The Silent Revolution of AGI

The arrival of AGI will not be marked by fanfare; it is quietly transforming everything at an exponential rate. From the isolationist policies of the Qianlong era to the explosive growth of ChatGPT, history often deceives the present with the speed of past revolutions. This time, humanity may not have twenty years to establish a safety framework. As the technology iteration cycle compresses from centuries to mere months, how should we respond to this silent revolution?

Historical Patterns of Technological Revolutions

Looking back at history, we find that every technological revolution does not announce its arrival. Whether it’s the agricultural revolution, the industrial revolution, or the internet revolution, we only recognize their starting points in hindsight. AGI will be no exception; it will not notify us of its arrival, and by the time we realize it, significant changes will already be underway.

A more chilling thought is that, after the advent of nuclear weapons, it took humanity 20 years to figure out how to ensure safety. This time, history may not grant us that long.

For individuals, this window may be even shorter.

The Threshold is Behind Us

In 1793, Lord Macartney was sent by Britain to Qing China to request the opening of trade ports. Qianlong wrote a calm and measured response to King George III, stating, “The Celestial Empire is abundant in resources and lacks nothing; there is no need to rely on foreign goods to meet our needs.”

At the time of writing this letter, steam engines had been roaring in British textile factories for nearly thirty years. Qianlong was unaware of this and felt no need to know. He operated within a framework that had been in place for thousands of years, where Qing was the center of the world and foreign nations came to pay tribute.

Fifty years later, the Opium War broke out.

AGI will not have a singular moment of arrival but rather a moment of reflection.

A research institution called Meter specializes in tracking how long AI can independently sustain a complete task without intervention. By early 2024, cutting-edge models will reliably complete tasks in minutes, such as answering a question or writing a piece of code. By 2025, this will extend to five to ten hours, equivalent to the time an engineer would take to independently complete a full functional module. These figures double every seven months, showing no signs of leveling off.

During this time, due to project needs, there have been numerous exchanges in some cutting-edge laboratories, leading to the feeling that the threshold may not be far off, even suggesting that it has already begun to blur.

I do not know when or how society will reach a consensus that AGI has arrived.

The Speed of Change is Unmatched

In 356 BC, Shang Yang began reforms in the Qin state, re-measuring land, abolishing noble hereditary titles, and implementing military meritocracy. Twenty years later, he was executed by being torn apart. A hundred years later, Qin Shi Huang unified the six states.

The fruits of Shang Yang’s reforms were not reaped by the one who planted them.

In 1765, Watt improved the steam engine. Factories began replacing manual labor with machines, and coal mines started using machines for pumping water. The first factory law limiting child labor hours in Britain was not passed until 1833.

It took nearly seventy years from the steam engine to the factory law.

In the 1870s, the second industrial revolution simultaneously erupted in Germany and the United States. After the Franco-Prussian War, Bismarck received reparations and coal mines, and immediately began constructing steel mills. Germany transformed from an agricultural nation to Europe’s largest industrial power in less than thirty years.

In 1994, China connected to the internet, and the first email was sent from Beijing. A decade later, Taobao was launched. Another decade passed, and WeChat reached one billion monthly active users.

From the first email to transforming daily life, it took twenty years.

In November 2022, ChatGPT was released, and within two months, it gained two hundred million users. By 2025, even that speed began to seem slow. Manus was acquired for billions of dollars less than a year after its launch. OpenClaw, referred to as “Little Lobster” in the AGI Bar community, surpassed Linux in GitHub stars within a month, a feat that took thirty years for Linux to achieve.

Every technological revolution deceives the present with the speed of the past, and each time, humanity’s ability to establish coping mechanisms lags behind technological advancement. But for this time… I do not know.

The Answers to Safety Lie in Politics

In December 1952, London was shrouded in thick smog for five days, with visibility so low that bus drivers needed passengers to guide them with flashlights. Over four thousand people died within those five days, with another eight thousand succumbing in the following months.

Solutions existed on a technical level. Churchill’s cabinet reviewed the report and shelved it. Industrial lobby groups claimed the costs of reform were too high, the Treasury said it was not the right time, and the Health Ministry argued that the death toll was within normal winter levels. Four more years passed before the Clean Air Act was finally passed in 1956.

This history has been compressed into four characters in our textbooks: the London Fog.

At that time, it was a national and epochal pain.

“How dangerous is too dangerous” is a question that engineers cannot answer.

The steam engine will not tell you how much sulfur dioxide is excessive. That figure requires doctors to tally death counts, journalists to report, the public to protest, and lawmakers to vote; each step is a negotiation between people.

Perhaps we still need to endure countless lessons before finding safe answers.

This Time, We Do Not Have Twenty Years

In October 1962, news of the Soviet Union deploying nuclear missiles in Cuba reached Washington.

The next thirteen days marked humanity’s closest approach to nuclear war. The exchanges of telegrams between Kennedy and Khrushchev still send chills down the spine today; both men knew that any slight miscalculation could lead to the end.

Fortunately, from a god’s-eye view, everything passed smoothly.

Prior to that, the two nations had spent nearly twenty years navigating crises like the Berlin Blockade and the Korean War, exploring each other’s boundaries in real high-pressure confrontations. Khrushchev knew how far he could retreat, and Kennedy knew how far he could not push.

This “knowing” was earned through twenty years of probing.

During those twenty years, the RAND Corporation established a complete theory of nuclear deterrence, a hotline was set up, and the first round of arms control negotiations began. A framework was created from scratch.

Humanity took twenty years to understand each other’s boundaries after the emergence of nuclear weapons, but we do not have twenty years to address the risks of AGI.

When the crisis truly arises, there may not be thirteen days to exchange telegrams. The RAND Corporation of the 1950s knew what it was doing. Corresponding work today is still unknown to us.

New Tools First Belong to Old Trust

In 1455, Gutenberg printed the first batch of Bibles. Before this, books in Europe were hand-copied by monks, taking months to produce a single Bible. After the printing press emerged, the first to use it on a large scale were the churches. They printed indulgences, papal decrees, and mass manuals, maximizing the new tool’s potential through the institution that controlled the most widespread communication channels.

In 1517, Martin Luther nailed his Ninety-Five Theses to the church door. This document spread throughout Germany in two weeks, leveraging the printing and dissemination network established by the church over decades. Luther used the infrastructure built by the church to undermine the church itself.

Those who quickly adopt new tools are always those who have built the deepest trust on old tools.

Leverage amplifies existing advantages.

Trust is a slow variable; it does not suddenly emerge with the introduction of a new tool. With frameworks not yet established and regulations not yet in place, trust is one of the few things we can work on in advance.

This window is still open.

I cannot guarantee that these five judgments are correct, but if the general direction is not wrong, the time left for adjustments may be shorter than most people think.