64 Years from Watt's Patent to the Factory Act — Can AI Regulation Catch Up?── The time gap between adoption and regulation determines the scale of harm

In September 2026, an Anthropic researcher resigned after warning that the probability of AI causing human extinction exceeds ten percent. Will AI become a general-purpose technology that reshapes society the way the steam engine did, or will it become a risk that outpaces our ability to control it? The historical record suggests that the answer depends on one variable: the gap between the speed of adoption and the speed of regulation.

01A developer resigned and a promoter spoke of trillions in the same week

The Anthropic researcher's resignation was unusual as a warning from inside an AI company. The same week, NVIDIA's Jensen Huang called AI an infrastructure on par with electricity and said its scale would be measured in trillions of dollars, not billions. A survey reported the same week found that 39 percent of Americans believe AI does more harm than good.

The builders talk of vast potential. An insider walks away citing existential risk. Nearly four in ten members of the public fear the harm. The fact that all three are happening simultaneously sharpens the question. It is not whether AI is good or bad. It is how far behind the mechanisms for preventing harm will fall relative to the speed at which the technology arrives. The record of the Industrial Revolution tells us what that gap has cost before.

02Benefits arrived first, and regulation lagged by sixty-four years

In 1769, James Watt received a patent for the separate condenser. The improvement cut the coal a steam engine consumed and turned it from a mine pump into a factory power source. In The British Industrial Revolution in Global Perspective, Robert Allen argued that Britain's combination of high wages and cheap coal made it rational to replace human labor with machines. The Industrial Revolution was driven not by genius alone but by the price ratio between labor and fuel.

Mechanization raised output. It also left a clear record of harm.

From Watt's patent in 1769 to the Factory Act in 1833: sixty-four years. From the Luddite unrest in 1811 to the Factory Act: twenty-two years. From Chadwick's report in 1842 to the Public Health Act in 1848: six years. Technology spread first, harm became visible, and regulation followed. That sequence never reversed.

03AI spreads orders of magnitude faster than the steam engine did

It took decades for the steam engine to reach factories across Britain. Electricity followed a similar timeline. Edison's Pearl Street central station in New York began operating on September 4, 1882, serving fewer than ninety customers. Electricity needed decades more to reach every corner of daily life. During that time, wiring standards and safety rules were put together piece by piece.

AI is different. The "trillion-dollar infrastructure" Huang describes is already under construction. The fact that a researcher's resignation and a promoter's massive investment announcement occurred in the same week signals how fast this technology moves. With steam and electricity, slow adoption gave society time to adapt. With AI, that time is shorter. In Superintelligence, Nick Bostrom argued that once an intelligence exceeds the human level, the subsequent rise may be steep. His point was that no time for preparation might remain.

04General-purpose technologies do not stay inside one industry

The steam engine did not change only one process. It changed where factories stood, how labor was organized, how cities were structured, how children spent their days, and how public health was managed. Economists call such technologies general-purpose technologies: they reshape not a single industry but the arrangement of an entire society.

When Huang called AI "infrastructure on par with electricity," he was claiming that AI is a general-purpose technology. Electricity was used for lighting, power, and communication. AI, too, spans text generation, image recognition, and data analysis, with no boundary confining it to one domain.

The defining feature of a general-purpose technology is that its reach cannot be predicted in advance. The builders of the steam engine did not foresee urban sanitation crises. How far AI will reach is likewise not yet visible in full.

05AI has already entered pharmaceutical materials review

In pharmaceutical promotional materials review, AI is already in use. It generates drafts, searches for supporting literature, and suggests alternative phrasings. A literature search that once took half a day now sometimes takes minutes.

At the same time, the shape of the harm is becoming visible. A plausible-looking citation points to a paper that does not exist. A phrase hinting at off-label use slips into smooth prose. An author never opens the original paper. None of this is a defect of AI itself. It occurs where the process for using AI has not been defined.

The core problem is that organizations have not drawn a clear boundary between processes where AI is used and processes where it is not. In 1833, the Factory Act drew lines: no employment under age nine, inspectors to enforce the rule. AI raises the same structural question. Which processes may use AI? Who checks the output? Without a boundary, harm spreads quietly.

06Regulation lags behind adoption for three structural reasons

Why does regulation always fall behind technology adoption? Three structures are at work.

First, the benefits of a technology are visible from the moment of adoption, but the harm appears with a delay. The steam engine raised output immediately. It took decades for the deterioration of children's working conditions to be recognized as a social problem.

Second, those who benefit from a technology resist regulation. As Allen showed, mechanization was a rational investment for factory owners. Regulation raises costs. What the Luddites tried to break was not the machines themselves but the power that rewrote their lives without consent. The Factory Act later gave that anger institutional form.

Third, as Joel Mokyr argued in The Lever of Riches, technological creativity within a single society rarely lasts long. Vested interests and inertia accumulate, and innovation slows. Regulation, once created, is also prone to rigidity. Even when the pace of adoption accelerates, the pace of regulation does not keep up.

These three structures operate in the Industrial Revolution and in AI alike. What changes is only the speed of adoption. The faster adoption moves, the greater the harm that accumulates before regulation catches up.

07Internal rules can be built before national law catches up

There is no need to wait for national regulation. In pharmaceutical workplaces, internal operational rules can be built first. The Factory Act of 1833 began with a national inspectorate of just four people. Starting small is enough.

The Factory Act of 1833 put just four inspectors in charge of every textile mill in the country. Four people could not cover them all. Yet the fact that a boundary existed as an institution mattered in itself. Internal AI rules work the same way. They do not need to be perfect. What matters is that they exist and that they are updated based on actual failures. That is how the gap between adoption and regulation can be narrowed from the ground.