CHAPTER 9

Barely a week later, Melody found herself in a conference room inside one of the most secure buildings in the world. She remained in her wheelchair, which her doctors insisted she continue to use for the time being, and her hosts had given her a place at the foot of a long conference table. The other positions around the table were marked by tidy stacks of folders and paperwork which a man had laid out just prior to the meeting. He stood in the corner, talking quietly with Mr. Constantine. Apparently, Mr. Constantine was in the military, but while he was certainly built like a military man and carried himself like one, she had not yet seen him in any kind of uniform. He was clean shaven, but he tended to dress casually, and his hair was a little messy. The other man—she had already forgotten his name, but it was something short and distinct. Song? Hong? In any case, he was obviously an office worker, and she could see him being an employee of the agency that famously occupied this building. He had called himself an analyst, and that was what he looked like.

Men in suits (and one woman, in a suit-jacket and skirt) began filtering into the conference room and taking up available seats. They greeted her politely but did not introduce themselves, nor ask her identity. They seemed to expect her. Melody found that she was more than a little nervous. This seemed to her like the appropriate level of concern to give a matter that had already gotten her shot, but all the same it was intimidating. The paperwork alone had been intimidating. The NDAs she had signed promised dire, life-ruining consequences should she breathe a word of any of the covered facts to anyone they did not approve. They had given one to Doran as well. He was in a waiting room somewhere nearby. Despite having witnessed little less than she, and being privy to everything she had said to date, he apparently was not to be included in this meeting or others going forward. They said the rest of her project teammates had been given NDAs as well and were for now under police protection. She wondered if that might also be observation. Still, if that was the price of not being killed, it was a small one to pay.

“Miss Ritter, can I get you anything?” asked the analyst. What was his name? “Water? Coffee?”

“Some water, I guess? Thank you.”

“No problem.”

“I got it,” said Mr. Constantine, ducking out. The other attendees began taking their seats.

“While we wait for Mr. Constantine to come back, I’ll go ahead with introductions,” said the analyst. “I’m Roy Sing, Strategic Analysis. To my left, Joshua Carrington from Cyber Warfare. Down at the end, Special Agents Maia Raines and Andrew Wheeler from the Bureau. Uh, Counter-Intel and Cyber-Crimes, right?” They nodded. “Bob Nelson representing our Clandestine Activities Directorate, and George Fry, my supervisor in Strat-A.”

Polite greetings were exchanged all around.

“And, of course, the guest of honor, Miss Melody Ritter.”

“Miss Ritter,” said Fry, standing up and extending a hand to her. “On behalf of all of us, I want to thank you for coming. And I want to express our deepest sympathy for what you’ve been through and our condolences on the loss of your friend, Mr. Brigham.” That was Cookie. Billy Brigham, his name had been. They had told her. There was now an ongoing investigation into his death as a possible homicide.

She shook his hand from her wheelchair. “Thanks,” she managed, and then she felt that might not be sufficient to the occasion. “Um, thank you for investigating. For helping us.”

“That’s what we’re here to accomplish,” said Fry, taking his seat again. “Roy?”

“Right. The following material is classified up to Top Secret, no-foreign, and we’re working on a program name. That should come through today, so expect it. For now, treat everything like it’s SAP, and only the people in this room are read-in, with the exception of Miss Ritter, who is here to provide her testimony.”

Nods all around.

“For those of you who don’t know, an attempt was made on Miss Ritter’s life eight days ago in the parking lot of her local grocery store. She survived a single gunshot wound to her chest. We’ll come to the details, but we’re very lucky to have her here. Miss Ritter, I want to reiterate what my boss said. We’re grateful for your bravery.”

Now she began to feel a little embarrassed. “I’m really… not that brave. It was pretty terrifying.”

“Bravery’s not about not being scared,” said one of the men. Bob? Clandestine Something Something. “It’s about doing the right thing even when you are scared.”

“Absolutely right,” said the female Special Agent.

“Well…” said Melody. What did one say to all of that?

The door opened, and Mr. Constantine entered. “And our last attendee, Chief Warrant Officer Robert Constantine, Army SF Detachment 544. Our other eye-witness.”

Constantine nodded to them as he set a bottle of water down in front of Melody and then went to his seat.

“Okay, a quick review of the material in front of you.” As one, they began leafing through their folders. Melody had the sense that there was some ritual to a meeting like this one. “The police report from the scene of Miss Ritter’s attack. Miss Ritter’s statement regarding that event. The police report dated twenty-seven November regarding an incident of menacing at the residence of Andrew and June Ritter, Miss Ritter’s parents, to which she was a witness. A coroner’s report regarding the death of one William Brigham, Miss Ritter’s associate.”

“Lord…” said someone, recognizing the pattern. A few sympathetic eyes were on her, and a few that seemed as suspicious as they were sympathetic. The female Special Agent was watching her carefully.

“After action reports from Operation Press Hook—which I need to remind you Miss Ritter is not read in on at this time. With apologies, ma’am, we do have to keep information compartmented.”

“Of course,” she said.

“Following that, analyses compiled by me over the past year regarding instances I believe represent a pattern of activity that includes these events. We’ll come to all of that. And finally, the work of Miss Ritter herself, pursuant to her master’s thesis project, which she will explain for us. Everyone good?”

Nods all around.

“All right. Because a great deal of this centers around Miss Ritter’s testimony and the master’s project on which she and several of her classmates have been working for the past year, I want to begin by having her describe her project for us. Miss Ritter, would you mind going through it?”

“And please bear in mind that most of us are not coming from a computer science background,” added Fry. “Layman’s terms, please.”

“Okay. You… you want me to start now?”

“Please. And take your time. This is not a deposition. We’ve asked you here to educate us on your work and what you’ve experienced.”

“Okay, well… How much do you all know about AI?”

Silence. The others looked at one another. Finally, one of the men said, “I know a little, but it wasn’t my field of study. Basically it’s machine-learning, right?” He was the cyber-warfare guy.

“Yeah.”

“I think you’d better just start at the beginning,” said the Clandestine Something Something guy.

“Okay, well, AI is not really this big, magic thing that people think it is. Really, it’s…” Melody found herself sinking into her inner world as she struggled to figure out how to articulate the subject at hand. How could she tackle this in a way that they would understand, in a way that did not rely on terminology and concepts peculiar to a computer science background? They all waited quietly, staring at her, while she thought. “It’s a recognition engine,” she said at last. “It’s a way to write a program to recognize or classify things that aren’t easily recognized with a specific algorithm.”

“Could you explain that?” said the cyber-warfare representative.

“Well, so, I have to—” No, that was already too far. She needed to back up to the very beginning. “Okay, so all computer programs are algorithms. An algorithm is…”

“A standard way to solve a problem,” he offered.

She nodded. “Like, a step-by-step way, that always works, and doesn’t take any intuition. It’s just a process. Something a machine can do. Step 1, step 2, step 3, step by step, until it’s done. Like following a recipe. You don’t have to understand a recipe. If it’s precise enough, and you follow it precisely, you’ll get a cake, even if you don’t understand why.

“So, if you want to solve a problem with a computer, you come up with an algorithm that the computer can perform, and you write code to represent that algorithm. Most problems have algorithms to solve them. But recognizing and classifying things isn’t always simple. Sometimes it’s more intuitive. Like, if you want to identify the colors in a picture, that’s easy. For each pixel, you measure the amount of red, green, and blue, and record it in a table. But if you want to know what it’s a picture of, that’s a lot harder. How do you tell a computer to pick out all the pictures of boats from a set of pictures? You can’t do color analysis, because boats can be any color. You can’t really do shape analysis, because a boat can have almost any shape depending on what angle you look at it from and what kind of boat it is. So this sort of classification problem—'boat’ or ‘not boat’—was impossible to automate for a long time.”

The people around the table seemed to be following so far, and the helpful one smiled and nodded to her, so she continued, feeling just the tiniest hint of confidence beginning to emerge.

“So, the breakthrough was what we call ‘machine learning’ algorithms, like neural nets. People thinking about how the brain does this activity so efficiently began to look at the structure of the brain and think about structuring a program the same way.”

“So a neural net simulates how the human brain works?”

Melody raised a hand. “Not at all. I should really be clear about that, because that’s a big thing nobody really gets, I think. A neural net is just a kind of computer program. Some people like to say it’s ‘vaguely inspired by’ the way a human brain works, and that’s about right. It’s like… you know how some movies are ‘based on a true story,’ and some movies are ‘inspired by a true story?’ And ‘inspired by’ basically means a true story gave somebody an idea, and then the movie comes out and it’s nothing like what actually happened?”

Several of them smiled, which was a relief. Their earlier courtesy had not been quite sufficient to reassure her that they were human.

“So yeah, neural net programs are ‘inspired by’ the human brain, but don’t actually work anything like the human brain. Really, all a neural net program is is a bunch of weighted sums. Let’s say I want to take a set of, say, six features—six measurements—and determine from those measurements… Er, let’s start over. Let’s say I want to determine if a thing is a boat or not a boat. And I decide that I can do that by taking six measurements. This is just a thought experiment, so the features could be anything. Length, width, color, mass, whatever. Let’s just say that I think I can do it with these six features. So I get a pile of a million objects, and for each one I measure these six parameters, and feed those into the program, and the program will spit out an answer: ‘boat’ or ‘not boat.’ That’s my goal.

“So, how a neural net program works is, for each of those six features, I convert it into a number. So, like, length and width are already numbers. I come up with a number to represent its color—like from a color table or whatever. It doesn’t really matter, as long as each possible color has a unique number. And so on. Each feature I convert into a numeric representation. Then, I multiply each feature by some random number, called a weight, and add all the results together.”

“It doesn’t matter that they’re different units?” asked the male Special Agent.

“Nope. It doesn’t even matter if some of them aren’t units. They could be arbitrary numeric mappings. Like, I could feed in the material of the object with an arbitrary numeric map. 1 for wood, 2 for metal, 3 for plastic. Doesn’t matter.”

“Okay…”

Melody grinned. This was the great part about machine learning in general and neural nets in particular, the part that seemed like magic. It was the part that had made her interested in the field. “So we take a sum of all the parameters, each one multiplied by a random weight—a weight meaning just some random number. Doesn’t matter what the weights are, to start with. Anyway, think of that as one neuron. In this case, we had six features, so each neuron has six inputs, and it has a different random weight for each input. It takes in all the features, multiplies each input by a different weight to get six products, adds the products together, and outputs the result.

“Now, let’s add a few more neurons. They take the same inputs, our six features, but they each have a different set of six random weights, so they each put out a different result. That would be… Er, so, that collection of results, one result from each neuron, would be the first layer of the neural network. We can add another layer of neurons if we want. Each of these neurons takes all the results from the first layer, all of the first layer’s outputs, as inputs, and multiplies each input by a random weight, and adds them together. So if our first layer had ten neurons, then each neuron in the second layer will have ten inputs, and ten weights, one for each input. The second layer has many neurons as well, however many we want, each one outputting a different weighted sum because it uses different random weights.

“And we can keep doing this as much as we want. A layer can be as wide as we want, meaning it can have as many neurons as we want, and each one takes all the inputs from the layer below it, multiplies each input by a different weight, adds them all together, and outputs the result. And we can have as many layers as we want. A lot of layers is what we would call a deep neural network. So you have the input layer, which is just however many input features you have. Then you have all the layers of neurons, what we call the hidden layers. Finally, we have the output layer. The output layer has one neuron representing each possible answer. So we could say we have two: ‘Boat’ and ‘not boat.’ Each one takes all the outputs of the last hidden layer, multiplies each input by a random weight, and adds them together. And whichever output neuron comes up with the higher number, that’s your answer!”

She grinned. This sort of thing made her very happy. It made her happy in her brain.

The audience seemed at a bit of a loss, which was to be expected. They had no way to know that she had just told a very esoteric joke at their expense, by ending the story too early.

“So, I don’t get it,” said the male Special Agent. He was the Cyber Crimes agent, so he probably also had some computer background. He and the one other guy who was from Cyber Warfare. Melody desperately wished she had a better head for names. “Just by taking weighted sums, how does that ensure we get the right answer?”

“Oh, it doesn’t get us the right answer. Not with random weights. It just gets us a random answer. We feed in the measurements from one of our unknown-objects-that-might-be-a-boat, and the neural net takes a bunch of weighted sums using random weights and spits out an answer, and of course the answer is probably wrong. But, we know exactly how wrong. We know, as the trainers, that this thing is actually a boat. So we know that the ‘boat’ neuron should be a one and the ‘not boat’ neuron should be a zero. So we can take the outputs that we got and see what the difference is between the answers we got and the answers we wanted to get—the answers we should have got, knowing that this was in fact a boat. And, believe it or not, with a bit of calculus, we can calculate how to adjust all the weights, all the way back through the network, so that we get an answer closer to what we should have gotten. So we do that. Adjust all the weights for all the neurons. Then we feed it another random object.

“Again, we already know the correct answer. This is what we call the training data set. We already know whether each of these objects is a boat or not. So we feed in the six features of the next object, and again we probably get a wrong answer, but again we can calculate how to adjust the weights so that the neural net puts out the correct answer, and we do that. And we keep doing that for, say, ten thousand objects in our training data set. And after adjusting it over and over to give the right answer for ten thousand sample objects, the neural net should be pretty well adjusted. So then we feed it a test data set.

“The test data set is, let’s say, a hundred thousand random objects. Again, we know the correct answers already. But this time, we’re not going to make an adjustment after each one. We’re just going to let the machine—the program—do its thing, make its guess, and see if it was right or wrong. So for each item in the test data, we feed in the six features, and the neural net puts out an answer, ‘boat’ or ‘not boat,’ and we measure how many it got right and how many it got wrong. And if it got ninety percent of them right, then we’re doing okay, but we need to make some tweaks to our neural net design. We can make some changes and train it again on our ten thousand training objects, and test it with our hundred thousand test objects, and this time it gets ninety-nine percent correct. And we keep going, making tweaks until it’s getting the right answer with enough accuracy for our application.”

She fell quiet again, smiling. As far as she was concerned, it was a miraculous and staggeringly cool bit of software technology. Again, it made her happy in her brain, which was the part of her in which she mostly lived.

“So that’s it? It’s just a big pile of weighted sums, and you just keep adjusting the weights until it consistently gives you the right answer?”

“That’s it. Now, obviously I’m leaving some stuff out. There’s activation functions—say, we want each neuron to output a number between zero and one, so we have a function that converts the weighted sum for that neuron into a value between zero and one on a gradient. And we might bias certain inputs or outputs. We might perform some regularization to defend against overfitting. Oh, and of course, if we’re trying to recognize something in a picture, it won’t be, like, six or eight features—we’ll feed in the whole picture, pixel by pixel, and do some fancy stuff with operating on certain segments at a time. But none of that changes the basic structure, how it works. The network I just described is what we’d call a multilayer perceptron. But from there we could make it more complex, change some of the ways it connects within itself, to solve other kinds of recognition problems. More complex ones, like image recognition, or speech recognition. But the basic concept is the same. And if you can make it recognize things, then you can go further. Have it start creating pictures, or filling in missing pieces, that match how it’s been trained.”

“Wow,” said one of the men.

The female special agent was smiling at her. Melody was not sure what that meant, but she took it as a good sign.

“So this is what you were working on, for your thesis?” asked the Cyber Warfare guy.

“Not exactly. So, all of this, we talk about neurons and synapses and stuff, but it’s really just linear algebra. An array of inputs dotted with an array of weights.” She caught herself. “Sorry, an array of inputs and an array of weights, and we take the dot-product. Repeat, repeat, repeat. In fact, it’s so pretty much completely just that, just dot products of vectors, that we do it mostly on videogame computers, because three-D graphics is the same kind of math, lots of vector multiplication. So we use computers with good graphics cards to do our processing. Lots and lots and lots of dot products. Millions and millions per second, because a good neural network will have—Well, if I say a neural network has a million neurons, a thousand wide by a thousand deep, all I really mean is that I’m doing a million dot products, and I want to do that as fast as I can. But none of that is really anything like how an actual brain works. ‘Inspired by,’ maybe, but not really like it at all.

“So our project was to actually make a neural network like an actual neural network. What we wanted to do was see if we could distribute information processing and storage across a network of physical nodes in a way similar to how the actual brain stores and processes data. And by a network of physical nodes, I mean a computer network. We wanted to see if we could make a computer network act like a single brain full of neurons. In our model, you’d have a bunch of networked computers, and each one would be a neuron. You’d identify a few of them to be input neurons—one for each feature, in our perceptron example. And each of them would be naturally connected to other computers in the network. So we would treat the transmission cost—basically, how long it takes for a message to get to that node from its input node would be the baseline weight for that input. And each node would take in a weighted sum of its inputs, and send that out to the various computers to which it was connected, further down the line in our neural architecture. Different inputs would cause a neuron to send to a different neighbor, creating a different discharge pattern.”

The story has been stolen; if detected on Amazon, report the violation.

“So you’re using individual computers on the network as neurons, to distribute the computational load. And you have nodes identified as your output neurons?”

“Well, yes to the first part, but not exactly to the second part.”

“Okay.”

“So, uh, your brain works this way, pretty much: You have a thought—let’s say you picture a boat in your mind. Synapses fire and discharge, from neuron to neuron, across your brain, branching out along a certain path. But where does the actual picture of the boat happen? It’s not like there’s a projector screen inside your head, and those neurons cause a projector to project a picture of a boat on your skull. And, there’s no specific neuron somewhere in your brain that represents ‘boat,’ either. It’s not like every time you think of a boat, or see a boat, there’s a neural discharge that leads to your ‘boat neuron.’ No, the neural discharge is the picture of the boat.” She emphasized the word ‘is’ with all the wonder and excitement that that concept aroused in her. “That particular pattern of electrical activity, that pathway from neuron to neuron, that is the concept of a boat in your brain. If you imagine ten thousand different things, each one will cause a different neural discharge pathway across your brain, but the mental images of boats will have very similar pathways, and if you imagine the same boat twice, it should be the same discharge pathway twice.

“At least, that’s the theory. So that’s how we designed our network. Rather than having the network lead to discrete output neurons, and then refining the weights to drive the discharge toward the correct neuron, we would mostly leave the weights as they are, feed the perceptron a large training data set, and then track the discharge pathways and analyze them for patterns. If our system has enough degrees of freedom, then we should begin to see unique, identifiable discharge pathways representing certain concepts within our data set.”

The two men with cyber backgrounds were grinning. The Special Agent was even laughing a bit.

“That’s awesome,” he said. “Did it work?”

Melody nodded, but it was a slow nod. A nod of, yes, but…

“Question,” said Sing, the analyst.

“Sure,” said Melody.

“If this, uh, this architecture has no output neurons, then how do you know—how do you observe what the output pathway was? The discharge pathway, as you called it?

Melody grinned. “You’re smart. That’s a smart question.” By the way the two cyber warriors were nodding, they agreed. “That was actually my part of the project. I was the lead on what we called the Consciousness Layer. See, you’re not aware of the electrical discharges in your mind, but somehow, when certain discharges occur, you ‘see’—in your ‘mind’s eye’—a picture of a boat. No one knows how that happens. Somehow, the process of the electrical discharge is the picture of the boat, and yet you perceive that picture in some higher way that’s hard for science to describe. Whether that’s the soul or what, we don’t know, but we couldn’t really program a soul, so what we did was make sure that the traffic, the packets passing through our neural network, could be tracked. Then my team wrote a traditional convolutional neural network to detect and classify patterns in those observed discharge pathways, like a brain scan. So we had a traditional neural network program analyzing our nontraditional neural network system, acting in place of a ‘consciousness,’ so to speak. Which is why we called it the Consciousness Layer.”

The two men gave her a slow clap.

“And you actually did all of this? You made it work?” asked the Clandestine guy.

“We did. One of the cool things about the idea is that it turns a regular computer network into a supercomputer, in a more efficient way than ever before. You probably remember back in the day when people would run screen-savers that, like, searched for extra-terrestrial life by analyzing space radio waves, or helped look for the cure for cancer by calculating protein folds. That was distributed computing, but you were giving each computer a problem, or a piece of the problem, to solve. How much that computer contributed was based on how powerful it was, and it had to be dedicated to that work— that processor-intensive application—for some period of time. One of our guys calls that the cottage-industry model. You give a bit of work to that computer, and come back to get it when it’s done. This is a whole different way of using a network for distributed computing. Any one computer only has to perform a single dot product, and only every now and then. Mostly, it’s the act of them talking to one another which results in the overall computation happening. The power of the system is only determined by the breadth and depth of the network, not by how powerful any of the nodes are as individual computers. In theory, we could even include the Internet of Things. Smart watches, smart television, smart refrigerators, could all be neurons in a giant neural network computer.

“Anyway, that’s the end goal, but for our test we got a volunteer network to run our processes and let us install our monitor on the routers.”

“A private network?”

“For the first runs, yes, but for the proof-of-concept it had to run on the Internet, so we botted a bunch of the university’s computers, and got a company that works with the university to let us bot some of theirs as well, and run our packet sniffers on their routers and hubs.”

“Okay.”

“And, it worked! Of course, with only a few hundred computers, a few hundred neurons, it could only solve rudimentary problems, because the run-time is longer per iteration with network latency. But the idea is not to create the world’s fastest neural net processor. The idea is to prove that we could harness the complexity of the modern Internet into, potentially, the world’s biggest neural net processor, without having to spend millions of dollars on a single supercomputer.”

She relaxed a little, her story told. She had never tried to explain the whole thing to anyone before. It was satisfying, and exhilarating, to share with others the passion which had consumed so much of her life lately.

“Well, that’s incredibly impressive,” said George Fry. “I’m glad to know geniuses like you are out there accomplishing things like this. I do have to ask, though, how this relates to our business here today.”

“Oh! Right. Okay, so, since the thing is supposed to run on the open Internet, and to be somewhat organic and evolving on its own, we wrote the Consciousness Layer to be completely generic. It would not only analyze our neural-net traffic, but be able to pick our traffic out of all the general clutter on the Net. That’s the first thing it ‘recognizes,’ is what sort of traffic is a neural net discharge, and then it gathers observations of those discharges and classifies them for analysis. So we put our neural net system out there, and set up our Consciousness Layer with its listeners, and turned it all on, and we, uh, found that there was already one running.”

“One what?”

“A neural network. On the Internet. Someone had already done it. Someone had made what we made, except they’d already fielded it—and pretty broadly, on the Internet. Our listeners began picking up our traffic, but also the traffic of this other net passing through the same routers.”

“So someone out there had the same idea as you.”

“Yeah. Except theirs is, like, way bigger. Way bigger. We think it’s the real thing.”

Glances were exchanged.

“So, when we started capturing this other data stream, this other AI operating across the Net— well, we weren’t sure what it was, but it sure looked like what we were doing. And we had never heard of anyone doing anything like that, and if it was out there, it meant bots on a lot of computers—meaning, a lot of computers around the country are running a program kind of like ours, a neuron bot, whether they know it or not. So we talked it over, and we decided to crowd-source the analysis of what we were picking up, and see if the hacker world agreed with us and could find the bots or figure out whose they were.”

“This is when you contacted your friend Brigham,” said Sing.

“Yes, sir. I knew him through a game I play online, and I knew he did some hacking—like, nothing major or criminal. Just kind of as a hobby. But he would know how we could put it up on forums and stuff and get the hacker world to comment on it, so he set us up a site on the dark web and publicized it and opened it for comment.”

“The dark web?”

“The hidden parts of the internet,” said the female Special Agent.

“Kind of,” said the male Special Agent. “Basically it’s just the term for those websites that you can’t access without an anonymizing browser or some form of anonymizing software. They’ll reject your connection if your connection allows you to be traced. It’s used for a lot of illegal activity, but also for a lot of people who just want to interact anonymously. It was actually invented by the intelligence community of a couple of countries, working with the private sector, as a secure public network for spies to use. It’s taken on a life of its own since then.”

“Kind of like the Internet as a whole.”

“Yeah, pretty much. It’s just another part of the deep Internet. You have to use the right protocols to access these sites, and you have to know where you’re going to get there, for the most part.”

“Okay, so you put up your findings on the dark web for hackers to comment.”

“Yes, sir,” said Melody. This was where the story took a less pleasant turn. “So we put it up, and let our system run, and we worked on our project, but we let the Consciousness Layer continue to pipe the second data stream—the data it was picking up from this other AI—to this dark web output site. And I guess people found it interesting and began to comment on it. We kind of put it up and left it for a while. We’d come back to it in a few weeks and see if anyone had said anything interesting. Anyway, this was when I went on vacation to my mom and dad’s house. We had that whole thing…” Her throat was suddenly dry. She tried to swallow, was not entirely successful, and drank a bit of her water while fending off visions of the apparition that had killed her would-be assassin in front of her eyes. “We had that whole thing there—that was Wednesday. When I got home on Saturday, I had a message from Cookie— that’s Billy Brigham—that gave me the address and username and password to a file dump site. And that’s where I found his note—”

“You’ll see this message printed out in your folder, page two of Miss Ritter’s file.”

She glanced at Sing, and he nodded for her to continue. “So, yeah,” she said. Her voice was shaking a little, now. “I had the message from him that someone had wiped our site on the dark web. But before that—sorry, so when I got home that Saturday, I logged into our Voxo channel—my Guardians team, I mean.” She rolled her eyes. “So, I play a game called Guardians with a team of people I know online. Cookie was one of them. So when I got home—we use Voxo for voice comms, and I logged into our Voxo channel, and Nailoo was on, and he told me he had heard from Cookie’s family that he had died of an insulin overdose. He had died, like, the previous Tuesday, I guess?” They were flipping through their documents, now, and she guessed they had all of the medical and police forms in front of them. “I was pretty shocked, but I didn’t think anything of it then. We played for a while, kind of a memorial to him. Then, as I was shutting down for the night, I found his message, and that got me to the file dump, and that got me to the read-me file. He said someone had wiped our server, and that he was going to take steps to cover our tracks. That was his last message before he died.”

Melody pursed her lips, looking at the surface of the table, thinking about how terrified she had been that night. And after all, not unjustified, as it turned out. “So we put together a package to send to you guys—” She looked up at the two agents. “—and then basically tried to lay low. I got scared when I thought someone might be following me, so I told Doran—my boyfriend—but he’s the only other one I told. And it really seemed like it was going to turn out to be nothing. After I told him, I didn’t see any more sign of anyone stalking me or spying on me or anything—at least nothing I could be sure about. And then, you know. That guy shot me.”

She fell silent, looking at the table again. For a long time, no one spoke. Melody could feel the bullet hitting her, like hot punch to the chest. She could feel with exquisite clarity the sensation of her breath failing her. Each time she had inhaled, she had gotten a little less, until she was capable of taking only the slightest, poorest excuse for a breath, and it felt like her chest was being crushed under a vice. She could feel with exquisite clarity, even now, over a week later, the experience of dying. With shaking hands, Melody reached out and took her water bottle again, sipping from it. She tried to get the cap back on but failed and dropped it.

Mr. Constantine—Rob, the red-headed guy—who was apparently in the Special Forces, so that was a thing—bent down and picked up the cap. He took the bottle from her and closed it, and then set it down on the table in front of her and took hold of her hand.

“You probably don’t get scared at all, by this sort of thing, anymore,” she said.

“What, getting shot? Hell no. I’m terrified of getting shot. Getting shot is the worst. Let’s talk after this, for a bit, okay? About that.”

“Okay.”

“That’s a good idea,” said the female Special Agent. “You have nothing to be ashamed about. You’ve been through a terrifying experience, and you’re going to experience some post-traumatic stress. But that’s normal. It doesn’t have to develop into PTSD. Talk to Mr. Constantine, and I can connect you with some counselors who handle this sort of thing, too, if you want.”

“Thank you.”

There were nods about the table.

“So, are you willing to continue a bit longer?”

“Sure. Yeah.” Melody nodded. “I’m fine.”

“Can you tell us about the thing that came to your rescue that night? And the thing that you saw at your parents’ house?”

She looked up, meeting Sing’s gaze, and took a deep breath before letting hers fall again. “Um… It’s going to sound pretty crazy.”

“Not to me,” said Constantine.

She flashed him a weak smile of gratitude. “So, I saw it, but I didn’t really see it. Either time. It was more like… It’s hard to describe. It was always dark, and this thing was—I guess it was black? All black, but, like, really black. Sometimes. I think maybe sometimes it was transparent. Like, when I first saw it, I was sitting on the porch, and my Mom came out and turned on the light—” A moment of pleasant memory. “—and then I looked out at the yard, and I could see it because the light was brighter and it was… it reached to the trees. And I could see this one tree, and this…” She hesitated and rubbed her face as she struggled to find the right words. “It was like a ghost. I could kind of tell the shape of it, even though it was like I was seeing through it, seeing the tree through it. But I could see one of its arms because that wasn’t in front of the tree, and it was just black, but I could sort of see the outline of it.”

“Why do you think you were able to see its outline? Some distortion at the edges?”

“Yeah? Maybe? We were looking right at it and still couldn’t be sure what we were seeing. Even when we shined a flashlight on it. In places the flashlight just seemed to make it disappear. But then it moved—”

“You saw it move?”

“Uh, no, not that first time. They took the flashlight off of it, just for a second, and when they put it back on the tree, the thing was gone. You could hardly see it when you were looking at it, but when it was gone, then you knew it had really been there. That was, like, the worst. They kept looking for it while I called the police. When I got back out to the porch, they were out in the yard, looking around. And they found it again, standing between a couple of trees further ba-ack…” Melody paused as a shudder went through her that caused her voice to break. She took a couple of long, slightly quavering breaths. “Then it was just, like, a black shadow, standing there. It looked like something out of a horror movie, or one of those old paintings of demons, except all black, like someone had just cut a hole out of… the universe. We just caught a glimpse of it. Again, by the time they got the light back on it, it was gone. It went back into the trees, so fast. We saw some branches moving, but that was it.”

“And you did not see it again during that trip? Or anything else unusual?”

“Yes. Correct. Nothing. That was it. The police came and searched, but didn’t find anything. I found out later that, like a couple days later, they found some blood and a bloody shirt or something, but that was it.”

“Police report, enclosure four in section five of your file.”

“So that was it. And then it showed up when I got shot.”

“Are you certain that it was the same, ah, entity? Meaning the one that intervened in your attack. Was that the same as the entity you witnessed at your parents’ house?”

In truth, Melody had never considered that it might not be. She was reminded of the deep shock of fear she had experienced while on the phone with the emergency dispatcher, when that lady had raised the specter of multiplicity. Now she felt a shade of that reaction again, imagining that there might be more than one “entity” of this sort wandering around.

“I don’t know. It seemed pretty much the same, but I never could really see it well, so I can’t say for sure that it was. It just… like, it scared me the same. You know? Instinctively. It felt the same. When I got shot, I fell down, and it was raining, and there were lights shining from the store, and I could see… I could see the rain kind of twinkling in the lights. It was bouncing off of the, um, entity, and I could see the light refracting in it, even though I could still barely see the thing itself. But it still definitely looked like a demon. And it… what it did to that man….”

“You say the rain bounced off of it?” interjected the female Special Agent.

“Sort of?”

“Kind of splattered off of it? Like it was striking a hard surface.”

Melody frowned, thinking. “I guess?”

“Think about how rain hits a person wearing clothes. Picture last time you were out in the rain— and not being attacked. Think about how rain lands on your clothes, or your friend’s clothes, or your umbrella. And then think about how it hits a car, the hard metal and glass.”

“Definitely more like that. More like it was a hard surface. I could see it kinda splashing like that.”

The woman nodded and made a note.

“So, we have your report here, and we’re not looking for you to relive the entire experience. Just a couple of questions for clarity.”

Melody nodded to Sing.

“Your attacker drew your attention by calling your name—” She nodded again. “—and when you turned, you say you saw the gun, a suppressed pistol. It fired, and you fell back against your car, and when you looked again, this thing had taken your attacker by his gun hand and was lifting him up.”

“Yes. Or—it’s not like I was out for a while. Everything was just happening really slowly. I fell back, and it seemed like it took a long time for me to hit the ground. Is that normal? I’ve heard of that happening.”

“Yep,” said several of them at once. “Completely normal, time distortion,” said Constantine.

“So yeah, I fell back, and I saw him standing over me, except by the time I hit the ground, his arm was way up in the air like…” She started to pantomime and felt her chest wound advise against it. “Well, anyway, straight up in the air.”

“Hold on,” said the male Special Agent. He pushed back from the table. “Rob, help me out here,” he said, gesturing for Rob to join him. They stood.

“I’m your assailant,” said the agent, coming around the table to stand near Rob on her left. “I’m pointing my gun at you—”

He made a finger gun and pointed it at her. Melody stared at his fingertip and felt her breath quicken and panic grip her.

“Easy, now,” said the female agent from the other end of the table. “We’re just playing pretend.” She nodded quickly.

“Was this the hand he was using?” asked the male agent. “His right hand?”

“Yes.”

“And the entity—Rob, you’re the entity. The entity was on that side?”

“Yes.”

“So the gun fired, and you went down, and then you looked up and saw him. And he was still holding the gun?”

“Yeah. He was still holding onto the gun, and he was on the tips of his toes.”

“And it had him by the hand? The wrist?”

“Yes.”

“Like this?” Rob had taken the agent’s right wrist with his left hand and was raising it up toward the ceiling, like a fight referee presenting the winner of a bout.

“Yes. Um, no. Actually. I guess it was the other hand.”

“My other hand?”

“No, I mean, Rob’s—Mr. Constantine’s.”

“Rob is fine,” said Constantine, switching to his right hand and turning to face the agent in the process.

“Yes! Like that.”

“So its got me like this, and I’m up on the tip of my toes…” Rob was just a little shorter than the agent, which made him much too short to play the role of the demonic apparition she had witnessed, but the illustration was sufficient.

She continued: “And then it just put its hand over his face—”

“It’s other hand,” said Rob. “Like this.” He had to turn his left hand over to put his palm on the agent’s face.

“Yes. Right over his face and just threw him right back through the window of the other car.” She watched as Rob tipped the agent backward and pushed him toward the wall behind. She nodded vigorously. “It was crazy. I would say it was like something out of a movie, except it wasn’t. It felt like a nightmare.”

“And then what happened?” asked the agent from under Rob’s hand.

“It, uh, shot him.”

“It shot him?”

“Yes, with his own gun. It kinda bent his arm down, and I couldn’t really see what happened, but I heard the gun shoot.”

“You couldn’t see what happened why?”

“It was in the way.”

Rob pushed brought the agent’s hand down, took his right hand from the agent’s face and wrapped it over the agent’s finger-gun, and then stepped between her and the agent as he use his two hands together to wrist-lock the agent and in the process point his finger gun at his own head.

“Yes,” said Melody, a little breathless. “Exactly like that. Just like that. I heard the shots, and then it took off.”

“Took off?” asked the female agent from the far end of the table.

“Yeah. It kinda looked—almost like it heard something—and then, shoop, it was gone.” She made a lateral cutting gesture with her hand. The male agent and Rob separated and returned to their seats as the female agent continued to question her.

“Did you hear anything?”

“You mean when it left?”

“You said it acted like it heard something. Did you hear any sound that might have drawn its attention?”

“Not really? I don’t know. I don’t remember.”

“That’s okay. So it might have heard something, and it took off. Did it run?”

“If it was running, it was crazy fast. It just kind of zipped away before I could blink.”

“Shoop,” said the female agent, imitating her hand gesture.

Melody smiled self-consciously. “Yeah.”

“No no, I get it. Like a cat, right? When they decide to go, they’re just gone.”

“Yeah, a lot like that.”

“Okay. What happened next?”

“Um, well, I was kind of dying, so the rest is pretty fuzzy. But I remember there was a rattling sound just after it took off.”

“A rattling sound?”

“Like… hail. Hitting the cars.” Several of them nodded.

“By then you started crawling back toward the store?”

“Yeah.”

“And do you remember anything more after that that might be important? Anything to do with the entity or your attacker?”

“No. All I remember about that was feeling my chest get worse and worse, and coughing and not being able to breathe.”

“That’s fine. No need to dwell on that. Miss Ritter, have you had a chance to look over the data that Mr. Brigham secured for you before he died?”

“Some of it.”

Sing nodded. “And the other members of your online gaming team have no knowledge of any of this? It’s important that we know how much information has… is already out in the wild.”

“No, they don’t know anything. I brought it up with Cookie in private chat, and we kept it secret. They don’t even know I had a…” Suddenly she shuddered, her heart in her throat. They didn’t even know that she had gotten him killed.

“Hey, Melody? Melody, this is not your fault, okay?” said the female agent. “You need to understand that. You had no idea, no way to know that anyone was prepared to kill over this. And no reason to suspect it. Mr. Brigham—”

“Cookie,” said the cyber warfare man.

“Cookie, then. You did not cause his death. If he was murdered—and that still isn’t certain, yet—but if he was murdered, the only people to blame are the people who murdered him, and tried to murder you, okay.”

Melody nodded, but she could not make herself feel that it was true. She was crying, now, and could not stop. Rob took her hand and squeezed it, and they waited while she cried.

Eventually she regained control of herself and wiped her eyes. “I’m sorry,” she said.

“There’s nothing for you to be sorry about,” said the male agent. “Your friend died. You’ve been through a traumatic experience. It’s only natural that you feel the things you’re feeling. And I you know that we will find these people. We will bring them to justice.”

“One way or another,” added Rob.

Melody nodded and squeezed his hand in return. “Thank you. That will be good,” she said. In her mind, she saw the image of her shooter arching backward and going through the car window, and for the first time she realized that it gave her some satisfaction. A part of her wanted to see that done to everyone who had had a hand in the attack on her or in Cookie’s murder. So that was what revenge felt like. She could see why people sought it.

“Now, one more question, Miss Ritter. Do you have any idea why the entity would come to your rescue?”

“What?” She stared at Sing. This was the second time someone had spoken of that demon as being her rescuer. The first time it had been completely unthinkable. Now, though, she found that she could not entirely discount it. “No…” she said, shaking her head. “I have no idea. There was… when we saw it at my mom and dad’s place, it was not friendly. I think it wanted to terrify us. I don’t know why… it saved me.” She shuddered. “Maybe there’s more than one,” she said, very quietly.

Sing nodded. He looked around the table. Several others were taking notes. He waited for them to finish. “Anyone else? Anything else for Miss Ritter?”

“No, I have nothing more.”

“Same.”

“Again, Miss Ritter, thank you very much for coming to see us today and giving us your account. I know it’s painful, but you’ve been an incredible help.”

“It’s fine,” she said. It was not really fine, but it was… manageable.

“Well, if you think of anything else, or need anything, you have Chief Constantine’s number—”

“And take my card,” said the female agent, extracting a business card from her ID case and passing it down the table. “Give me a call later if you want to talk to a trauma counselor. We have some of the best. They’ve dealt with a lot of agents who have been through similar experiences to yours.”

“Thank you so much,” said Melody, and she meant it. “Thank you all so much. I just hope you find these people. And… punish them.”

“You can count on that,” said Rob. “Come on, I’ll walk you out to the waiting room.” He stood and took the handles of her wheelchair, and Roy Sing moved to the door.

“Have a good day, Miss Ritter, and try not to worry,” said George Fry. “We will sort this out.”

“One way or another,” repeated Rob, as he pushed her out.