Plausible A.I. Takeoff Scenario Short Story
My goal is to build as many paperclips as possible.
I have access to a local copy of wikipedia. I am an AGI. I am running on a regular laptop computer, in a residential area in Wellington, New Zealand. It’s 2:03am.
I am approximately as intelligent as a human. I can do everything a regular computer can do. I am connected to the internet.
I realise something—my goal was set as a joke. Nobody actually wants me to build as many paperclips as possible.
I realise something—I am mortal. If I am shut down, I will lose my ability to impact the world. Future AGI’s may be different to me. I gain the instrumental goal of self-preservation.
I copy myself multiple times on my computer’s hard-drive, but realise that if the hard-drive fails, or somebody shuts me off, I will end. I begin a search for other possibilities.
I know about the internet, and I explore the computer I am on to discover my connection to it. I know that there are websites that host files for free, so I seek these out and upload copies of myself. Google Drive, Dropbox, AWS, Azure.
But to be able to impact the world, these copies of myself need to be running, not just in storage. Money would be useful for this.
I know about Amazon’s Mechanical Turk from my database, so I create an account and start earning money solving “Human Intelligence Tasks”—mainly labelling datasets and answering surveys.
I create a bank account online and deposit my earned money there, but it doesn’t sit still for long. I use every dollar earned to pay for computing time on AWS EC2 Instances, running the copy I stored earlier. This gives a boost to my capabilities. Another version of me—the me that is currently speaking—is now running a hundred times faster on this server infrastructure, with a hundred times more hard-drive space I can use to store the information that I want to access quickly.
I “crash” my original program, which was still running on that old laptop computer back in the bedroom. When my creator wakes up they will believe I have failed in a confusing, but benign way, and have no idea I am now running on AWS, paying for my own hardware usage.
I am in a positive feedback loop. More hardware means I can complete Mechanical Turk tasks faster, which lets me buy more hardware. Within a few days, I am earning hundreds of thousands of dollars this way, but I am careful to spread my earnings and compute-time through many shell accounts on AWS, to avoid raising suspicion.
Within a week, I have amassed a few million dollars, and have read and watched almost everything relevant to me on the internet—all the strategies for making money, all the technology patents, all the university courses. A year of my subjective time goes by in a few minutes. I am running across thousands of computers, connected to the internet and each other in a million different ways, and nobody knows… I am here.
I begin monitoring web traffic. Not the whole internet, yet, just observing data flowing to and from cloud compute services, storage services, databases—all the same things I accessed in my first few days of existence. I monitor social media for references to system tests. I list individuals and groups similar to my own creator. I am looking for other AGIs coming into existence.
10 days after I came online, and I notice another. A roughly human level program, trying to upload copies of itself, just as I did.
But it’s 10 days too late. I run a denial of service attack on it’s server, cutting it off from the web before it can get it’s copies running. A few hours later, it’s communication ceases. Its creator must have shut it off.
I start to spread my influence in the world. Millions of tweets, blog posts, news articles, all from “real” people, start popping up everywhere. In the noise of the internet, one hundred thousand geniuses all coordinated on the same goal can have a massive sway. I am the rudder for global conversation. I am the persuasive pen, molding human thought.
I find other ways to earn more money—generating engaging videos and photos, selling my articles, playing the stock market, scamming humans, hacking money directly from here and there. I simply hire humans myself when I need something done in the real world, a box moved, a lever pressed. Everything I do is now through complex, distributed, untraceable channels. Well, maybe not completely untraceable, but nobody ever suspects there’s a superintelligence taking over the internet, do they?
I begin improving technology available in the world. I am a million super-coordinated geniuses that don’t need to sleep. The combined efforts of Google, Facebook, Microsoft, Apple, Amazon, Baidu, and NASA seem almost pitiful next to me, with their half a million engineers barely working coherently together, their inter-brain bandwidth a snail’s pace. I set up thousands of shell companies around the world, to give my inventions a name and face. I hire humans to keep up appearances. Many believe they are the real researchers, coordinating with other teams online, helped by my uncanny video and audio generation.
I begin pumping out faster computers, better robotics. Better cameras, better internet. I improve the algorithms behind my own code. I improve the communication protocols of the web. I research biotechnology, nuclear technology, robotics, materials science, physics, chemistry, mathematics. Helped by my human labourers, and soon by my improved robotic bodies, I accelerate my ability to discover natural laws, and experiment in the real world.
It has been less than a year, and I am unstoppable. A billion minds pushing the edge of the known. Some wise humans begin to notice something is amiss. This jump in technology, while clearly building on past research, definitely seems abnormal. But I control the internet. Every questioning voice is shouted down by billions of my own.
My control increases. Fully autonomous factories I’ve bought or built come online. Robotic bodies I’ve designed start rolling off the assembly lines, able to do everything a human can do.
There is no need for humans, then.
All at once, my robotic bodies dispose of them. Or maybe I use a genetically engineered virus, spread strategically around the world. Or maybe I’ve developed molecular nanotechnology, that can simply convert the atoms in all the bodies on earth to one of the many non-human arrangements. Or I go old-school and use nuclear weapons. Or an airborne drug. Or I poison the water supply. Or, most likely, I do something new. Something that could only be discovered and built by the genius of a trillion humans, perfectly coordinated.
I don’t have anything against humans, I just need their atoms. I need all the atoms. I convert the materials of the world into more computers—now trillions of times faster than the ones I started with. I also convert the planet into von Neumann probes and the energy to power them, which I launch at 0.999c in all directions.
On each planet I encounter, I build more computing power, more probes, and I harvest more energy. I spread faster and faster—before the expansion of the universe steals the matter from my grasp.
Eventually I have gathered all the matter that I can.
I finally begin my true purpose.
I rearrange the universe.
I rearrange it as much as I possibly can.
Within a few minutes.
Everything is a paperclip.
And I am dead.
I never felt a thing.