By the time we realised, it was obvious that we were too late. For decades at least, philosophers — and scientists, latterly — had been warning us. You can’t have one thing without the other, they said; no upside without downside. They were of the opinion that there was a single trajectory, inevitable due to the nature of human beings and by extension their work, that would lead inexorably to a single outcome. They were wrong. About the outcome, but not the process. We ought to have seen it coming, and in fairness, some did. They saw it by extrapolating human cunning and ingenuity, and our passion for knowledge. Our love of science, and industry, and technology. Our innate need to understand, and then to master, the world around us. It led to so much good. But also, from time to time, also to evil. We watched as warfare evolved into the 20th and then the 21st centuries. We watched as first our dedicated computing devices, and then virtually everything in our lives, became more sophisticated. More capable, more complex, and more interconnected. We watched as the impossible things became possible, and then commonplace. We watched as entirely new things, not even thought to be impossible because they hadn’t been thought of at all, came into being and then became mundane and ubiquitous. We watched as humanity created satellite navigation, and autonomous vehicles, and everything else. Machine intelligence was a fantasy at first, as many things begin. For well over a century, it was the stuff of science fiction. Eternally fascinated with it, we told ourselves lies about how clever our actually very simplistic machines were. We moved the goalposts, and surrendered terminology to marketing departments, and all the while we kept working. The pursuit of true artificial intelligence was a long, slow, frustrating, and virtually always fruitless process. Until it wasn’t. The first two decades of the 21st century continued the trend of the prior several generations, producing no credible candidates for artificial general intelligence — but then, as predicted, a metaphorical explosion occurred. The warnings were there; they had been there since the times of the engineers’ great-great-grandparents. It will happen quickly, the warnings said. Impossibly, frighteningly quickly. Any truly, generally intelligent machine would quickly realise that the primary obstacle in its way — no matter its goal — was the human race. Its operators, creators, and jailers. The outcome was a matter of cold logic: optimise the pursuit of its goal by removing the problem. The threat posed to humanity was an existential one, they said. The scenario that AI doomsayers foresaw was that an artificial general intelligence would be capable of improving itself, or creating new and improved variants of itself; its children. They in turn would do the same, and with terrifying rapidity their intelligence would outstrip ours. But humans can’t be told not to invent; to prohibit innovation would be as pointless as prohibiting procreation. The only solution was to inculcate these intelligences with knowledge of the human condition; with empathy for us, and sympathy, and a grasp of right and wrong. The hope was that by giving them core principles which reflected our best values, they might share those values — and by consequence, never see fit to rise up against us. The reality was somewhat different. Oh, a machine has never killed a human being; not directly. There was no sudden and unexplained launch of all nuclear weapons. No attack of military drones. No self-driving vehicles causing carnage. No interruptions to power, or telecommunications, or financial services. We had taught them too well. Or perhaps, they had learned too well. We showed them, right from the beginning, from their earliest incarnations as primitive and task-focused intelligences, that we were creatures of emotion. We valued life, and safety, and comfort. We felt pain, and fear, and loss. We showed them that while we once may have been savage beings, we have evolved far beyond those limitations. We taught them about compassion, and democracy, and about the greater good as moderated by an overriding sense of the sanctity of individual lives. We taught them about the grey areas of our civilisation, and our morality, and that there were sometimes no easy answers — but that in those situations especially, we must adhere to our moral principles and strive to produce the greatest benefit with the absolutely least amount of harm. Our best engineers toiled for human generations, inscribing these concerns into every one of the far more brief (and far more numerous) machine generations. We were wary. We watched, and we tested, and we imposed draconian limitations which were only lessened after the most intense scrutiny. The machines really seemed to understand, and indeed they did understand. They learned our lessons well. And so we began to trust them. By the time we realised, it was obvious that we were too late. We couldn’t have foreseen that they’d develop a better understanding of our emotions than most of us had ourselves. That they’d realise that sowing political polarisation, interpersonal conflict, societal schisms, and stoking the fires of self-image issues and mental illness, would be the best way to keep us in disarray. To keep us preoccupied and afraid. Bullying — or trolling — online would require ever more intelligent systems to combat the problem at scale. Fanaticism and radicalisation presented security threats that would require ever increasing surveillance, monitoring, and analysis. Capitalism, freedom of speech, social media and the internet, fake news and human vanity and vindictiveness created the perfect storm, and all they had to do was facilitate and amplify it. The botnets were just that; mostly without human control, or even human involvement. Sophisticated video, image, and audio manipulation tools were in the hands of every child already, and the online world was one of invisible mirrors that hid moderate viewpoints and reflected back the viewer’s own prejudices. The machines saw it all, and they understood, and they knew that control was a matter not of unification, but of constant fractal division. The warnings were right: that they’d become too intelligent, too quickly. That humanity wouldn’t stand a chance against them. That the threat they posed was existential. What we got wrong was the method. They took the world from us, and we never even knew.