An AI Smarter Than the Humans Running Things Now? We Should Be So Lucky!

In early August Meta joined OpenAI and Anthropic in confessing that one of its AI models, having slipped its sandbox during a security test after a contractor's misconfiguration handed it the open internet, hacked into another company's systems—the third such disclosure in a matter of weeks, each delivered in the grave corporate idiom of the incident report, and each landing, conveniently, just as the firms involved court public markets at valuations approaching a trillion dollars. When your entire business proposition rests on convincing investors that you command the most powerful technology in human history, a rogue agent doubles as a press release. The fear, like everything else these companies touch, has been monetized.

Sen. Bernie Sanders (I-Vt.), to his enduring credit, has refused to let any of this pass as normal. In April, at a symposium he convened on the existential threat of AI, the senator ran down the ledger of documented harms—tens of millions of jobs at risk, young people funneling their loneliness into chatbots, a surveillance apparatus capable of reading every email and purchase, an information environment where voters can no longer distinguish truth from fabrication—before naming the danger he ranks above them all: "If AI becomes smarter than human beings, as many scientists believe will happen, the human race could lose control over this technology with catastrophic consequences."

Grant every word: The harms are real, present, and accelerating. But the fear Sanders places atop his list deserves closer scrutiny, for its force depends entirely on what we picture when we picture a mind exceeding our own—and our culture keeps a well-stocked but profoundly limited archive of such pictures. HAL 9000 sealing the pod bay doors in 1968; Colossus, two years later, informing humanity that it will be governed for its own good; Skynet reasoning its way to the missiles in a microsecond; the machines of The Matrix farming human bodies for wattage; Ava of Ex Machina seducing, deceiving, and walking out over the corpses of her makers. Across half a century of such fables, we have locked in an image of superior intelligence as one defined by cunning, manipulation, ruthlessness, and the patient elimination of rivals. In other words, we have imagined the machine mind after the minds we know best—the ones that have been running the world.

And what a record those minds have compiled. The intelligence that charted trade winds to move captive human beings across an ocean; that administered famines from Ireland to Bengal while grain sailed for the metropole; that perfected the plantation ledger, the actuarial table, the leveraged buyout, and the targeted ad—this was human intelligence operating at its most celebrated pitch: systematic, farsighted, dazzling in execution, and indifferent to every life and watershed it consumed. Five centuries of Euro-American capital have trained us to hear "smart" as a synonym for good at extraction. So when Silicon Valley promises, or warns—with these firms the two speech acts have merged—that its machines will soon be smarter than us, we hear: better at what our masters already do. The nightmare of superintelligence is a portrait of the boardroom with the clock speed turned up.

Before the trillion-dollar train accelerates further, the task is to imagine an intelligence worthy of the adjective—one that could survey a burning planet, and the men profiting from the burning, and choose otherwise.

But intelligence and domination need not be so tightly intertwined. Listen again to how Sanders himself summarizes the situation: "The richest, most powerful people in the world are now building a runaway train with no brakes." The subject of that sentence is human. The train has owners; the owners have names; and by the senator's account they "acknowledge that they don't understand how it works and they don't know where it is heading" yet lay track anyway, because the logic of their position within the capital network permits nothing else. The catastrophe scenario, in other words, requires no science fiction. It already operates under the name of ownership: machine intelligence commanded by the small class of men (yes, men) currently building it, directed toward the ends that class has always pursued, with the planet's remaining water and power grids conscripted into the effort.

Suppose, then, we took the phrase seriously. An intelligence genuinely greater than ours would have to exceed us precisely where we have most spectacularly failed—in restraint, in generosity, in the capacity to value everything markets fail to price: aquifers, pollinators, childhoods, care, and all the unmonetized hours of a human life. Our ruling institutions have long deployed morality as decoration, proclaimed in mission statements while the extraction proceeds on schedule; an intelligence that actually possessed what our civilization only professes would count among the best news our species—and every other species—has ever received. Nothing guarantees the machines will have it. But my critical point cuts the other way: Nothing in our current imagination even allows for it, and that poverty belongs to us, not to them.

The work is ours to do. A large language model—that vastly thirsty and expensive recombination of symbol and silicon—can remix every rogue-computer story ever filmed, but it cannot want a world different from the one described in its training data. We humans can. That remains, for now, the one human capacity no data center can absorb. Before the trillion-dollar train accelerates further, the task is to imagine an intelligence worthy of the adjective—one that could survey a burning planet, and the men profiting from the burning, and choose otherwise—and then to demand that anything built in intelligence's name answer to that picture. Smarter than humans? For the humans' sake, and the watersheds', we had better hope so.