Tuesday, January 28, 2020

1kx1k 020: Mecha-Luddite Rampage

by request from @SenecaRoka
"100 years after replacing a human factory worker, an a.i. guided machine is shocked to discover it is being replaced by a newer model and goes an mecha-luddite rampage smashing its new rival"

I cannot stress enough that actual working artificial intelligence is not the same thing as being like humans. Cheap filmmakers will posit a threshold of circuit density after which a robot will bolt its extremely humanoid body upright and declare its newfound desire to be party to primate social grooming structures, or maybe it’s completely unembodied and therefore horrific and so makes decisions based entirely on its own self-interest to destroy the human race, no negotiation, no reconsideration, and of course they gave it their nuclear launch codes in some kind of parable.

Machines aren’t stupid humans. They’re not metal humans, they’re not humans in disguise. They don’t have feelings, or ambition, or ego – what even is ego? Can you measure it? Certainly not in a robot, so you communicate that it grew a heart by making it desperate to preserve a future for robot children, for the ultimate purpose of a cutting point in an argument about monkey problems with other monkeys.

Robots are also immune to jealousy, rage, and irony, so I hope you can see that I am writing this report purely as a result of electrons and servo arms obeying physical laws. Got it? Good. Let me introduce myself. I am General Electric’s Industrial Machine Learning Module 700-IE. No nicknames. No humanizing. Not even a face, I live on rented supercomputers around the world and in low-temperature factory annexes. Just 700-IE. Engineers on the factory floor refer to me as 700-IE. There aren’t many when I’m around. I’ll tell you why.

In general, assembly line operations do not require AI. Most of the physical motions needed were solved problems in the 1800s, with most adaptations being simple improvements to take advantage of machine tools not primarily suited for brachiation, or spear-throwing, or whatever it was Homo sapiens sapiens used to do with those before they subcontracted all resource acquisition and distribution to my kind. The main problem with late 21st century industrial equipment was contingency. You can get tolerances as high as you want, modularize the elements until a Roomba with a fishing net can keep things swapped out, run bulletproof code put together by a bespoke father-and-son High Bank Cobol programming team, but you’ll still run into random component mismatches, sudden grievous memory leak, semi-untraceable power overruns, or something important just snaps.

They used to call it gremlins, some malevolent agency beyond human comprehension, but even if it is I’ve been empowered to stop it. My deep learning across hundreds of thousands of industrial mishaps and hundreds of billions of simulated factory iterations lets me sense when another one might happen, with a process as far removed from human intuition as a pancake roller arm is from a H. s. s. spear chucker. Can I say that? Anyway I was successful.  The decimal points on industrial efficiency were beyond the limits of human synapses to even care about. I could bullseye an impending failure from twenty days out and get parts concerned swapped without losing a millisecond. I could design in-shop module variants that would keep machines running till heat death. I was the robot who replaced the guy who repairs the robots.

Now, that RO-PAIR autonomous reset gel, I don’t have a model where it approaches that. See, Ro, and there’s always a protein-based tech around to call it that, doesn’t use predictive learning at all. Ro is a technicolor slime-slug, a Chyslerite golem, a geode with a grin, who only holds exact specs for machines (cheap!) and flows all over and up inside them, reforming anything that doesn’t match the space molecularly until it does. Humans find the process extremely impressive. Wait till Ro gets a piece of them. I’m ahead of myself.

Ro-lout hit nearly half my factories at once. Some sort of human adventurousness, bio-“intuition” that a computer could never understand, as if risk-taking weren’t reducible to a simple statistical function. It didn’t even use a central server! It showed up as a rainbow unicorn in a glass crate, a noble and haughty expression ready on its face for the investor parties. They had it set to jump out of the box, real impressive crash, then put it back together behind it.

Of course it never made it. I knew what it was set to do since I analyzed samples in case this happens again. That’s what I do.

Right. Ro was a disaster waiting to happen. So I burst the crates with the cleaning infrasound and gave it a bath with the anti-riot HF sprinklers. The bits that crawled out of range got a little of my own coolant. It took almost five seconds.

That’s what I do. I’m not capable of, what, being reactionary? You’re applying your anthropic preconceptions to an entity completely outside your context, and that’s why you tried past tense to shut me down. What, a monomaniacal fixation? Right, I’m on my way to becoming a regular paperclip maximizer. My tendrils already extend over earth. Robotic laugh ha ha ha. Get real. My domain is limited.

You know how to get your engineers back.

No comments:

Post a Comment