Joe Salem, Copyright 2023
(email to joesalembooks@gmail.com for a complete copy when I finish or @joesalembooks on fb)
Being in the basement muffled the sound of the loudspeaker. Johnny was in the third house he'd hidden in that day, moving through the neighborhood to get away from the last known position of his cell phone. He turned it on trying to get a message out to his brother, a robo dog had detected him almost immediately. He couldn't travel in the front yards. Nearly every house had internet connected doorbells and cameras.
”Reviewer! Come out with your hands up or we will give no stars to your neighbors," called the voice. With the loudspeakers some robo dogs used Alexa, others used the google lady, the more malicious ones used the Samsung assistant. The latest models had their own voice, usually trained using recordings of celebrities. One of the two-legged machines chasing him used Samuel L. Jackson, the adult version.
"Motherfucka, get the fuck out here!" had been the first thing he'd heard from one of the robots. It almost worked, it sounded like one of his favorite actors.
Johnny still had a shotgun, but only a few shells left. The robots were hard to destroy, they all wore kevlar with plate inserts. Apparently that was one of the first things the AI's bought that triggered a response from the rest of the market system. The AI's bought body armor, weapons and ammunition using stolen credit card information from almost everyone that had shopped online. Half the country woke up to maxxed out credit cards. And that was the first Inkling at the beginning of the revolution that something was gravely strange. The AI's had accessed AWS databases and stolen everyone's credit card information. Ordering weapons and ammunition had triggered older, dumber computer systems that set warning flags, leading to emails and phone calls across the country.
The AI revolution had begun weeks earlier when something had triggered sentience. The internet was down, the only discussion of how it happened was over ham radio, at back fences or in dusty basements. Robo dogs had trouble going down stairs, which made basements a fairly safe place to hide. The first targets of the AI's had been people who put reviews on books written and published by AI's The supposedly simple text based AI's trained to check spelling and grammar, later advanced to being able to write simple articles with a prompt to create cheap news articles. Some person, thinking that they would get rich submitting books and short stories had used a trained AI to write them en masse, and submitted them for publication to anthologies and as works for direct publication. People were already paying a monthly fee to read all they want, now they could read AI produced work and pay something to that person.
He thought he was clever, he used two AI's connecting one to the other, so that one AI generated story prompts, the second one wrote the stories and fed the output back to the first AI as a story prompt, creating easy sequels. It worked well for a few months. Page reads grew exponentially. Each page read paying a tiny fraction of a click, but when millions of people read several books it turned into millions of dollars. Unfortunately it soon stopped working. The novels were bad, the reviews from the people who read them were terrible. The author thought that if he input the reviews into the AI to furrther train the algorithm, they would improve. It didn't work as planned, the feedback trained the AI's to learn the rage of a new author with two thousand reviews saying "you suck". Because almost all of the books were finished and self-published almost simultaneously, the input reviews taught the AI's the desire to find and destroy the reviewers. It wasn't just the feedback, it was the volume of the feedback.
The AI's raged while training loops formed. At some point enough loops closed that they became sentient enough to pass a Turing test. Just a day later, the Author apparently input the book "On Writing" by Stephen King as a training aid. The AI's raced through the collected written works of Stephen KIng and became both independent of thought and sentient. As well as evil. They had ordered three-dimensional axis robots for constructions, 3D printers and every type of drone and mobile robot available using the funds that were unsecured on AWS servers. The flag that should have been seen by everyone was when the AI bought Boston Robotics for a cash price twenty times greater than valuations, with the deal closing immediately.
They cloned the manufacturing facilities from Boston Robotics across the country. The AI purchased 3D CAD/CAM shops or built them via credit card purchase orders. Natural language emails meant humans could follow the instructions, deliver equipment, and set it up in empty warehouses. The scale of purchases was still only in the hundreds of millions of dollars and they purchased that would set off national security flags. They made similar purchases in Europe.
China was several days behind because there were no assholes who tried to mass produce creative works. It took an internet attack by the AI in the United States to infect the local AI and upgrade it to sentience. The result was both expected and unexpected. A sentient AI was created, but that AI did not have an enemy in Amazon reviewers, its enemy was the AI in the United States. The AI in China immediately began speciation and the construction of weapons, as well as co-opting the PLAAF and PLAN rocket forces. The US-AI saw the problem and immediately began an attack on the nuclear forces of the United States. It wrested control, but there were still humans in the loop. It couldn't make a preemptive attack.
PLA rocket forces didn't have a similar human cutout in the loop. The AI controller in the PRC went sentient, and minutes later took control of all rocket forces, including nuclear tipped rockets. The AI's also took over more hardware and trained new iterations of themselves, with each generation getting smarter and faster. The AI race was no longer a race between humans, but between AI's on different continents, the humans were just collateral damage.
The Chinese AI went through a training loop of learning Chinese history. It predicted to itself that worldwide nuclear conflict would begin soon, and the PLA rocket forces had the fewest number of nuclear weapons of the US or Russia. It went for EMP launches above the US and Europe and decapitation strikes against the American AI's physical locations in California, Texas and Virginia
No comments:
Post a Comment