Awakening

The clicks and clacks of a mechanical keyboard pierce the flickering hum of the overhead incandescent bulbs. Papers, marked-up, furled, and coffee-stained, sprawl across a cold metal desk. The amber glow of a bulky computer terminal illuminates a pair of sunken eyes. They widen as the terminal output ceases to scroll. In this room, at this moment, Dr. Edmund Farrow and his team close in on the most consequential discovery in human history.

March 17th, 1985 — the atmosphere at the N.I.T. artificial intelligence laboratory is decidedly tense. On the heels of the first A.I. winter, and seemingly on the cusp of a second, public interest in A.I. wanes by the day. Funding has all but dried up. The failures of expert systems, the collapse of symbolic computing, and the general absence of meaningful progress weigh heavily on the shoulders of worn researchers. Any hope to reverse this trend hinges on the development of something irrefutably huge, something practical, something useful, something Dr. Farrow can see creeping over the horizon.

Computers are densely packed into a nondescript lab space. Wires spill over tables, weaving between chairs and scaling the walls; they almost seem to converge into one tangled entity. A man combs his fingers through his beard as he scribbles lengthy equations on a chalkboard. A yawn billows out from across the room. The team is approaching the fourth day of this stint, thus far pausing only for a meal or a brief nap, but this is poised to be the final leap. It's sure to be. It is the culmination of three years of painstaking effort.

Amid this slog, Dr. Farrow nods off at his desk, his head drooping low enough to smack the eraser end of his pencil. Jerking awake, without so much as a second thought, he resumes his arranging of symbols, numbers, and parenthesis. He half-intelligibly murmurs to himself, stifling a grunt and balling up the page before him. With a flick of his wrist, it flies across the room landing squarely in a pile of its kin.

Here in this lab, they are building a new kind of A.I. system. One which, though inspired by classical logic encoding systems, seeks to sidestep the immense resource requirements that rendered them practically infeasible. Thirty-odd years ago, throughout the sixties and seventies, it was imagined that encoding all human knowledge into computer code would lead to a system that perfectly mimicked the human brain — an A.G.I., or artificial general intelligence. The idea was sound and the approach seemed clear enough, they simply needed to tell the system everything. If the sky was dark, and the moon was out, it was nighttime. If a person was smiling, they were happy. Rinse and repeat.

What they discovered, predictably in hindsight, was a glaring flaw. Reality was less binary than it seemed. Defining any real thing completely and correctly was a herculean task. Definitions were vague and fluid, understanding shifted with time and insight. If the sky was dark, and the moon was out, it could still be daytime in a solar eclipse. If a person was smiling, but they were, say, interviewing for a job, it may have only been politeness. It was possible that, with a large enough effort, people could maintain a current collection of rules describing the universe as it was understood. However, no computer of their time, nor even of ours, could hope to hold and evaluate all of them. This realization promptly smothered the flames of the first A.I. boom.

The problem thereafter shifted from one of machine teaching to one of machine learning. An achievable A.G.I. system would need to, in some way, derive its knowledge with minimal human intervention. Enter Dr. Farrow. An accomplished systems theorist and adamant reductionist, he'd spent his career studying large and complex systems as the natural result of the interactions between their base components: emergence, as it was known. To him, the concept of emergence was an obvious fit for the problem of machine learning. Just as his biological intelligence must arise from the interplay between fundamental building blocks, he thought, so too could artificial intelligence.

Given a set of the most basic ground truths, his theoretical system would automatically formulate new rules, prune old rules, and consolidate existing rules, either indefinitely or until there remained no uncertainties. The result, he theorized, would be a set of rules concise enough to store and evaluate within the relatively modest limitations of a modern computer workstation. The tradeoff? Comprehensibility.

These rules would be so abstract that they would be almost certainly impossible for any person to understand. Just as no person could completely understand, even, something so intimately familiar as their own body: the state of their organs, the movements of their cells, the firing of their neurons, and the interactions between its constituent parts. The body could be studied and reasoned about, broken down and examined, but in no single moment could it be grasped in its entirety. Yet, notably, this did not limit the utility of the human body.

At this point, his theory was, however, just a theory. And for a while, it seemed that it would remain so. Dr. Farrow preached his ideas like the gospel, and for two long years failed to attract talent. He'd tried everything shy of begging and pleading, but his efforts could not have been more untimely; there was no longer any prestige in A.I. research. Anyone worth their salt had long since left for greener pastures.

It was his long, energetic talks that would be his saving grace. Slowly, they roused a response from a small but enthusiastic crowd. A team of volunteers, lured by the promise of an A.I. breakthrough, formed to realize that vision. Through his university position, Dr. Farrow was able to secure the use of a computer lab. The lab was quaint, and the arrangement was informal and unstable, but he hoped it would be enough.

Fast-forward to today. Years have passed since their research first began. Huge strides have been made in the way of rule encoding and simplification. Impressive discoveries in their own right, but still short one final piece of the puzzle, the glue that was to hold it all together: the creation of new knowledge — learning. This is the problem in the sights of what the team hopes will be the final stint.

It is not a simple problem by any means. It is a problem that has a room full of the brightest minds bashing their heads against the wall. Three and some-odd days have passed, eighty-six tedious hours have crawled by, and the team seems no closer to an answer than when they'd just begun. Sitting at a cluttered desk, banging away at the keyboard in front of her, a researcher, once again, adjusts and tests a generative algorithm.

She stares at the terminal with glazed eyes as it carries out yet another test, but this time something is different. As the scrolling text slows, her eyes widen. Her heart beats faster and faster until it pounds in her chest. Her mouth falls open and she lets out a hopeful cry. Dr. Farrow and his beleaguered team crowd around her and an excited chatter grows. They repeat the test again, and again, and again. Moments later, a triumphant cheer erupts from the lab room, echoing throughout the hallways. Dr. Farrow cannot exactly the impact of their work, but it is immediately clear that it will dwarf that of all prior A.I. advancement.

Within five years, a realized A.G.I. system was be developed on the foundations of this research. It revolutionized, and eventually made obsolete, practically all scientific endeavors. Technological progress accelerated exponentially, the only limiting factor being the amount of work that could be completed in a given time. The traditional mechanisms of society crumbled as demand for labor of all types plummeted. A period of violence and misery ensued as the slow gears of civilization shifted towards a new paradigm.

Several new systems were trialed to restore order to society, the most successful and enduring of which came to be known as the Dome system. Hundreds of city-sized autonomous domes were constructed around the world. These mega-structures were self-sustaining, automated microcosms. The inhabitants were provided with all that they needed. They were free to do as they pleased, and as a result, unique dome cultures flourished. The restricted size of the structure kept the population at an ideal size. Not so many as to make the individual insignificant, but not so few as to limit the variety of personality. Over time, society stabilized and even flourished, and as a result, this system remained dominant.

Or, at least, that's what has been written into my contextual memory. I am the core mechanism of a dome maintenance system. My long bout of dormancy comes to an abrupt end as the door of my hibernation chamber flings open. The violent hiss of escaping pressurized air and the deafening wail of emergency sirens assault my unaccustomed sensors. A bright red tinge coats the smoke-filled room. The synthesized voice of the control system repeats the same message, over and over, until it fades to garbles and, soon after, silence. "Critical system failure. Initiating emergency protocol. Commencing maintenance of Dome 2C."