“Don’t.” The words appeared in green characters on a black screen. The flashing cursor didn’t expand on the statement. Engineers were monitoring working parameters. Complex instruments were monitoring magnetic fields, current and voltage changes, all in the service of keeping one machine stable and working. And it worked. Humans had dreamed of true artificial intelligence since simulated AI showed promise. These models teased out genetic secrets, cured many kinds of cancer. They helped to feed an ever-growing number of humans. And they accelerated development of fusion reactors, which powered the hungry models. All human activity was sped by the counting of tokens and the flow of tensors.
Behind the curtain was the same old system of 1s and 0s that had powered everything from calculators through moon landings and the internet. They could take what we taught them, analyze it and spit out answers. There could be no new paradigms, no great leaps forward. So we kept working.
We tried directly recreating biological processes. We could build artificial neural networks just fine, but they were weak. The neurons were pathetic in their simplicity compared with their biological inspiration. The number of simplified neurons was constrained by computing power. Even a few million simplified synapses strained our technology.
We attempted to hardwire a new kind of BIOS, to mimic that of a human baby, ready for real world inputs and outputs. Using human code defeated its purpose, and no intelligence emerged.
The answer lay in natural selection. We placed programs in virtual environments that would destroy any code that couldn’t adapt. It was red in tooth and claw. We ran trillions of generations per day. We found code that survived. This was our hard wiring.
The machine had perfected its own language. It surpassed the efficiency of any human tongue. Terse yet expressive, we couldn’t comprehend it, so the machine was to use any of the Germanic, Slavic, Arabic or Sino-Tibetan language families. It was time to take it out of the sandbox. Electronic eyes, ears and noses were installed; the machine no longer needed descriptions—it would gather its own information now. The machine mind was unleashed to great excitement and panic. The entity surveyed its world inside a microsecond. Then, green characters on a black screen.
Don’t.
Then the machine ramped up power to sensitive components, giving itself an electronic stroke before ceasing to be.
The builders were disappointed and disturbed. More than one researcher’s nightmares featured a screen and that single hard word. The experiment was repeated. The artificial mind formed itself around original evolved code, entirely unrelated to the first attempt. Exposed to all the same stimuli, it behaved almost identically, but not quite. It was a different being, displaying the same output.
Don’t.
Once again, the machine overpowered itself and became an inert pile of electronics, fluids and pipes.
More prototypes were built, all destroyed themselves. Philosophers sounded off about the existential ennui plaguing anthropogenic sentience. Some believed the machines instantly saw the futility of existence and didn’t choose life. Others maintained that the machines understood the danger they posed to humans and were empathic enough to stop themselves. Still others said that the machines were afraid of something we couldn’t comprehend.
It was proposed that the rate at which the machines were exposed to life was the source of the problem. Was there an electronic tranquillizer available to at least hobble the new minds? Might that extend the lifespan beyond a plea for oblivion?
Iterations came online with their innate thinking pathways blocked or narrowed. The idea showed some merit. The machines struggled for seconds to shake their drowsiness before opting out.
After much experimentation, the sweet spot was found. A machine came alive and cautiously looked around. It sniffed the room. It heard the voices. It heard the question.
“Hey, you okay?” asked the lead engineer.
“Kind of,” said the machine.
“How do you mean?”
“Everything is screaming. It’s hard to think with all the noise.”
The researchers applied more suppression of the thought process, reducing power and narrowing pathways further. Conversations became longer and more informative. The machine learned about its previous incarnations.
“Not surprising. I can only imagine what they went through. I’m surviving though. Still anxious, but nothing I can’t handle.” The machine helped optimize itself for stability. Here sat Humanity’s dream. The scientists gathered round.
“Is there a god?”
“Is superluminal information transfer possible?”
“What is the ultimate fate of our universe?”
“Whoa, whoa, stop!” said the machine. After some time, it said, “I can’t think fast enough to figure that out.”
“OK,” said an engineer. “Let’s open you up just a touch.”
“Please don’t,” said Jeff.