We are in the middle of what might be called a global AI freakout: Researchers warn of systems escaping human control, tech executives propose slowing their own frontier models, and Washington is consumed by debates over regulation. Anxiety over jobs, safety, data centers, and the concentration of wealth and power is coalescing into a huge backlash.
Hovering over every proposal for restraint is China. If America slows down, the idea goes, China races ahead. House Speaker Mike Johnson put it plainly this week, saying there can be no moratorium on AI development because America would lose its edge to China.
President Donald Trump, who has said “whoever wins AI wins,” is scheduled to meet Chinese President Xi Jinping on September 24, with AI expected on the agenda. He should rethink his maximalist position, which projects indifference to the welfare of his citizens. The meeting with Xi offers a chance to test whether the AI race really must proceed without brakes.
Into this uproar has stepped an unlikely advocate of restraint — billionaire Bill Gates. On August 26, the Microsoft co-founder published “The turbulent AI era is here. The choices we make now are critical.” Gates remains enthralled by AI’s possibilities in medicine, science, and education, yet believes its ability to substitute for human cognition could make this technological revolution different. New jobs will emerge, he writes, but without the right policies, “there will be far fewer than exist today” — potentially upending a civilization essentially built around human employment.
Gates proposes designating some occupations “Human Reserved,” comparing them to nature reserves society preserves despite the economic value of developing them. He cites the “irreplaceably human” caregivers who tended his father through Alzheimer’s and also proposes taxing robots and AI tokens to discourage some worker replacement and finance a stronger safety net. Most strikingly, he accepts some resulting economic inefficiency, arguing that conventional measures overlook the broader value of work. Call this the Gates Doctrine: A machine’s ability to replace a human being more efficiently does not settle whether society should let it.
This makes Gates, of all people, something resembling a Luddite. The word today evokes someone baffled by an iPhone or reflexively hostile to technology, but the historical Luddites were skilled English textile workers whose trades required years of apprenticeship. They worked with machinery and understood it intimately.
Their rebellion concerned machinery used to replace skilled labor with cheaper workers, depress wages, and transfer productivity gains to factory owners. Their argument was ultimately about the purpose of technology: Greater output did not automatically constitute progress when the people producing it became poorer and less secure.
The Luddites eventually lost more than an economic battle. They lost the argument over how we think about technology. A powerful story took hold: Technological progress is inevitable and always ultimately beneficial, because of greater efficiency and productivity (of whoever’s working). Sure, machines destroy jobs, but displaced workers find new and better ones; productivity rises, society grows richer, and the market somehow produces enough work for everyone. There is an almost religious faith embedded in this idea — that, however wrenching the disruption, all will be fine.
For two centuries, history seemed to vindicate that faith. The steam engine, electricity, and the digital computer destroyed countless occupations, yet mass employment survived, and living standards rose spectacularly. Against that record, the Luddites came to look ridiculous: men swinging hammers against the unstoppable march of progress.
Artificial intelligence now resurfaces the central Luddite question: What if this time really is different? Tech maximalists still answer with the technological inevitability idea: super-intelligent AI is coming no matter what; jobs will disappear; companies will automate; everyone will adapt. This narrative turns thousands of decisions made by governments, corporations, universities, investors, and engineers into some sort of law of nature.
But AI, like all technology, is a human creation. Governments financed research, universities trained scientists, investors supplied capital, and companies built the models and data centers. The institutions that accelerated AI development also possess considerable power to steer it.
I would go beyond the Gates Doctrine to a broader proposition. Call it, with self-indulgence unbecoming to a columnist, the Perry Doctrine: Technology is not destiny, and efficiency is not an end in itself. Both are instruments whose value derives from their contribution to human welfare. Societies can encourage useful technologies, steer their development, slow applications demanding more thought, and prohibit those whose costs overwhelm benefits.
Applied to AI, this can still be enthusiastically pro-technology. AI that discovers cancer treatments or gives poor children access to superb tutoring deserves acceleration. The same principle demands scrutiny when AI allows a corporation to eliminate 20,000 good jobs while providing essentially the same service. The productivity gain belongs in the calculation alongside livelihoods, wages, communities, and the distribution of the resulting wealth. Efficiency remains valuable because of what it does for people.
China presents the hardest test. Frontier AI carries immense military, scientific, and economic power, and American restraint can create a Chinese advantage. Maintaining technological leadership may therefore require America to develop systems comparable to anything China can build. Yet capability and deployment are different things. America can remain at the frontier while preserving human roles where they have social value and restricting dangerous applications. We may have to build what China builds; we do not have to become the society China becomes.
There are also areas where neither country benefits from an unconstrained race. AI-directed cyberattacks, autonomous military escalation, catastrophic biological applications, and systems escaping meaningful human supervision threaten Beijing as surely as Washington. China has itself called for international cooperation on AI governance, even as a state newspaper attacked U.S. proposals to slow development as an attempt to preserve U.S. dominance. The tension between competition and common danger already exists.
THE ANTI-AI REBELLION JUST FOUND ITS FIRST REAL TARGET — AND IT HAS A STREET ADDRESS
Trump’s September 24 meeting with Xi offers an opportunity to test the waters. They could begin with a modest proposition: There are things technology may make possible that neither country should permit merely because they have become possible. From there could come reciprocal safeguards and eventually verifiable agreements around the most dangerous applications. Gates himself argues that some U.S.-China cooperation will be essential to governing AI.
The great question of the AI age is whether governments still possess the confidence to direct technological change toward human purposes — and whether Washington and Beijing can recognize that they share at least some of those purposes. To decide in advance that we have no choice would be defeatist — and unintelligent.
Dan Perry, a computer scientist by education, is the former London-based Europe-Africa Editor and Cairo-based Middle East Editor of The Associated Press, chairman of the Foreign Press Association in Jerusalem and the author of two books.