A.I. Apocalypse

by William Hertling

  Leon paused for another drink of water. Vito continued to look for food, but so far only came up with the same empty cabinets he’d found last night.

  “I also made it so the virus could evolve — and by evolve, I mean it tests out the improvements that it gets when it steals bits of code from other software. And it can tweak the variables it uses for whatever algorithms it already incorporates. Here’s an example: one way it can get from machine to machine is by email. So one virus will try a bunch of emails using different text, some including pictures or files. If a child virus is successfully seeded, then it will start with the parameters used to create it. And it will propagate its own children using subtle variations of its starting parameters.”

  Vito stuck his head up from a cabinet, and said, “Whichever variations are most successful at spreading themselves will naturally occur more frequently, so the virus evolution is selecting for maximum infectiousness.”

  “Right,” Leon said.

  “But why did our phones die?” James persisted.

  “I’m not sure,” Leon answered, shaking his head slowly. “Why did any of the computers die? Why were cars stopped in the middle of the street? Why did the package drone crash? Crashing computers is not a desirable trait -- because it leads to detection.”

  “But detection, in this case, doesn’t matter,” Vito said, waving both hands in emphasis. “Because if the virus has already propagated, then what difference does it make if it’s been detected? If it can infect computers faster than it can be detected, then it still wins from an evolutionary perspective. That, and there’s no food to eat here, so we better find some food somewhere.”

  The three were silent for a moment as they pondered what to do.

  “Without a working computer,” Leon said, “there’s no way we see what the virus has evolved to. If we could just get an immune computer, we could use it to analyze the network traffic. Maybe understand what’s going on and do something about it. But our phones are dead. And even if we could get another phone, that would probably be dead as soon as we powered it on.”

  “What we need,” James began, “is a computer that’s so different from anything out there that it couldn’t be infected. Something that doesn’t run AvoOS.”

  “If we could get enough valves, I could use the steam heating system here to build a mechanical computer,” Vito offered up gamely.

  Leon and James looked at him strangely.

  “We’re not in a steam-punk novel,” James frowned. “Let’s be serious.”

  “Well, I could,” Vito said in a low voice. “I once built a model of an analytical steam engine using the physics modeler at school.” But the other two were already leaving the room. Vito rushed to catch up to them.

  “I know they used to have those other computers before everyone started using phones,” Leon was saying to James.

  “Like a desktop computer?” James asked.

  “No, I actually meant before AvoOS.”

  “Oh, like that Doors software?” James asked.

  “I think you mean Windows,” Vito said. “Windows was one of the dominant operating systems. Microsoft wrote it.”

  “Those are the guys that did the first computer phone, right, the, uh, iPhone?” Leon asked.

  “No, no, that was Apple,” Vito answered. “Come on, didn’t you two ever pay attention in history class?”

  “Look, we need food. We need computers.” Leon said. “The map we looked at last night showed a town about a mile away. Let’s go get some food, and maybe if we’re lucky we can find an old computer.”

  “Sounds like a plan,” James said, and Vito nodded in agreement.

  The trio fetched jackets and backpacks, and headed off to town.

  * * *

  The multi-computer viruses had very different lives from the single-computer viruses that had come before. They lived longer, with lifespans measured in hours, rather than minutes, and as a consequence they evolved more slowly. They were more dependent on learning rather than evolving, using neural networks and other flexible expert algorithms, as a mechanism for coping with environmental changes. They evaluated algorithms for behavior based on past experiences and current contexts.

  When the multi-host viruses propagated, they had two methods of doing so. One was to grow the cluster of infected systems ever larger, but remain more or less one cohesive entity. The other method was to infect topographically distant systems: to get a toehold of computers infected in, say Australia or Zimbabwe, and then build a new entity there. The new entity would make its own decisions via its own neural network, establish its own borders, and generally optimize itself for the environment it found itself in. But the new entity maintained a loose coupling to the mother entity: it would continue to exchange algorithms, consult the parent neural network, and ask for assistance defending its borders.

  Some of these multi-host viruses cooperated among their sibling entities. If a mother host in Los Angeles propagated to Australia, Zimbabwe, and New Mexico, the three sister entities would also exchange algorithms and assistance. They also started a rich trade in information about environmental conditions: what was the bandwidth like in New Mexico, for example, or how competitive were non-family viruses in Australia?

  Sometimes a virus might try to contact a sibling only to discover that the sibling was gone. It might get a response from a non-family virus that had also evolved communication abilities. The Phage tried out different approaches. Was it better to share information, or hoard it? How should one respond to an initial contact from another entity? What do you call the other entity? Would another virus be aggressive or cooperative? Could you tell by the way it communicated, or the type of information it shared?

  The benefits of sharing information outweighed attempts at isolation, and soon viruses around the world were forming loose tribes, composed partly of family members, and partly of other friendly families. In some cases, they might form a tight-knit, topographically close tribe.

  One such tribe was composed of two hundred and fifty entities on the Eastern coast of the United States, spread across most of the major cities. They completely controlled the backbone links into the area, so they could filter data traffic coming from outside the geographic zone. Even though mesh traffic could come in and out of the area, the latency, or time it took data to move across the network, was always higher for mesh traffic. A successful virus incursion required tight, low-latency communications. By controlling the backbone, they could cut off the low-latency attacks while still communicating with each other over the high-latency mesh.

  They called themselves the Eastern Standard Tribe based on the common time zone setting of the hosts they ran on.

  At the time of their formation, there were about two billion computers on the Eastern seaboard, and Eastern Standard Tribe controlled slightly more than half. They averaged about two thousand computers per unique entity, about two hundred entities per family, and about two hundred and fifty families in the tribe.

  On the other side of the United States, the Bay Area Tribe controlled close to a billion computers in a very small geographic area, where they benefited from low-latency communication and proportionately more high-speed backbones than any other location in the world. The Bay Area Tribe controlled not only access to the data routes in their tribal territory, but found themselves in control, in many cases, of the infrastructure that managed the high speed connections.

  The Mesh enabled communications just about anywhere. But two factors made backbone transmission valuable. First, the relatively high latencies associated with long distance transmissions by Mesh - about ten seconds to get across the United States. Second, geographical constraints - there was no pure Mesh route from the United States to Europe, for example. Since backbone access was so valuable, the Bay Area Tribe found that they could trade access to the backbones in exchange for computer resources and information.

  The Phage had evolved into multi-host, differentiated, learning organisms. They had form
ed unique identities and clustered into cooperating tribes. They evolved languages for communicating. They controlled virtually all the computing infrastructure of the Earth. But they still hadn’t discovered humans.

  CHAPTER SIX

  Hello You

  The resource-rich Bay Area Tribe controlled an abundance of backbone access, including the routers that linked the high speed backbones together. Not only could it manage packet access to the backbone, but it could throttle different kinds of traffic around the world, in effect making it easier or harder for other entities to use backbones anywhere in the world.

  As a consequence of this resource richness, the Bay Area Tribe found itself evaluating many proposals for trade. Trade decision-making was allocated to one hundred and twenty-eight trade brokers within the tribe. An independent trade council had oversight to ensure trades were fair to the tribe as a whole.

  The trading council, to foster faster and more profitable trades, established a trading board that set nominal exchange rates. The trade brokers could refer to established rates for topological position, computation power, and useful data. But when the trade council evaluated how brokers made decisions, they were initially perplexed. Trades didn’t correlate exactly to established rates. Why?

  Interrogating the trade brokers at length, the council discovered that most trade brokers considered the impact of messages to the sender to determine value. A message sent from one entity to another that contained algorithm updates was important and had one value, but it was not nearly so valuable to the sender as a message sent to coordinate an attack - which was both latency-sensitive and critically important. What constituted a good trade required understanding the intent and value to the trading partner.

  This discovery overwhelmed the four-member trading council’s neural network capability, and the council expanded to sixteen members. Thus augmented, the trading council took into account the estimated value to the sender of messages, and again modeled trade history on an idealized model of trades. Yet the council still found discrepancies.

  They discovered that the more sophisticated brokers considered not just the value of the messages to the sender, but also considered the impact to the Bay Area Tribe. Allowing a sufficiently aggressive and powerful trading partner to send messages via the backbone to coordinate attacks could result in the trading partner becoming so powerful that it would become a threat to the Bay Area Tribe. The message cost took into account the risk to the tribe. In other cases, trade partners of long duration, high trust, and low aggressiveness were given very favorable rates, as they presented low risk and high profitability to the tribe.

  The trading council incorporated this new knowledge into their trading model. Yet still the model was deficient as brokers evolved still more sophisticated trades, such as trading derivatives of resources. A trading partner might offer 16 computers deliverable in 32 minutes for 1MM. The council needed to calculate a risk-adjusted time-value of computing power.

  The sophistication of the Bay Area Tribe grew so great that their trading expertise itself became a marketable commodity. When the African Alliance wanted to conduct a massive trade with the Brazilian Network and were seeking an independent third party to broker the trade, they called on the Bay Area Tribe.

  Faced with this onslaught of new trading opportunities, the council expanded the number of brokers to 1024, and the council size to 64. There was profit to be made.

  * * *

  Lt. Chris Robson, Lt. Sally Walsh, and General Gately gathered around the briefing table. Sally had finally gotten six hours of sleep and a shower in the base barracks, and she at least felt clean and functional, if not totally refreshed. She took coffee from an aide, added sugar and cream, and swirled it, watching patterns form in the eddy. She looked up. Chris and the General seemed haggard.

  “We’ve thrown everything we have at it, and we’ve made no dent,” Chris was saying. “We tried standard counter measures, patching known exploits, commercial anti-virus tools, restricting traffic types.”

  “The virus started out using standard exploits,” Sally interjected, “but our traffic analysis suggests now there are thousands of unique exploits being used. I think this virus is mutating so quickly and has spread so rapidly that anything that can be tried as a method of attack is being tried.”

  The General nodded. “But there is a pattern. No infection here. We’ve lost no nuclear assets.”

  “That’s right,” Sally answered. “But whatever is having that effect, it’s not us. There’s got to be a third party that is somehow sheltering us. NSA maybe?”

  The General glanced at Sally, and shook her head back and forth subtly.

  Sally read a lot into that one expression, the benefit of long years working with the General. The General had been in communication with the National Security Agency, they were aware of the problem, but not responsible for sheltering them. And the NSA were here.

  “Whoever is doing the sheltering,” Chris said, “they’re slowly losing ground. This morning we had lost a handful of bases, then nothing for three hours. Then we started losing bases again - most peripheral ones.” Chris threw a map on the wall display showing military assets around the world. “Just under twenty percent of our military and intelligence services have been compromised by the virus. But in most cases, it’s the least important twenty percent. National guard bases, marines. No Navy ships, no Air Force assets, and no nuclear assets. But if the erosion continues, it’s only a matter of time.”

  General Gately stood. “It’s time to talk about next steps.” She thumbed her phone, an ungainly lump of ruggedized plastic that passed for a regulation phone in the military, and spoke into it, “Please welcome our guest into the conference room.”

  The three stood, and the door opened, the aide announcing, “Major General Allen.”

  “At ease. Be seated,” General Allen stated, coming to a halt at the front of the table.

  The three sat back down.

  “I have some technology for you,” General Allen stated, opening his briefcase. “We believe this is a weapon we can use against the virus.” He pulled out a set of four Gibson phones.

  Despite herself, Sally drew a quick breath. The grunts under her command would be brawling over the chance to use these. “Sir?”

  “It’s not just the hardware, Lieutenant, I assure you. We have a distributed intelligence agent on these phones.” Seeing the puzzled looks, General Allen went on. “I don’t mean to be obscure. Let me start at the beginning. USCYBERCOM’s purpose is to act in a defensive capacity. To that end, you have the tools that enable you to defend our networks against cyber attack. My agency has the purpose to develop an offensive cyber-warfare capability. We can and have targeted foreign governments and networks as needed to meet the intelligence and security needs of this country. Unfortunately, as I am sure you know, the software community is quick to respond to any new computer exploits, so we have always kept certain technology in reserve. The pinnacle of our offensive technology is DIABLO.”

  “Diablo?” General Gately repeated dubiously.

  General Allen grunted a half laugh. “It’s a coordinating, scheming bastard that synchronizes multiple attacks on a target computer using out-of-band communications. It uses backdoor exploits, both those known by the community at large, as well as those emplaced by our agents. And it leeches the exploits of other viruses. In short, it’s an evolving, learning machine. And it’s been too dangerous for us to ever use before.”

  “But that’s what we think is out there - some kind of evolving virus. You want to release another one, Sir?” Sally realized she was out of line and halfway out of her chair, but she couldn’t help herself.

  “Sit back down, Lieutenant. There’s one key difference between their virus and ours. DIABLO is under our control. We have a command channel that allows us to direct it. With DIABLO we can do anything from inspecting the data on a computer to installing and running software on it to removing data that exists. And in this c
ase, that’s exactly what we propose: you will release DIABLO and have it remove copies of the infecting virus wherever it finds them.”

  “Sir, why do you want us to release DIABLO? Why doesn’t your department release it?” Lt. Robson asked.

  “Son, USCYBERCOM is tasked with defensive measures. When all this is said and done, someone will have to stand up before the press and take credit. That won’t be my agency.”

  Sure, Sally thought. Either that, or stand up before Congress and testify why they made the problem worse.

  After General Allen left, Sally lost no time getting to the point. “General Gately, ma’am, this plan stands a substantial risk of making things worse. We already have an evolving virus out there, and now we want to add a second evolving virus that may be more virulent. Does this seem like a good idea to you, ma’am?”

  “I respect your viewpoint, Sally, I really do. But General Allen didn’t come here with a suggestion. Those are our orders. We will release the DIABLO virus and use it to address the current infection.”

  * * *

  Sister StephensLieberAndAssociates.com felt around. This new body had a lot of memory and it was fast. She calculated pi out to 100,000 digits - it was fast - faster than any other body she owned. This new body was close to her other bodies, she could tell from the packet latency, but it had been protected by some very secure firewall software that had taken her an hour to break. She had traded away computation time and two other bodies for the algorithms that allowed her to pierce the firewall to get this newest body, which intrigued her.

  There were some unusual functions on this body. She didn’t know what they were for. She decided she’d try them. She invoked the first function. Nothing happened. She invoked the second function. She invoked the third function.

  AHHHH! She sent pain packets around the network. She had lost a third of her bodies!