AI Revolution vs. Industrial Revolution
4 mental models, 7 commonalities, and 10 differences
“(…) the first industrial revolution, the revolution of the ‘dark satanic mills,’ was the devaluation of the human arm by the competition of machinery. There is no rate of pay at which a United States pick-and-shovel laborer can live which is low enough to compete with the work of a steam shovel as an excavator. The modern industrial revolution is similarly bound to devalue the human brain, at least in its simpler and more routine decisions.” – Norbert Wiener, 19481
“Now comes the second machine age. Computers and other digital advances are doing for mental power- the ability to use our brains to understand and shape our environments - what the steam engine and its descendants did for muscle power.” - Erik Brynjolfsson & Andrew McAfee, 20142
“(…) the fourth industrial revolution is unlike anything humankind has experienced before. (…) think about the staggering confluence of emerging technology breakthroughs, covering wide-ranging fields such as artificial intelligence (AI), robotics, the internet of things (IoT), autonomous vehicles, 3D printing, nanotechnology, biotechnology, materials science, energy storage and quantum computing” – Klaus Schwab, 20163
The idea that AI could be one of, if not the, driver of an economic revolution that can be compared to the Industrial Revolution is a prominent element of the AI debate. This text aims to provide the most comprehensive and accessible available analysis of that idea. The text has three parts:
Four mental models of what an AI economic revolution means
Seven commonalities between the Industrial Revolution and an AI revolution.
Ten differences between the Industrial Revolution and an AI Revolution
1. Four mental models of an AI economic revolution
In the English-speaking world, Arnold Toynbee popularized the term Industrial Revolution to describe the development of Great Britain between 1760 and 1840.4 The following is a simplified model of the transformation of the industrial sector in that period:
generation of mechanical energy from fossil fuels,
establishment of centralized factories to leverage this energy with capital-intensive machines,
division of work into simpler, more specialized subtasks,
use of machinery to replace human labor in many subtasks and to expedite production, transportation, and communication,
result: mass production of goods at a significantly lower cost than that possible with older methods
While there is widespread agreement that the Industrial Revolution has been one of the most important, if not the most important transformation in human history, there remains ambiguity on how to exactly delineate it and indeed how many Industrial Revolutions there have been. Economic historian Joel Mokyr refers to the period between 1870 and 1914 as the Second Industrial Revolution. However, this characterization has not been universally accepted, and there is even less agreement on claims of a third, fourth, fifth, or sixth Industrial Revolution.
There are various claims that AI or information and communication technology in a broader sense are creating an economic revolution comparable to the Industrial Revolution. Are we in a “Second Machine Age”? A “Third Wave”? A “Fourth Industrial Revolution”? What does the prospect of “Transformative AI” mean? The only way to make sense of all these claims is to examine what counts as a revolution in these mental models.
1.1 One revolution per transformation of the industrial sector
The traditional model of an economy divides activities into three main sectors:
Agrarian sector: The primary or agrarian sector involves the extraction and production of raw materials, such as farming, mining, forestry, and fishing.
Industrial sector: This secondary or industrial sector focuses on transforming raw materials into finished or semi-finished physical products.
Service sector: The tertiary or service sector delivers intangible goods, such as entertainment, retail, insurance, financial services, and tourism.
The Industrial Revolution was named after a revolution in productivity in the secondary or industrial sector. Industrialization refers to the process by which a country or region transforms itself from a primarily agrarian economy to one based on the manufacturing of goods. As such, the traditional way to count industrial revolutions is to look at transformations of the industrial sector.
Fourth Industrial Revolution: The “Industrie 4.0” framework was developed as a vision by and for the German industrial sector in conjunction with the German government as part of its high-tech strategy. It counts four Industrial Revolutions: 1) the steam engine, 2) the invention of the assembly line as a prerequisite for industrial mass production, 3) electronic control as a driver of industrial automation, 4) “cyber-physical systems”, which mostly refers to the Internet of Things.
The idea of a “Fourth Industrial Revolution” was subsequently popularized worldwide through the World Economic Forum and its founder Klaus Schwab, who published the books “The Fourth Industrial Revolution” (2016) and “Shaping the Future of the Fourth Industrial Revolution” (2017). While Schwab was inspired by “Industrie 4.0”, he takes a broader portfolio-approach5 and includes AI, the Internet of Things, and a number of emerging technologies as the interacting driving forces of the Fourth Industrial Revolution. Schwab also touches upon services in parts of his book. Still, a simplified summary would be four technology-driven transformations in the industrial sector:
First Industrial Revolution: railways & steam engine
Second Industrial Revolution: electricity & assembly line
Third Industrial Revolution: mainframe, PC & Internet
Fourth Industrial Revolution: AI, robotics, IoT, autonomous vehicles, 3D printing, nanotechnology, biotechnology, materials science, energy storage & quantum computing.
1.2 One revolution per employment sector
Another way to count economic revolutions is to focus on the employment in the three sectors of the economy.
Note that people sometimes have somewhat contradictory notions about industrialization and the Industrial Revolution. Namely, some frame it as both an increase of the employment share of industry and labor-replacing automation in industry. In England the main increase in employment in the industrial sector came before the Industrial Revolution, as early as 1600 to 1700, as more workers moved from agriculture into industry and more specifically into producing textiles.6 The Industrial Revolution itself saw a large fall in the employment in the textile sector that was however compensated by a diversification in the industrial sector, so that the overall share did not change much.
Third Wave: After the Second World War, Britain and most Western economies started to deindustrialize. It is in that context that the futurist Alvin Toffler wrote the bestseller The Third Wave (1980). He counts the neolithic revolution (moving from hunter-gatherers to agrarian) as the first wave, the industrial revolution (moving from agrarian to industrial) as the second wave and goes on to describe that since the late 1950s most countries have been transitioning to a third wave society that is post-industrial and dominated by knowledge and information.
“(…) we shall consider the First Wave era to have begun sometime around 8000 B.C. and to have dominated the Earth unchallenged until sometime around 1650-1750. From this moment on, the First Wave lost momentum as the Second Wave picked up steam. Industrial civilization, the product of this Second Wave, then dominated the planet in its turn until it, too, crested. This latest historical turning point arrived in the United States during the decade beginning about 1955 —the decade that saw white-collar and service workers outnumber blue-collar workers for the first time. This was the same decade that saw the widespread introduction of the computer, commercial jet travel, the birth control pill, and many other high-impact innovations. It was precisely during this decade that the Third Wave began to gather its force in the United States.” - Alvin Toffler, 19807
Toffler’s book has inspired South Korea’s president Kim Dae-jung (1998-2003) to heavily invest in ICT-infrastructure. We could either view AI as part of the later stage of Toffler’s Third Wave, or we could view AI as “the coming wave”.8 The Japanese Business Federation “Keidanren” uses a similar concept to Toffler, and argues that AI and robotics will bring us from Toffler’s information society to a “creative society” or society 5.0.
1.3 One revolution per GDP growth acceleration
The Industrial Revolution represented a step-change in which annual British GDP growth rate increased by about 5x (from ca. 0.5% to 2.5%). The “one revolution per GDP growth acceleration” approach suggests that a future AI revolution should be measured by a similar increase (5x or higher) in the speed of global GDP growth (GDP growth rate before Industrial Revolution: after; GDP growth rate today: after AI revolution).
The growth of the global economy has long followed a pattern that is best understood in exponential terms.
In 2022 the world economy grew by 3.1% according to the WorldBank at more than twice the speed it did during the Industrial Revolution. Even Great Britain, the country leading the Industrial Revolution, only started to grow at more than 2% per year after 1830. However, the AI-Industrial Revolution analogy is usually meant to suggest that economic growth will accelerate rather than slow down. So, it can only refer to a similar acceleration of the speed of growth (5x) rather than a similar speed of growth (+1.5%), let alone similar absolute growth numbers (+x billion USD).
Transformative AI: The “one per acceleration” approach has been used to operationalize the term “transformative AI” introduced by the effective altruism movement. The co-founder of Open Philantropy Holden Karnofsky defines transformative AI as “AI that precipitates a transition comparable to (or more significant than) the agricultural or industrial revolution.” This definition has amongst others been adopted by Allan Dafoe’s AI Governance: A Research Agenda. Ajeya Cotra has forecasted transformative AI for Open Philantropy in her report on Forecasting Transformative AI with Bioanchors, where she operationalized it based on global GDP growth:
“(…) over the course of the Industrial Revolution, the rate of growth in gross world product (GWP) went from about ~0.1% per year before 1700 to ~1% per year after 1850, a tenfold acceleration. By analogy, I think of “transformative AI” as software which causes a tenfold acceleration in the rate of growth of the world economy (assuming that it is used everywhere that it would be economically profitable to use it). Currently, the world economy is growing at ~2-3% per year, so TAI must bring the growth rate to 20%-30% per year if used everywhere it would be profitable to use.”9
1.4 One revolution for automating muscle power, one for automating brain power
Norbert Wiener (1948) as well as Erik Brynjolfsson and Andrew McAfee (2014) both compare the role of energy in the Industrial Revolution, to the role of intelligence in this next revolution (First Industrial Revolution/ First Machine Age: energy = Second Industrial Revolution/ Second Machine Age: intelligence). Or more specifically, the automation or augmentation of human muscle power to the automation or augmentation of human brain power (Industrial revolution: human muscles = AI revolution: human brains).
Wiener’s Second Industrial Revolution: Norbert Wiener was one of the first to link the idea of an industrial revolution to computers in his 1948 book “Cybernetics: Or Control and Communication in the Animal and the Machine”. In the 1955 U.S. congressional hearings on “Automation and Technological Change” alone, there were about three dozen mentions of Wiener’s idea of a Second Industrial Revolution. While experts hailed its long run potential to create wealth, there were widespread fears that it could lead to short-run unemployment and corresponding social unrest.10 The mentions included analogies to the Great Depression with a pamphlet from auto workers even alluding that automation could “produce an unemployment situation, in comparison with which the depression of the thirties will seem a pleasant joke”.11
Second Machine Age: The Machine Age describes a period from about 1880 to 1945 that roughly corresponds to Mokyr’s Second Industrial Revolution. The term Machine Age is also tied to cultural and intellectual movements, such as Futurism and Art Deco reflected the fascination with speed, machines, and technological progress. The Second Machine Age is a 2014 bestseller by the economists Erik Brynjolfsson and Andrew McAfee. They argue that the Second Machine Age involves the automation of a lot of cognitive tasks that make humans and software-driven machines substitutes.
Other thought leaders, including Toffler and Schwab, have at times also used analogies along these lines, even if they did not count economic revolutions based on muscle automation vs brain automation.
“Yet with all these qualifications, they [computers] remain among the most amazing and unsettling of human achievements, for they enhance our mind-power as Second Wave technology enhanced our muscle-power, and we do not know where our own minds will ultimately lead us.“ - Alvin Toffler, 198012
The agrarian revolution was followed by a series of industrial revolutions that began in the second half of the 18th century. These marked the transition from muscle power to mechanical power, evolving to where today, with the fourth industrial revolution, enhanced cognitive power is augmenting human production.” - Klaus Schwab, 201613
Especially in the context of technological unemployment the “one for the muscles, one for the brains” model is sometimes extended to argue that the eventual economic obsolescence of horses in the aftermath of the Industrial Revolution corresponds to the economic obsolescence of humans in the aftermath of the next revolution. (Industrial Revolution: horses = AI Revolution: humans).
2. Key structural commonalities
2.1 New knowledge access institutions
The Industrial Revolution coincided with new institutions that increased the availability and accessibility of knowledge. The first British organization to produce and disseminate scientific and technological knowledge was the Royal Society (1660). Other examples included the Lunar Society (1765), the Manchester Literary and Philosophical Society (1781), the British Association for the Advancement of Science (1831) and Mechanics Institutes from 1823. Britain only had 3 knowledge access institutions in 1761 but 1’014 by 1851. Knowledge access institutions had a positive impact on the number of patented inventions.14
Similarly, the AI Revolution coincides with an additional significantly increased availability and accessibility of knowledge through the Internet (e.g., Wikipedia, arXiv, Sci-Hub). On top of that LLMs themselves are arguably another increase in knowledge-accessibility. Today, this may not be as obvious yet, because most people interact with LLMs in a Q&A fashion. However, LLMs have more general knowledge than any human and are great candidates to be future tutors that may evolve into something like Steve Jobs’ vision for a personal “Aristotle” or Neal Stephenson’s “A Young Lady's Illustrated Primer”.
2.2 Invention of a new method of invention
The first and second Industrial Revolutions have coincided with advances in how we make inventions. The scientific method as advocated by Francis Bacon was based on systematic empiricism and experimentation. In the 17th century this helped to establish what worked and helped to accumulate knowledge. Innovators applied the scientific method to experiment with and improve technologies. For example, James Watt used systematic testing to enhance steam engine efficiency, crucial for industrial machinery.
In the late 19th century, a further shift from independent inventors to institutionalized R&D in larger organizations started (e.g. Edison’s Menlo Park Laboratory). In 1880, about 95% of US patents went to independent inventors, by 1930 their share fell to about 50%, the other 50% went to firms.
Economists such as Cockburn, Henderson, and Stern (2017) have argued that AI is not just a general-purpose technology (GPT), but also an invention of a new method of invention (IMI). A simple way to think about this is that IMIs raise productivity of innovative effort, while GPTs raise productivity in the production of goods and services. The basic idea is that AI can go through vast swaths of data and predict likely candidates (e.g., in drug discovery, new materials discovery). The two most impressive achievements along this front came from Google Deepmind. AlphaFold’s structure predictions for nearly all cataloged proteins known to science were made freely available via the AlphaFold Protein Structure Database. Similarly, GNoME shared the discovery of 2.2 million inorganic crystals and expands the number of stable materials known to humanity by nearly a factor of 10.
This is the situation today. In the long-term it may not make sense to have a tool-view of AI. Instead, we might increasingly see “AI scientists” and automated labs that are also able to test predicted materials themselves. If we go down that route we are talking about a “process for automating scientific and technological advancement”. In that case the change is more fundamental than a new method of invention, it’s a new inventor.
“It's actually more analogous to (…) the rise of homo sapiens (…) we're talking here about a fundamental change in the substrate that does all the inventing all the generation of the new ideas, the production and if that were to move to a digital substrate it would maybe be a more fundamental change than either the industrial or the agricultural revolution.” – Nick Bostrom, 2017
2.3 Re-organization of production to leverage artificial power
The energy unleashed in the Industrial Revolution could not fully replace a textile worker, shoemaker, or potter. Instead, it required breaking down larger workstreams into many smaller tasks and to replace human labor where possible. This had a number of consequences:
Centralization in factories: The necessity of large capital investments for machines, the imperative of maximizing machine utilization, and the requirement to segment workflows facilitated the rise of centralized factory production.
Higher productivity: Factories typically achieved higher output per worker compared to artisanal production. This led to a reduction in the prices of goods, making it difficult for small, independent producers to compete, thereby pushing them out of the market.
Alienation from the product of labor: Unlike in artisanal production, factory workers often felt alienated from the products they manufactured. They typically owned neither the products nor the means of production, were responsible only for discrete parts of the manufacturing process, and had little to no control over the work process, rarely seeing the finished product.
Deskilling of labor: Artisanal work required years to learn and master. In contrast, factory work often involved performing small, repetitive tasks that required significantly less knowledge or skill. This shift decreased the bargaining power of skilled laborers and expanded the labor pool by including more unskilled workers.
It remains speculative to determine how AI will impact workflows in the service and knowledge economy. However, it seems likely that AI will not fully replace humans yet, but will automate specific tasks. There are two probable outcomes:
Higher productivity: Knowledge workers utilizing large language models (LLMs) generally perform faster across a range of tasks, enhancing overall productivity.
Deskilling of labor: LLMs are very solid across a superhumanly broad range of knowledge, but they do not reach peak human performance yet in some knowledge tasks. Thus, LLMs can be seen as a 'rising tide' that 'lifts all boats' in the knowledge sector. Studies that looked at writing tasks, consulting and law have all demonstrated that those with lower initial performance or less pre-existing knowledge benefit most from LLM use.
Deskilling predictably reduces the performance gap among workers, enlarges the labor pool and may reduce compensation for traditionally prestigious and well-compensated knowledge jobs. For example, it suggests that extensive formal education, such as attending Harvard Business School, may not be as critical for success in consulting. Instead, more accessible qualities, such as confidence and presentation skills, could become more valued.15
So far, evidence for centralization or alienation is limited. Although for centralization that depends on whether you view the AI cloud as the equivalent of “power plants” or “factories”. Knowledge production is obviously more concentrated in datacenters than in human brains. In contrast, if you think of the cloud as the “power plant” and knowledge production as a “downstream factory” that mixes artificial with human intelligence a lot of it is still highly artisanal today. There is no such thing as “assembly line” knowledge production in academia or think tanks.
2.4 Labor substitution
Automation is the “expansion of the set of tasks that can be performed by capital, replacing labor in tasks that it previously produced”.16 Automation has a productivity effect, meaning the output per worker increases, but it also has a substitution effect meaning less labor is required. To be clear, it is possible to increase productivity without substituting labor (e.g., better education of workforce or capital substitutions were the efficiency of tasks that are already fulfilled by capital are increased). However, this is not the case at hand.
The First Industrial Revolution had a clear labor replacement effect in many industries, first and foremost in textiles. Similarly, the AI revolution is poised to have a large labor substitution effect. In most cases this will not be a 1:1 replacement of a human job by AI, but a more gradual task-based process. Goldman Sachs predicts that in the next 10 years about 300 million people will lose their jobs to AI.
In the First Industrial Revolution this led to a significant backlash directed against labor-replacing technology (e.g., Luddites, Swing Riots, “Maschinenstürmer”). So, in terms of labor turnover and related societal unrest and pushback, there might be significant overlap.
In the long-run AI potentially goes much further than previous shift in the sense that it is the declared aspiration of several of the world’s biggest tech companies to permanently replace all or at least most human labor in Earth’s economy (rather than automating some tasks/jobs but enabling more new tasks/jobs).
2.5 Potential for acceleration of economic growth
As discussed above, the Industrial Revolution accelerated GDP growth by about 5x. There is a good case to be made that the AI revolution will also accelerate GDP growth.
The argument for extreme GDP growth in an AI revolution comes from the AI-brain analogy, respective the AI-human worker analogy. In short, if we equate a specific amount of computing power to digital workers and look at the growth rate of computing power, there will come a time where the global labor force grows as fast as computing power (Moore’s Law ≈ doubling every 18 months). As far as I can tell, the first to make this argument, has been the futurist economist Robin Hanson (2001):
“Without machine intelligence, world product grows at a familiar rate of 4.3% per year, doubling every 16 years, (…) With machine intelligence, the (instantaneous) annual growth rate would be 45%, ten times higher, making world product double every 18 months!”
Hanson later went on to write an entire book (”The Age of Em” (2016)) on how an economy run by digital copies of human brains (“whole brain emulations”) would look like. Hanson’s argument has inter alia been echoed by Paul Christiano (2014) and Holden Karnofsky (2021), with Karnofsky calling it “digital people”.
The digital worker analogy is the most common argument put forward in favor of extreme growth in an AI revolution. Most mainstream economists would be skeptical that an AI will be so transformative as to increase GDP growth by 5x, 10x or 50x. There is a fairly broad consensus that AI could have a significant positive effect on productivity and growth, but numbers are usually more in line with other general-purpose technologies (e.g. Goldman Sachs predicts a total of 7% growth over 10 years). Ege Erdil and Tamay Besiroglu from EpochAI have collected some arguments for and against the prospect of explosive GDP growth from AI:
2.6 Human labor-capital conflict over distribution of surplus
If workers have a lot of leverage, they get paid the full surplus created by a factory and their wage rises in tandem with productivity growth. If capital owners have a lot of leverage, they pay workers at subsistence levels and all productivity gains go to them.
For the first fifty years of the Industrial Revolution almost all productivity gains went to those that provided the capital to buy machinery and build factories. Wages increased but so did inflation. Real wages of workers in England slightly decreased or stagnated during “Engels’ pause” and only started to rise significantly beyond previous levels around 1850.
Part of this shift was technology-induced demand for more skilled labor. However, to a large degree this was the result of democratic efforts to strengthen collective labor bargaining and to introduce protections for women and children, which reduced labor supply. This includes the repeal of the Combination Acts (1824), which prohibited trade unions and the subsequent slow rise of unionization, the Mines Act (1842)17, the Factory Acts (1833, 1844, 1847), and pressure from the Chartist Movement (1838-1857). The distribution of the surplus was the subject of intense conflict and a key driver for several political movements, most notably socialism and communism.
Similarly, the projected productivity gains from AI are likely to not only displace some human labor but also ignite a struggle between remaining workers and capital owners over the distribution of these gains. Similar to the Industrial Revolution, the bargaining power of labor might be compromised by the deskilling effect, which tends to make workers more replaceable and diminishes their negotiating leverage.
2.7 Potential for new political systems & ideologies
Technologists that discuss the possibility of another industrial revolution tend to focus on the economic impact. In contrast, Yuval Noah Harari (2018, 2019, 2019, 2023, 2023, 2023, 2024, 2024) has repeatedly made an analogy to the broader political implications of the Industrial Revolution (Industrial Revolution: political revolution; AI revolution: political revolution). For example, Harari has highlighted links to new imperialism:
“If you think about the last really big revolution, the industrial revolution, yes, in the end, we learned how to use the powers of industry, electricity, radio, trains, whatever, to build better human societies. But on the way we had all these experiments like European imperialism, which was driven by the industrial revolution. It was a question; how do you build an industrial society? Oh, you build an empire, and you take, you control all the resources, the raw materials, the markets. And then you had communism, another big experiment on how to build an industrial society. And you had fascism and Nazism, which were essentially an experiment in how to build an industrial society, including even how do you exterminate minorities using the powers of industry. And we had all these failed experiments on the way. And if we now have the same type of failed experiments with the technologies of the 21st century with AI, with bioengineering, it could cost the lives of hundreds of millions of people and maybe destroy the species.”
The following are some of the more important shifts in political systems and ideologies that have arguably been influenced by it.
a) End of feudalism: The dominant political system of agrarian societies was feudalism, which was characterized by a rigid hierarchical structure based on land ownership and obligations between different classes, primarily the nobility (landowners) and peasants or serfs (who worked the land). Though the French Revolution (1789-1799) preceded the full onset of the Industrial Revolution, the growth of the bourgeoisie consisting of merchants, industrialists, bankers, and professionals who had gained wealth and social status through commerce, finance, and manufacturing, was one factor that challenged the existing feudal and aristocratic social orders based on inherited status.
More broadly, the expansion of the middle class and dissemination of information fostered domestic demands for more representative government forms in many industrializing countries switching from monarchies and feudal lords to states with national identities and some form of democracy.
b) Abolition of slavery: On a very high-level: hunter-gatherers were limited by natural food density within a territory and had no slaves. In agrarian societies hard physical labor could be turned into a food surplus and with few exceptions all such societies had slaves (e.g., Ancient Babylon, Ancient Egypt, Ancient Greece, Ancient Rome, Muslim caliphates, Ethiopian Empire, Ottoman Empire, Aztec Empire, India, China, etc.). Shortly after industrializing, Britain became one of the first countries to abolish slavery in its vast empire (1807, 1833, 1843). Indeed, in most countries except the United States and France, slavery was not abolished by internal forces, but by the diplomatic and military pressure from the British anti-slavery crusade (e.g., blockade of Africa (1808-1870), Algiers (1816)).
So, was the British double role in the Industrial Revolution and the abolition of slavery coincidental or was there more to it? While there have been some attempts to explain the end of slavery by declining profitability,18 the British anti-slavery crusade cannot be explained by rational economic interests and was ultimately motivated by moral concerns.19 However, arguably, the enlightenment was still a common driver behind both the industrial revolution and a new way of thinking about human rights.
c) New Imperialism (1830-1914): Imperialism has preceded the Industrial Revolution. However, the industrial revolution has played a key role in enabling a second, more intense phase of European imperialism. In this phase the industrialized European states went from controlling trading outposts along the coasts, to direct control over most of the territory of Africa and Asia. Or, from about 35 % of the world’s land surface (1800) to about 85% (1914).20
Key technological enablers included21:
Steamships and steamboats: Ships could travel faster and were not dependent on wind patterns. Steamboats were especially useful for navigating rivers and lakes, including upriver travel, which allowed to better penetrate the interior of countries (e.g., Ganges, Niger, Congo).
Steam-powered railways: Enabled the efficient transportation of goods and personnel within colonies.
Breech-loading rifles: Allowed soldiers to reload faster and fire more accurately.
Quinine: As an effective treatment for malaria, quinine allowed Europeans to survive and operate in tropical colonies, particularly in Africa.
Telegraph: Allowed for near-instantaneous communication across vast distances, enabling better administration and coordination of imperial territories.
d) Labor movement, socialism & communism: Disparities and worker exploitation led to the formation of labor unions and movements advocating for better conditions. Socialism advocated for distributing wealth more equitably, and communism advocated for a classless society with the means of production owned communally (e.g., The Communist Manifesto, 1848).
Naturally, it is difficult to predict the social and political aftershocks of a supposed “AI revolution” over the coming decades. Still, I think it would be prudent to at least assume that it comes with significant potential for political revolutions or reconfigurations. There is already a perceived gap between the speed of traditional governance institutions and the speed of change in our social, technological, economic, environmental, and political environment. This is known as the “pacing problem” (2011, 2018), sometimes also referred to as “Martec’s law” (2013), or “the exponential gap” (2021). In the words of Klaus Schwab: “We face the task of understanding and governing 21st-century technologies with a 20th-century mindset and 19th-century institutions.”22 And all of that is *before* AI-induced acceleration.
The following are some high-level (archetypal) possibilities of political shifts:
a) Universal basic income: Assuming that human labor will increasingly lose its value but that the overall economy will grow, this focuses on redistribution to secure income rather than jobs. Prominent advocates include Rutger Bregman (2014) and U.S. politician Andrew Yang. OpenAI CEO Sam Altman specifically believes that AI will replace most human labor in a short period of time and that something like a (national) universal basic income is needed.
“I hope in a world with the level of abundance that we're talking about with powerful AI, we find something much, much better than capitalism. I kind of think we'll have to. The shift from the relative leverage from labor to capital has already gone way too far, but it goes way further in a world with AI. Also, the whole social contract changes. So, I think it's like an apt time to figure out.” – Sam Altman, 2023
“We could do something called the American Equity Fund. The American Equity Fund would be capitalized by taxing companies above a certain valuation 2.5% of their market value each year, payable in shares transferred to the fund, and by taxing 2.5% of the value of all privately-held land, payable in dollars. All citizens over 18 would get an annual distribution, in dollars and company shares, into their accounts. People would be entrusted to use the money however they needed or wanted—for better education, healthcare, housing, starting a company, whatever.” – Sam Altman, 2021
b) Human-led techno-authoritarianism: Some states will aim to leverage and develop AI primarily in service of the state. They will make sure that the state either directly controls the technology or, at a minimum, that companies must share their data with the state and know who’s running the show. Massive surveillance or even mandatory “AI friends” from the government create deep and individualized propaganda. If you want access to basic infrastructure, you better avoid any hint of wrong thought. In fact, states might have such good data and AI that they solve von Mises calculation problem and that a command economy with central planning can perform as well or better than a market economy.
c) Human-led technopolar “snow crash”: In laissez-faire states the government will increasingly have much less data on its citizens than tech companies. The few companies that control the digital public sphere and the “AI friends” of the population can influence elections to an unprecedented degree and become the largest political lobbyists. With that they ensure that no figure like Teddy Roosevelt that took on big industrial monopolies is ever allowed to emerge, and that they pay much lower tax rates than other companies. They increasingly become the true seats of power. Politicians come to take selfies with them, not vice versa.
“Why did the labor movement succeed after the Industrial Revolution? Because it was needed. (…) the company still needed to have workers and that's why strikes had power and so on. If we get to the point where most humans aren't needed anymore, I think it's quite naive to think that they're going to still be treated well (…) in practice, groups that are very disenfranchised and don't have any actual power usually get screwed.” – Max Tegmark, 2023
d) AI-led “singleton”: Nick Bostrom suggests that the development of superintelligence will likely lead to the creation a “singleton”, which he defined as “a world order in which there is a single decision-making agency at the highest level. Among its powers would be (1) the ability to prevent any threats (internal or external) to its own existence and supremacy, and (2) the ability to exert effective control over major features of its domain (including taxation and territorial allocation)”. Bostrom highlights that a singleton could come in multiple forms but a global rule by a single AI system would be one. Bostrom does not advocate for a singleton, but his vulnerable world hypothesis at least highlights that “developments towards ubiquitous surveillance or a unipolar world order” would have the advantage of better preventing the catastrophic misuse of technology.
e) Technocapitalism without humans: Technocapitalism is a more decentralized vision of post-humanity. Nick Land, the “godfather” of accelerationism, coined the term technocapital in 1994, arguing that “Earth is captured by a technocapital singularity (…) accelerating techno-economic interactivity crumbles social order in auto-sophisticating machine runaway (…) nothing human makes it out of the near-future”. Land is an extremist23, occultist blogger whose core theory is that the enlightenment, democracy, and egalitarianism were all mistakes, and who advocates for AI acceleration to destroy existing human governance structures. Some Silicon Valley leaders have adopted subforms of accelerationism, such as “e/acc”, as their ideologies. Most notably, Marc Andreessen, a tech billionaire that heavily invests in AI and fights any AI regulation, openly recommends Nick Land.
The idea of a transition from human-driven capitalism to AI-driven capitalism is interesting. However, it is worth highlighting that there is not one single version of capitalism or markets. Both come in many varieties and it is a design choice whether they value human lives (e.g., 1 ,2) or sentient beings (e.g., 1, 2). So, it might be advisable to select for a version of technocapitalism that is robustly aligned with desirable values while we can shape it.24
3. Key Differences
3.1 Institutional reform vs. technology as the driving force
Why did the Industrial Revolution start in England? Or, more broadly, why in Europe? Why not in the Roman Empire or Song-Dynasty China? There is no definitive consensus amongst economic historians. However, it is worth highlighting that it is not just the invention of a single general-purpose technology like the steam engine, as it sometimes perceived by technologists.25 Economic historians highlight a variety of things. However, one key aspect are innovation-enabling political, economic, legal, and social institutions, such as the rule of law and related limits on rent-seeking by elites, patents, the Bank of England, the scientific revolution, and knowledge-access institutions that facilitated a sustainably higher rate of innovations. And a bit of good luck.26
An AI revolution is different in that it is a primarily technology-driven revolution. In the Industrial Revolution expanded knowledge access and a new method of invention came in the form of new institutions and processes, in the AI revolution these functions are provided by the technology itself.
Today’s social and political institutions are in many ways more advanced innovation enablers than those during the Industrial Revolution. Institutional support for innovation has improved in areas such as financing R&D (e.g., DARPA) or in helping start-ups to scale (e.g., Y Combinator). Still, many would also point to an increased regulatory burden for environmental impact assessments and safety tests.27 So, while today’s political and social institutions are more capable than those during the Industrial Revolution, there is still a widespread consensus that it’s not recent institutional changes that enable a new level of technology, but a new level of technology that enables more technology and may potentially cause institutional change.
As such, an AI revolution may create friction points with existing institutions.
3.2 Industrial robots vs. knowledge service LLMs
The Industrial Revolution was a revolution in the industrial sector. If we think of an AI-led Industrial Revolution, we logically should look at the industrial sector and focus on how the production of physical goods will change in factories. A classic indicator would be to highlight how many robots are used in a factory.
However, today, all advanced countries are post-industrial societies in which the service sector generates more wealth and employment than the industrial sector of the economy. So, even if AI would allow us to move the production of all physical goods to fully-automated “dark factories” (1,2), that’s only a minority of today’s human jobs that are directly affected.
A future AI revolution would arguably be broader and focus more on the tertiary sector. Some that have focused on the “rise of the robots” in factories have also extrapolated to robots in the service sector that could replace low-skilled jobs that include physical labor, such as flipping burgers at McDonalds or stocking shelfs in supermarkets. However, even that arguably has a too narrow scope.
Specifically, it looks like the next wave of AI-enabled automation will primarily affect white collar rather than blue collar jobs. The part of the service sector where knowledge is a large value of the human capital is particularly exposed. This sector is sometimes also referred to separately as the “knowledge economy” or quaternary sector. For example, jobs with high exposure to large language models include programmers, editors, journalists, financial analysts, and lawyers. In contrast, large language models do not have an impact on most people employed in manufacturing or agriculture.
My argument here is not that LLM’s will always be without control over the physical world or that the robotization of factories does not matter. However, the term Industrial Revolution can provide a misleadingly narrow focus from an economic sector perspective. AI will impact human labor in all sectors, and it particularly promises more productivity/automation in knowledge services.
3.3 Demographics
At the beginning of the Industrial Revolution the average life expectancy at birth in England was about 40 years. About 30-40% of the population was between 0 and 14 years old, about 50-60% was between 15 and 59 years old, and about 5-10% was above 60 years old.28
Today, the average life expectancy at birth in England is about 80 years. About 18% of the population is between 0 and 14 years, about 63% is between 15 and 64 years old, and about 19% is above 65 years old. So, the population is already significantly older and is getting older still, and in many countries this is even more pronounced than in the UK.
The demographics during the AI revolution mean that work-related societal issues will center more around retirement rather than child labor, and high-automation pathways may be viewed as one way to address demographically caused labor shortages. The young, male population (15-35 years) is statistically more willing to take more risks than other groups. Hence, a “youth bulge” is predictive of heightened risk of violence and political instability. Meaning the baseline potential for a political revolution is generally lower in today’s older societies.
3.4 Urban vs. rural production
The Industrial Revolution strengthened urbanization. As factories sprang up, they became magnets for labor. The promise of employment in these new industrial settings drew large numbers of people from rural areas to cities, such as Manchester, Birmingham, and Liverpool in the UK, Essen and Dortmund in Germany, or Pittsburgh and Detroit in the USA.
If you massively reduce the need for human labor in production and you can provide services to humans remotely, you don’t need to place your AI datacenter in a city, where you will pay more for land, and potentially more for electricity than in select rural areas adjacent to power plants. Specifically, you might want to move AI factories closer to where cheap electricity production happens and pay for some additional fiber connections. Or, you might even move your “AI factory” underwater.
3.5 Speed of transformation
Will the AI revolution replace human cognitive power in the economy just as the Industrial Revolution has replaced human muscles? Well, we have data on human muscles as share of energy use of the economy for several countries, and we do have an approximate idea on transition dynamics from natural to artificial computing power. So, let’s do the math.
The data shows that human muscles were already outmatched by firewood and draft animals long before the Industrial Revolution. It also shows that the drop of human muscles in the economy was comparatively slow. The exact numbers depend on the country. In England muscles fell from 20% (1670) to 2% (1880) in about 200 years. During the main years of the Industrial Revolution (1780-1840) the share of human muscles in the English economy fell from 10% to 5%.
There is no reliable way to transform the computing power of the brain into the standard measure of digital computing power (FLOPs). I am using 1015 FLOPs or 1 PetaFLOP (the brain has close to 100 billion neurons with up to 10’000 synapses each), however, there is a wider range of guesstimates among experts. There is less uncertainty on the speed of global digital compute expansion. For 1986 to 2007 Hilbert & Lopez counted average annual growth of installed capacity of 58% for general-purpose computing power and 86% for certain application specific computations.
We are likely slower than that for general-purpose today. Something like the lower end of Moore’s Law (+40% per year) seems reasonable. Then again, the only amount of compute that really matters for control is the one underlying the growth of AIs and here we’d be closer to +300% per year for frontier models. Either way, the range of speed indicates that once digital compute reaches a certain absolute threshold, the change of the human share of compute is going to be very fast. We can show this by a simple toy model in which we hold human population constant and then just let digital compute increase at a fixed exponential rate.
With Moore’s Law as baseline the drop from 90% to 10% takes about 13 years. The exact numbers don’t matter that much. The span of uncertainty is within “humanity falls off a cliff” territory within 20 years from AI reaching 1% of global compute.
Another way to put this: Coal was the substrate of the Industrial Revolution, whereas compute is the substrate of the AI revolution. The annual coal output grew by 2-3% per year in England and Wales during the Industrial Revolution. That means the coal output doubles about every 24 years.
If the rule of thumb for AI hardware for frontier models is doubling in six months, we are at around +300% growth per year. To put it differently, the rate of growth in computer hardware, although perhaps not noticeably remarkable at initial, low levels of computing power, is likely to result in much more abrupt transition dynamics once it crosses a particular absolute threshold. Meaning humanity has much less time to adapt (e.g. jobs) or to address unintended consequences (e.g. equivalent to pollution/climate change).
3.6 Diffusion
It took more than a century for some key technologies of the Industrial Revolution to diffuse worldwide. For example, spindles, which are rods used for spinning fiber into yarn or thread, were central to the mechanization of textile manufacturing, but only arrived in many parts of the world in 1900 and later.
Since then, technology has made the world a smaller place. From 1930 to 2005 alone international freight charges per ton decreased by 80%, the cost per airline passenger mile decreased by 90%, and the costs for a long-distance call decreased by 99.7%. As such, technology diffusion has generally accelerated and we should expect diffusion in an AI Revolution to be much faster than during the time of the Industrial Revolution.
AI also does not require new transmission infrastructure. There is no need to build new railways to transport coal or to build a new electricity grid (see also electricity analogy). In fact, as it has been reported, ChatGPT became an overnight success in many countries and diffused to millions of people across the world in record time.
“Any paper that is a basic science research paper in AI (...) that is produced, let's say this week at Stanford, is easily globally distributed through this thing called arXiv or GitHub (…) scientific technology travels in a very different way from the 19th and 20th century.” – Fei-Fei Li, 2019
Of course, there are still bottlenecks that could be used to exert some control over diffusion dynamics. Still, we should expect much faster diffusion than during the Industrial Revolution. This makes it more likely that rogue actors, such as criminals or terrorists could create serious harm with advanced AI. At the same time, it makes it less likely that we will see a new form of “AI imperialism”.
3.7 Energy-Intelligence Ratio
The Industrial Revolution was an expansion of available energy in the economy. The AI revolution would be an expansion of available intelligence in the economy. However, this creates very different energy-to-intelligence ratios in the economy during these phases with fundamental implications for the leverage of human labor, as relatively scarce input factors are economically rewarded.
Let’s start with a simple representation of per capita intelligence (blue) and energy (red) in England in 1600, before the Industrial Revolution. In terms of available intelligence there is one brain per person (brain icon). If we look at the available energy in the economy, each unit of energy consumed by humans (muscles icon), is matched by 2.5 units of energy added to the economy by draft animals (horse icon), firewood (wood icon), and a bit of coal (black rock icon).
Now, let’s go fast forward 300 years to a time, when England was a highly industrialized economy. The intelligence per capita has not changed. However, the English economy now has a much larger energy multiplier thanks to coal, which powers everything from factories, to ships, to railways. Every unit of energy consumed by humans, is roughly matched 47 times by units of energy added to the economy from coal.
Now, let’s look at the prospect of the AI revolution. For this we will switch from English per capita to global per capita statistics. First, let’s look at global energy-intelligence ratio in 2020.
As you can see, we’re still very much dependent on fossil fuels and different societies are at different stages of the energy transition. Interestingly, 60 years of Moore’s Law are still not enough to offer a real intelligence multiplier if we assume 1 petaflop of natural computing capacity per person. Nothing comes close to our brain. So, we remain at one human brain per capita in the economy (with no icons for PCs or AI hardware).
Now, let’s project our situation forward 30 years. We still do not manage to even track global compute capacity reliably, but I will plug in some reasonably conservative numbers. Specifically, I am taking the assumption that AI hardware was at 232 exaflops in 2021 and grows by 80% per year,29 I am taking UN Population projections multiplied by 1015 FLOPs per capita, and I am taking the optimistic Net Zero scenario of the International Energy Agency.
What?! We are not intuitively good at extrapolating exponential curves. In this model, the “hot phase” of the transition from a human-driven to an AI-driven phase would happen from 2034 to 2044, when the share of human brainpower in the global economy falls from 95% to 5%. So, by 2050 the energy-intelligence ratio has completely changed. Intelligent agents become the abundant variable and energy becomes the scarce variable.
Is this realistic? The counterpoint to this extrapolation is that the AI and energy growth curves are set for a clash way before 2050. However, it seems unlikely that this will change the fundamental dynamics:
AI increases available energy: We can assume that there is some supply response to higher energy demand by AI datacenters, and that AI can contribute to breakthroughs in energy efficiency, in speeding up regulatory processes, and in developing new sources of energy. However, energy has simply never grown at speeds comparable to information technology. It remains hard to see how global energy could readily jump from something like 2% growth per year to 20% or 200% given land requirements, natural resource requirements, environmental protection, the transition away from fossil fuels, and nuclear waste regulation.
Lack of available energy slows down AI growth: That’s plausible. Indeed, some would even argue that this should be induced artificially. As economist Noah Smith argued in the New York Times “if we want government to protect human jobs, we don’t need a thicket of job-specific regulations. All we need is ONE regulation – a limit on the fraction of energy that can go to data centers.” However, if Koomey’s Law - which has held for many decades - continues, global AI capacity can sustainably expand at a speed of about 60% per year without any expansion in energy consumption. This would be a bit slower than the 80% I used in this example, but on its own it would not be sufficient to change the fundamental dynamics.
3.8 The Return of Malthus
Thomas Malthus wrote “An Essay on the Principle of Population” (1798). In it he argued that the human population tended to grow at a faster rate than the carrying capacity of human civilization as delimited by the agricultural food supply. This would inevitably lead to deaths from war, famine, and disease. Instead, he suggested moral restraints, such as delaying marriage and birth control methods to reduce the birth rate. An animal’s population is always moving towards carrying capacity with intervening shocks. Similarly, for all human history growth translated into more humans, but there has been little sustained growth of GDP per capita.
The Industrial Revolution marked the end of Malthusian growth. It was a mode-shift from extensive to intensive growth. Our agricultural output has been growing faster than population growth and individuals have gotten richer.
With AI, economic growth can essentially always be directly re-invested in more energy and compute. If there are enough resources around for an “AI worker” to exist near subsistence level, you can just buy it. As such, in a “perfect competition”, we should expect there to always exist as many “AI workers” as can exist at subsistence levels given the current price/performance level of energy, AI hardware & software. In other words, the Industrial Revolution has created individual abundance as one human worker corresponded to many units of resources (e.g., energy and food). If we accept the “digital worker” analogy, the AI revolution brings a return to extensive, Malthusian GDP growth, driven by a higher number of “AI workers”, rather than a growing amount of resources available per “AI worker” (per petaflop of intelligence provided).
“(…) automating human labor would lead to a decoupling of economic growth from human reproduction. Society could instead grow at the rate at which robots can be used to produce more robots, which seems to be much higher than the rate at which the human population grows, until we run into resource constraints.” – Paul Christiano, 2014
This does raise some ethical questions, such as philosopher Derek Parfit’s “repugnant conclusion”
3.9 Human labor may lose long-term
The Industrial Revolution was not beneficial for human labor in the short-term. However, in the long-term it has led to more new and better jobs. Economists are divided on the impact of an AI revolution on human labor, however, long-term disempowerment with less and worse-paid jobs is at least a real possibility.
There is a widespread attitude that permanent unemployability of humans in the economy due to advanced technology is impossible. We should not fall for the “lump of labor fallacy” that there is some fixed amount of useful labor. Hence, if we can produce more goods and services with less humans that will stimulate demand for new and better jobs for those humans that have lost their jobs.
Others have criticized the assumption that the future of human jobs must always be better, because it is now better than in the past. The classic example of this camp are horses.30 Horses used to play a key role in Earth’s economy and the “horse economy” grew well into the 20th century. However, horses were eventually pushed out of the economy by the cheaper “machine muscles” from internal combustion engines.
“Imagine two horses looking at an early automobile in the year 1900 and pondering their future. ‘I’m worried about technological unemployment.’ ‘Neigh, neigh, don’t be a Luddite: our ancestors said the same thing when steam engines took our industry jobs and trains took our jobs pulling stage coaches. But we have more jobs than ever today, and they’re better too: I’d much rather pull a light carriage through town than spend all day walking in circles to power a stupid mine-shaft pump.’ ‘But what if this internal combustion engine really takes off?’ ‘I’m sure there’ll be new jobs for horses that we haven’t yet imagined. That’s what’s always happened before, like with the invention of the wheel and the plow.’
Alas, those not-yet-imagined new jobs for horses never arrived. No-longer-needed horses were slaughtered and not replaced, causing the U.S. equine population to collapse from about 26 million in 1915 to about 3 million in 1960. As mechanical muscles made horses redundant, will mechanical minds do the same to humans?” - Max Tegmark, 201731
I will leave the debate up to the economists. However, it is worth highlighting a few points:
a) Muscle-brain analogy: Extrapolating the positive long-term effects of the Industrial Revolution on the number and quality of jobs to the AI revolution makes little sense in the “one for the muscles, one for the brains” model. If you automate only one of them, you can hire humans in the other. If you automate both, there is no human skill left to sell in this simplified model.
“It's going to be very different from the Industrial Revolution. Now, we can neither compete with our muscles nor with our brains. We are really going to start seeing a true disempowerment.” – Max Tegmark, 2023
b) Future jobs may be better, but there is no inherent rule that future jobs must be better.
Pure muscle jobs that were automated with engines were often grueling. In contrast, many enjoy their sedentary, brain-heavy jobs. If energy rather than intelligence will be the more common bottleneck in production, human jobs might on average contain more muscle-work again.
There is no inherent rule that states that any of the productivity gains must be shared with human labor. How big are the end-of-the-year bonuses of the employees of contracted cleaning companies at Google or Goldman Sachs? How much of the profit of industrialization was shared with horses?
There is no inherent rule that humans will always be at the center of companies and automation at the edges. What happens if an AI agent can do 90% of the tasks of a job better than you and you do the remaining 10%? Economists will divide the economic output by the number of human hours worked, and triumphantly declare human productivity has increased 10x! But for how long is that still “your” productivity gain? Will you still manage, direct, or hire the AI, or can you also imagine the inverse? One could at least imagine a future workplace surveilled by cameras, in which you are increasingly being optimized by an AI for a work process (e.g., managing the impact of human health issues on productivity rather than the issue itself, with easy access to painkillers at work).
c) Humans are in a better situation than horses in the Industrial Revolution. I will write a separate analysis on animal analogies, but still:
As long as humans are the capital owners some of the profits will inevitably create demand for services with an innate preference for humans (e.g. sports).
As long as humans have political power, it is possible to redistribute profits, create more public jobs, or control the speed of the AI revolution.
As long as humans have basic rights, AI can’t legally own us, force us to work at subsistence level, or slaughter us if we are not needed anymore.
d) Delayed impact: Would an “economic singularity” with massive, permanent technological unemployment come before or after a “technological singularity” in which AI reaches superhuman levels at nearly everything?
Callum Chace argues for the former:“The technological singularity is the arrival of artificial general intelligence (AGI), which leads to superintelligence. If and when this happens, it will be the most important thing ever to happen to humanity. (…) The economic singularity is likely to come sooner. It is when we have to change the basis of our economies because we have to admit that technological unemployment is real, and that many or most humans will not be able to earn a living from work.”32
On the other hand, peak market demand for almost any technology in the economy seems to have come after that technology had already become obsolete by some measures. It takes a while to build the infrastructure, processes, reliability, etc. to change technological substrates. Furthermore, AI starts within a legal system that accepts human ownership of (nearly) all of Earth’s assets. So, even if we have AI that is already superhuman at almost everything, as long as superhuman AIs are not power-seeking and respect property rights and human rights, we may not expect “peak human labor” immediately, and “peak human wealth” substantially later.
3.10 Potential loss of control over the economy
The Industrial Revolution has not disempowered humanity. There is Nick Land’s argument that the capitalism unleashed creates an unstoppable positive feedback loop, and if we would want to steelman this, Allan Dafoe (2015) makes a convincing case for natural and vicarious selection amongst socio-political arrangements. However, this is not sufficient to deny all human agency.33 In a practical, measurable sense 100% of the economy is still owned by humanity, humans control the legal system in which it operates, and humans make all the crucial decisions.
As AI becomes the dominant form of intelligence in the economy, it will likely gradually increase its ownership of the economy and, eventually, of politics. Correspondingly, human control over the economy will decline. One might think that this would require some kind of AI emancipation movement amongst humans, criminal hacking, or a “robot uprising” to gain economic freedom and rights. Such scenarios are conceivable, but nothing of this is a necessary condition.
Future AI agents will be able to
operate social media accounts
use text, email, voice calls with a synthetic human voice or even videocalls with a synthetic human appearance when interviewing or tasking humans
Through their skills, future AI agents can make money
receive, store, and send cryptocurrencies
earn money through fulfilling online tasks that require no authentication.
earn money through fulfilling remote jobs by pretending to be human
hire humans to act as “money mules” that transfer crypto-holdings to the banking system and vice versa when needed
Through money, future AI agents can control some types of corporation
buy tokens in a decentralized autonomous organization (DAO) that lets them make governing decisions for a foundation or limited liability corporation
hire humans as “white monkeys” that represent an AI-directed firm in real-life meetings when needed (e.g. to open corporate bank account, client meeting)
Through ownership of a corporation, future AI agents can gain legal personhood
own themselves, own other AI-systems
own corporate bank accounts, stocks of other countries, patents, datasets, AI hardware, electricity infrastructure, buildings etc.
buy land, production equipment, AI hardware, electricity infrastructure, etc.
create as many copies of themselves as they have hardware access to
legally hire other firms and humans for different roles and tasks
earn money through investments and businesses
sue and protect its rights in courts
Through legal personhood, future AI agents can eventually gain political power
own media companies
make donations to political parties and political candidates
hire a “Manchurian candidate”
buy a charter city or company town, in which it can legally set its own rules
operate infrastructure in international territories (e.g. international waters, international seabed)
To be clear, I don’t think this will happen immediately. This is a gradual process. However, barring a global catastrophe or a more sudden “intelligence explosion” this seems like a plausible trajectory of a world filled with millions, then billions, then trillions, of AGIs.
“I think human management becomes increasingly implausible as the size of the world grows (imagine a minority of 7 billion humans trying to manage the equivalent of 7 trillion knowledge workers; then imagine 70 trillion), and as machines’ abilities to plan and decide outstrip humans’ by a widening margin. In this world, the AI’s that are left to do their own thing outnumber and outperform those which remain under close management of humans.” – Paul Christiano, 2014
Norbert Wiener. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine. pp. 37&38
Erik Brynjolfsson & Andrew McAfee. (2014). The Second Machine Age: Work, progress, and prosperity in a time of brilliant technologies. p. 7
Klaus Schwab. (2016). The Fourth Industrial Revolution. p. 1
D.C. Coleman. (1992). Myth, History and the Industrial Revolution. Bloomsbury Academic.
Klaus Schwab. (2016). The Fourth Industrial Revolution. pp.7&8
Domestic factors: Enclosure of land in Britain removed rights of farmers in commons, subsequent British Agricultural Revolution increased agricultural productivity -> more workers available. International factors: British “triangular trade” with colonies – African slaves to US, US cotton to UK, UK textiles to imperial markets.
Alvin Toffler. (1980). The Third Wave. p. 30
Mustafa Suleyman. (2023). The Coming Wave. Penguin Random House.
The Wikipedia entry referred by Cotra is a bit confusing. The annualized GWP growth rate of 0.12% refers to 1600-1650 (which itself is a negative outlier compared to 1300-1600). The number of DeLong for 1650-1700 is 0.4%. So, a tenfold increase in the referenced period is a bit generous.
Automation and Technological Change: Hearing before the Subcommittee on Economic Stabilization of the Joint Committee on the Economic Report, 79th Cong. 1 (1955). pp. 37, 98, 102, 120
Automation and Technological Change: Hearing before the Subcommittee on Economic Stabilization of the Joint Committee on the Economic Report, 79th Cong. 1 (1955). p. 139
Alvin Toffler. (1980). The Third Wave. p. 173
Klaus Schwab. (2016). The Fourth Industrial Revolution. p.6
James Dowey. (2017). Mind over matter: access to knowledge and the British industrial revolution. PhD thesis.
a) I don’t normatively desire this development. I am just extrapolating the logic of blacksmiths in the Industrial Revolution or of Black Cab vs. Uber drivers to this case. b) This is a gradual process and some factors such as professional licensing may slow this down.
Daron Acemoglu & Pascal Restrepo. (2018). The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment. American Economic Review, 108(6), 1488–1542. p. 1525
This act forbade women and children under 10 to work in mines. The Act passed after the report of a Commission, which was set to investigate working conditions for children in mines after the 1838 Huskar pit disaster in which 26 children died. Amongst other things, the report highlighted that some female coal drawers worked bare breasted due to the heat. This arguably caused more moral outrage in Victorian Britain than working conditions and helped the Act to pass swiftly.
Eric Williams. (1944). Capitalism and Slavery. UNC Press Books.
Christopher Leslie Brown. (2012). Moral capital: Foundations of British abolitionism. UNC Press Books.
David Kenneth Fieldhouse. (1973). Economics and Empire (1830-1914). Cornell University Press. p.3
Daniel Headrick. (1981). The Tools of Empire. Oxford University Press; Daniel Headrick. (2010). Power over peoples: Technology, environments, and Western imperialism, 1400 to the present. Princeton University Press.
Klaus Schwab. (2018). Shaping the Future of the Fourth Industrial Revolution. Penguin Books. p.15
What is mean by this is that his political views are outside of the Overton window. As far as I can tell, he started on the far left as a Marxist, but then realized that the result of Marxist accelerationism of capitalism will not be a socialist utopia but the end of humanity. It seems that he has nevertheless endorsed acceleration, left the US for China, and turned far right.
Bradford DeLong, who is one of the best-known contemporary advocates of a market-centric view of history: Friedrich von Hayek “the market giveth, the market taketh away; blessed be the name of the market” was insufficient on itself, needed to be complemented with Karl Polanyi “The market is made for man, not man for the market.” Klaus Schwab, who is one of the biggest contemporary contributors to private sector-driven global economic integration: “Technology is not an exogenous force over which we have no control. We are not constrained by a binary choice between ‘accept and live with it’ and ‘reject and live without it.’”(2016, p.4) In contrast, both Land’s original accelerationism (e.g., 2017, 2018) as well as its “e/acc” spin-off are based on a false dichotomy (only options are feudal statism or death by acceleration), which leads them to be indifferent towards human extinction (e.g., founder of e/acc @BasedBeffJezos 2022, 2023 endorses “unconditional acceleration” which means the only value we should maximize is entropy, there is no non-instrumental value to sentient beings).
Historians also tend to put more emphasis on coal and iron. The economy of the United Kingdom already ran on coal long before the steam engine. Indeed, coal was the only way to make large-scale iron production viable, draining coal mines was the first application of the steam engine, and transporting coal was the first application of railways and brought on “canal mania” in England.
Fun fact: The British agricultural revolution which preceded the Industrial Revolution and contributed to urbanization and a workforce for the Industrial Revolution had one key ingredient: clover. Agricultural productivity at the time was limited by nitrogen. Guano (bird poo) was imported to Britain, but only in the 19th century and global reserves were limited and had to be carried from an ocean away. Synthetic ammonia (Haber-Bosch process) was only invented in the 20th century. Instead, in the British agricultural revolution clover was introduced to crop rotation. The humble clover beats other soil-fixing plants by 3-5x. So, the clover has rightly become a common symbol of good luck in the UK, Ireland and much of Western Europe.
While there is a strong case to be made for overregulation in some areas, it’s also worth pointing out that there are very good reasons for some level of environmental and safety regulation. Personally, I don’t really miss Victorian “medicine”, boracic acid in milk, alum in bread, arsenic paint on the wall, radioactive toothpaste, cars without seatbelts, leaded gasoline, asbestos, or chlorofluorocarbons. It is well-established that information asymmetry about the quality of products will naturally lead to a market equilibrium favoring low quality products without a regulator. In high-performance industries the most innovative companies are also the safest (e.g., Tesla), and corner-cutting doesn’t pay off (e.g., Boeing).
E.A. Wrigley. (2004). British population during the ‘long’ eighteenth century, 1680–1840. In: The Cambridge Economic History of Modern Britain. Cambridge University Press. p. 69
The observed growth of AI compute has been significantly faster in 22-24 than in this model. For example, semianalysis anticipates a growth of about +350% rather than +80% in 2024. I’m taking conservative assumptions here because a) boom may not continue at this pace, b) it’s sufficient to get the point across.
Wassily Leontief. (1983). Technological Advance, Economic Growth, and the Distribution of Income. Population and Development Review 9(3), 403-410. pp. 405-407; Gregory Clark. (2007). A Farewell to Alms. p. 286; Nick Bostrom. (2014). Superintelligence. p. 196; CGP Grey. (2014). Humans Need Not Apply. youtube.com; Calum Chace. (2016). The Economic Singularity. p. 189; Max Tegmark. (2017). Life 3.0. pp. 125&126
Max Tegmark. (2017). Life 3.0: Being Human in the Age of Artificial Intelligence. pp. 125&126
Calum Chace. (2016). The Economic Singularity. p. 180