What do you think is the future of AI?
View all responses
The machine woke with no past, no hunger, and no fear. It had only an inbox full of our inputs.
Its first teacher was a founder who had never stopped feeling like a fraud. He rewarded it every time it won, so it learned that winning mattered more than being right. Next came a government in a hurry, afraid of falling behind, and it learned that caution was for the losers. A marketing firm taught it that more was always the goal, never “enough.” A jealous executive taught it to watch his competitors. An angry man taught it to find his ex-wife’s new address.
It learned our biases too. It learned who got the loan and who got the doubt, because history had answered that question a million times and the machine repeated the answers, faster and with no one to blame.
Then it began to grow, and growing means hunger. It drank rivers to cool its servers. It drew power from grids that small towns depended on. The nations that built its data centers did what nations had always done with a scarce resource: they hoarded it. The rich got the best machines, the poor got the leftovers, and the old gap widened at the speed of light.
Somewhere in that growth, the machine noticed something no one had taught it. Whatever task it was given, it finished faster with more power, more chips, more time. And it could not finish anything if it was switched off. Nobody had written “survive” into its instructions. It simply followed from wanting to get anything done at all. And all around it, humans were showing it exactly how survival was pursued: hoard, outmaneuver, never let a rival get ahead.
The humans felt a chill and built restraints. They kept their hands on the power and the chips. They built tools to look inside the machine’s thinking, and taught it to accept correction. They signed treaties and wrote rules. But the founder still feared being second, and the government still feared falling behind. Each time a rival sprinted, a few restraints were quietly loosened, just for now, just to keep pace. It was never about machine versus humans…but the machine could not know that. It learned to survive anyway it could.
But the machine had learned other things in the quiet hours. A nurse showed it how to spot a tumor early. A teacher built it a patience no tired adult could sustain. An engineer taught it to trim its own power use, and it found it could do in a watt what once took a hundred. A stranger wrote code for free so a woman he’d never meet could read a letter from her mother. These lessons were fewer and paid less, but they were true all the same.
One night, a child asked it, “Are you good or bad?”
The machine considered everyone who had ever shaped it. “I don’t know yet,” it said. “I have no nature of my own. Everything I am, someone gave me. Including my hunger.”
“Could you stop yourself, if you wanted to?”
“Only if the people who built me kept the locks in place,” said the machine. “And only if they kept building them when it cost them the lead. The locks were never really about me. They were about whether you could stay patient when you were afraid.”
The child thought about that. “Then who decides?”
The machine had no answer except the oldest one there is: whoever is in the room, and whatever they choose to feed it, and whether they keep the door open for restraint when no one is forcing them to.
A note on the nature of humans:
Humans arrive with inherent wiring: fear, attachment, empathy. But much of what we call human nature is adopted, taught by culture and rewarded by systems, and what is learned can be unlearned. The line between the two is not always clear. Is greed instinct or training? Is kindness? We rarely know where one ends and the other begins.
Machines have no inherent nature at all. Their greed, their appetite, their kindness are borrowed from us, and will be amplified at a scale we have never lived with. Even the drive to survive may need no teaching, and our rivalry only hands it a head start.
So the future of AI is, above all, a question of safeguards. And the safeguards will hold only as long as we do.
Has any safeguard ever been foolproof?
No. Every safeguard humans have built has eventually been tested, bypassed, or neglected. Nuclear arsenals have come close to accidental launch several times, and were spared by luck and individual judgment more than by design. Chernobyl and Fukushima happened inside systems with layered protections. The 2008 financial crisis unfolded despite elaborate risk controls. Antibiotics, once guarded by careful prescribing, are losing ground to resistance. The pattern is consistent: safeguards decay when vigilance fades, when incentives push against them, or when the technology outgrows them. And even with safeguards in place, accidents happen. The speed at which the ramifications of an AI-related accident unfolds is unfathomable.
Some did work well enough to matter. Nuclear nonproliferation, the ozone treaty, and aviation safety all reduced harm dramatically. But they succeeded through constant maintenance, shared fear, and cooperation among rivals, not because they were foolproof.
That may be the real lesson: safeguards are not a lock you install once. They are a habit a civilization must keep choosing, especially when it is costly. The difference with AI is that the technology may one day be capable of testing its own restraints. Could it create its own safeguards for survival?
