over to their mobile communications, using the session identifiers the algorithm worked out.

“These twerps were half-smart, just enough to be properly stupid. They’d bought burner phones from newsagents with prepaid SIMs and they only used them to call each other. People who try that sort of thing, they just don’t understand how data-mining works. When I’ve got a visualization of all the calls in a country, they’re mostly clustered in the middle, all tangled up with one another. You might call your mum and your girlfriend regular, might call a taxi company or the office a few times a week, make the odd call to a takeaway. Just looking at the vis, it’s really obvious what sort of number any number is: there’s the ‘pizza nodes,’ connected to hundreds of other nodes, obviously takeaways or minicabs. There’s TKs—telephone kiosks, which is what we call payphones—they’ve got their own signature pattern: lots of overseas calls, calls to hotels, maybe a women’s shelter or A&E, the kinds of calls you make when you don’t have a mobile phone of your own.

“It makes detecting anomalies dead easy. If a group of people converge on a site, turn off their phones, wait an hour and then turn ’em on again, well, that shows up. You don’t have to even be looking for that pattern. Just graph call activity, that sort of thing jumps straight out at you. Might as well go to your secret meeting with a brass band and a banner marked UP TO NO GOOD.

“So think of the network graph now, all these nodes, most with a few lines going in and out, some pizza nodes with millions coming in and none going out, some TKs with loads going out and none coming in. And over here, off to the edge, where you couldn’t possibly miss it, all on its own, a fairy ring of six nodes, connected to each other and no one else. Practically a bullseye.

“You don’t need to be looking for that pattern to spot it, but the lads from the uni and their GCHQ minders, they knew all about that pattern. Soon as they saw one that the persistence algorithm mapped onto the same accounts we’d seen in the chat rooms, they started to look at its information cascades. Those mapped right onto the cascade analysis from the chat intercepts, same flows, perfect. Course they did—because the kid who told the best jokes was the most sociable of the lot, he was the one who called the others when they weren’t in the chat, desperate for a natter.”

I stopped him. “Thinking of your example of a group of phones that converge on a single location and all switch off together,” I said. “What about a group of friends who have a pact to turn off their phones whilst at dinner, to avoid distraction and interruption?”

He nodded. “Happens. It’s rare, but ’course, not as rare as your actual terrorists. Our policy is, hard drives are cheap, add ’em all to long-term retention, have a human being look at their comms later and see whether we caught some dolphins in the tuna-net.”

“I see.”

“We have their ‘command structure,’ we have their secret phone numbers, so the next step is to have a little listen, which isn’t very hard, as I’m sure you can appreciate, Mr. Holmes.”