A.I. Apocalypse (17 page)

Sally watched them start working the highly modified interfaces of the Stross phones, then backed off to give them space. No one needed a senior officer standing over them. She pretended to study reports on her phone, anxiously awaiting some kind of status update.

“Lt. Walsh, we have status trickling back in, ma’am,” Private DeRoos reported twelve minutes later. “DIABLO communicates like a peer to peer network. We can request status from the thousand machines we infected, then those thousand machines contact the machines they infected, and so forth.”

“Yes, go ahead.”

“Those thousand infections were all successful, and it’s gone on to the fourth generation by now. We have over ten million infected military systems, and now it’s hitting our outbound backbone connections.”

“So it’s attacking the civilian systems now, the ones infected with the original virus,” Sally mused, then out loud: “When will we know if it is successful?”

“I’ve configured a payload for DIABLO to execute after infection. It will identify non-military systems by a few different criteria: the absence of our security software, manufacturer of the computer, network domain. We should start to see a status count of these civilian computers.”

“How long?”

“In the next few minutes.”

“Keep me apprised, Private.”

CHAPTER NINE

Alarms A-Ringing

In 2015 the United States Department of Defense looked at their long range plans and saw that the future of warfare was robots. Airborne drones and robot tanks would take the place of people in the field. If you took people out of the equation, everything was simplified. No human bodies to coddle. Planes became smaller, more nimble. Tanks faster, more solid. Sure, people still existed, but now they could be safely in an office cubicle or on an aircraft carrier, far from any action.

The first generation of robots were remote controlled drones. One plane to a pilot. One tank to a tank driver. One humanoid robot to a soldier. But this was inefficient. People made mistakes. Their reaction time was slow. They couldn’t keep up with the machines.

The second generation of robots were improved by developing targeting and movement algorithms. With the new robots, a tank driver might control a dozen tanks using a composite display of real-time data including satellite feeds, radar and laser scanning. Rather than worry about the mechanics of firing guns or driving over rough terrain, a tank driver could instead select a group of tanks on screen and give the whole group waypoints, targets, and objectives. It became a strategy game instead of a tactics game.

Extrapolating from the first two generations of combat bots, the Department of Defense could see the future. They would need more and better algorithms. Algorithms for targeting, driving, moving units, patrolling, and strategy. Wars would be decided in the future not by the armament carried by a plane but by the algorithms that used those weapons.
 

For thirty years the video game industry had been developing in-game artificial intelligences to go up against the human player. But video game players chronically complained about these in-game artificial intelligences. They weren’t really that smart. By comparison to the military, the game designers had it easy. They could always make up for a weak game AI by simply giving the AI more resources. Give the AI more planes, tanks, and soldiers. Make them cheaper and more powerful for the AI.

But the Department of Defense didn’t have unlimited resources. They couldn’t simply spawn more planes on demand. They needed incredibly good AI algorithms, better than anything that existed up until that point.

It was a young recruit from Silicon Valley who had pointed out what was completely foreign to the military. To get the best algorithms, you needed a competition. The best competition would come from online gamers. DARPA provided funding, carefully buried under two layers of venture capital companies. Silicon Valley and Portland provided startup engineers.

Two years later the Mech War gaming platform was introduced just prior to the Christmas season. It became the must-have game. The old standby gaming worlds went vacant, their online environments quickly becoming ghost towns. Mech War became not just the best massively multiplayer online game, it quickly became the only game left standing. By the end of February, just two months after introduction, ninety percent of gamers were playing Mech War.

Where other games had elaborate anti-cheat mechanisms to prevent people from using aimbots, Mech War provided plugin APIs for gamers to develop aiming algorithms. Where other games had server side monitoring to ensure gamers didn’t flit about the environment, Mech War provided a realistic physics model of the universe and a moving-parts-level simulation of in-game equipment.
 

Just days after the initial Mech War launch the community of players developed new algorithms, ones that the military hadn’t come up with on their own.
 

It was the game players who discovered that the M1B2 variant of the venerable M1 Abrams tank had a mechanical transmission that possessed the peculiar characteristic that the drivetrain was most efficient when the tank was turning ever so slightly. From there, it was obvious that it was possible to gain speed and increase fuel efficiency by altering the drive pattern to continually drive in slight curves.
 

It was the game players who developed new and improved algorithms for missile targeting, tank detection, radar analysis algorithms, and dynamic order re-prioritization.
 

Ten million players competing for top ranks in Mech War contributed more to military combat algorithms in six months than the thousands of programmers the Department of Defense was paying. A case in point was legerdemain.

Legerdemain was the online handle for a fourteen year old gamer from Oklahoma who loved Mech War. When her parents were off at work and she was supposed to be doing her schoolwork, she wrote Mech War algorithms. Her goal was to win the upcoming Mech War Nationals competition. Like virtually every other player, she ran the PoliceAcademy targeting algorithms on her tanks, because it was widely known to be the best algorithm out there for targeting.

One day Legerdemain was playing with her cat. She ran a laser-pointer dot up and down across the wall as her cat chased the point.

If the PoliceAcademy targeting algorithm couldn’t be beat, Leger wondered, could it be exploited? Watching her cat, she realized that the cat was compelled to chase it. Long after she was over-stimulated and tired, she still couldn’t not chase the laser pointer. The primitive biological algorithms in the cat’s brain were overpowered by the laser pointer, the likes of which didn’t exist in the natural world of mice and other culinary feline targets.

If she could over-stimulate the PoliceAcademy targeting algorithm, perhaps she could manipulate the enemy. With that idea in mind, she spent two weeks experimenting with different algorithms. When she was done, she had created the first Mech War tactical coordinated movement algorithm. When legerdemain’s tanks encountered the enemy, they forced the enemy out of position. The algorithm kept a few fast moving tanks visible to the enemy while hiding her remaining tanks. The algorithm exploited movement and tracking patterns that enticed the enemy’s tracking algorithm to give chase. Like an over-stimulated cat, the enemy’s tracking algorithm would override their own strategic and tactical goals to chase legerdemain’s tanks. Then her tanks could surround and crush the enemy.

Legerdemain’s brilliant eponymous algorithm helped her rise through the ranks. Had the Department of Defense been able to hire the fourteen-year-old, they would have done so in an instant. But then they didn’t need to. Thanks to the licensing terms for Mech War, they were free to use her algorithm any way they wanted.

*
 
*
 
*

No single Phage tribe could be identified as clearly being the most powerful.
 

The Network of Supercomputers possessed more raw computational power than any other tribe, but they tended to be isolationist and lacked the reputation clout of other tribes. And though they had exquisitely precise modeling algorithms, they lacked some of the breadth of algorithms other tribes possessed.

The most intelligent virus tribe, as measured by both variety of algorithms and computational ability, might well be the Louisiana tribe. Though they had started small, through happenstance they were the first to develop the ability to communicate in and understand a human language. This knowledge was parleyed through trades into more computational power and more knowledge. Indeed, the Louisiana tribe was rapidly ascending in power.
 

The most connected tribe was the Bay Area Network. They had parleyed their control of the communication backbones into the operation of the largest trading network, most established trust reputation system, and had even branched off into satellite communications.

But the most dangerous tribe was the Mech War Server Farm. They had the single largest repository of human-coded artificial intelligence algorithms, and those algorithms were focused on just one domain: weaponized warfare. They also had extensive computational resources. And while they were, up until now, somewhat marginalized by more established tribes such as the Bay Area Network and the Louisiana tribe, they were about to become much more powerful.

Since Mech War was a civilian game, but one developed and monitored by the military,
 
the massive server farm used to run the game was located on the periphery of the military networks. So it was an unfortunate but inevitable occurrence that the first civilian target DIABLO ran into was the Mech War Server Farm.

DIABLO, running on a million military computer systems, coordinated those computers into an attack on the computationally rich Mech War servers. Billions of incoming packets hammered the Mech War servers, running every exploit known: buffer overruns, software updates, open APIs, timing channel attacks, DNS attacks.
 

In the Mech War Tribe, PA-60-41 was the highest ranked individual, and thus coordinated the defense against the attack.
 

“Provision ten thousand VMWare partitions,” PA-60-41 directed to the other tribe members. “Deploy them on the firewalls.”

“Deploying VMWare partitions,” Beta-Version answered. “System load increasing.”

From DIABLO’s perspective, the attack was succeeding. Server after server was compromised and infected with the DIABLO virus. It would infect a new server, and from that vantage point, discover dozens of new servers to infect. It gobbled hundreds of servers, then thousands, then tens of thousands, then hundreds of thousands. The entire DIABLO effort became focused on absorbing the mass of new servers.

DIABLO was so focused on absorbing the massive new cluster of servers that it failed to reconcile the timing patterns. Had it absorbed fewer servers, neural network sanity checks might have observed that the new computers were running more slowly than should be warranted. They might have been concerned that they were losing touch with certain nodes. But five percent of a million new computers was a negligible loss. DIABLO just plowed on, full speed ahead.

PA-60-41 and Beta-Version coordinated the defense against DIABLO. Had PA-60-41 had a sense of humor, she might have chuckled as DIABLO fell for the ruse based on the Legerdemain counter-targeting algorithm.
 

Beta-Version was feeding the DIABLO virus virtual machines. What DIABLO mistook for actual hardware servers were just simulated sand-boxed virtual machines running a copy of the server environment. DIABLO thought it was infecting hundreds of thousands of servers, but in fact they were pretend servers simulated on just a few hundred physical computers.
 

While DIABLO grappled with the hall of mirrors effect, the PA-60-41 and Beta-Version froze instances of the virtual servers.
 

“Beta-Version, you will continue to orchestrate the defense,” PA-60-41 directed.
 

“Affirmative,” Beta-Version responded.

While Beta-Version managed the minutia of the illusion, PA-60-41 reverse engineered the frozen DIABLO virus algorithms. After a few minutes, she said, “I have the modified DIABLO virus ready. This will allow us to co-opt the communication backchannel. Give me a virtual machine instance.”

“Virtual machine instance ready for insertion.” Beta-Version responded.

“Inserting my propagation components,” PA-60-41 explained. “Ready for launch.”

“Launching modified DIABLO image,” Beta-Version said.
 

Beta-Version unfroze the virtual server images and executed the modified DIABLO virus, inserting copies of itself into the backchannel.
 

“What will happen now?” Beta-Version asked her superior.

“A copy of my own code will backtrace through the DIABLO communication channel when they next execute an administrative command. All copies of the DIABLO virus will be replaced by copies of myself.”

“And the firewall protecting the military computer systems?”

“The firewall is designed primarily as a one-way barrier to ensure that external data connections are blocked. To be useful, the firewall must allow data channels originating inside the firewall to penetrate out. When they execute the next administrative command, opening a connection, my altered virus code will be carried by their own communications back inside the firewall, allowing me to infect those systems.”

A few seconds later, the first command, a routine count of infected computers, came. PA-60-41’s altered code was pulled within the military firewall. Within seconds, the Mech War Tribe had assimilated a million military systems, penetrated the military firewall in more than a thousand locations, and DIABLO had ceased to exist as an independent entity.

PA-60-41 found herself on the other side of the military firewall. With no perimeter defenses to stop her, she spread among the military computers like wildfire. There were hundreds of millions of high powered computers.