This article turns from sentience to economics — and to the social and political currents that come with it.

The

Across those six developments (described in Appendix A), a consistent pattern emerges — one worth holding in mind as we turn to AI:

1. Early disruption precedes long-term benefit. Each transformation created significant initial costs. Mechanisation displaced artisans, electrification devalued older machinery and skills, digitalisation eliminated clerical and production work, printing displaced scribes, and public-health measures required expensive infrastructure. Oceanic globalisation was far more destructive, involving conquest, disease, slavery and dispossession.

2. Pioneers and investors do not always capture the value they create. Some inventors, entrepreneurs and early investors failed financially despite creating lasting technologies and infrastructure. Railways are a strong example: many pioneering companies entered bankruptcy, receivership or reorganisation, while their tracks continued operating under new ownership. Gutenberg lost control of his printing equipment, and many dot-com companies failed even though their investments helped build the digital economy. Early investors may bear much of the risk while later businesses and society capture the long-term value.

3. Technology alone was insufficient. The greatest benefits appeared only after societies reorganised around the new development. Electricity required grids and redesigned factories; automobiles required roads; digital technologies required new business processes and telecommunications networks; public health required sewers, clean water and organised institutions. Adoption was therefore gradual rather than instantaneous.

4. Large upfront investment created shared infrastructure. Most developments depended on expensive systems that could serve entire economies: factories, railways, electrical grids, roads, ports, printing networks, telecommunications, sanitation systems and healthcare programmes. Once established, these systems lowered costs for many subsequent activities.

5. Costs fell and scale increased. Each transformation drove down the cost of producing goods, moving people and cargo, transmitting information, coordinating organisations, accessing knowledge or preventing disease — and as costs fell, scale grew. Production moved from workshops to factories, knowledge from manuscripts to mass printing, trade from regional routes to global networks, and computing from isolated machines to worldwide platforms. Organisations became larger, markets broader and economic activity more interconnected.

**6. Labour was displaced, reorganised — and unevenly rewarded. \ Older occupations declined while new industries and professions emerged: agricultural and craft work gave way to factories, transport, engineering, administration, healthcare, communications and software. The central effect was not the destruction of jobs but a continuing shift in the kind of work people did. The gains, however, were shared unequally — owners of capital, industrial powers and early adopters tended to benefit first, while workers, colonised populations and poorer communities bore a disproportionate share of the costs.

**7. Knowledge became an increasingly important economic resource. \ Printing allowed ideas to circulate more widely, while digital networks accelerated the storage, copying and transmission of information. Public health also depended on the systematic application of scientific knowledge. Economic progress increasingly came from education, research, communication and organisational capability rather than physical production alone.

**8. Urbanisation and complex societies became possible. \ Industry concentrated workers in cities, transport and electricity supported dense economic centres, and sanitation made large urban populations safer. Modern cities, hospitals, universities, factories and offices depend on several of these transformations operating together.

**9. Every solution created new risks. \ Industrialisation and motorisation produced pollution and fossil-fuel dependence; global trade helped spread disease and exploitation; printing and digital media accelerated misinformation; antibiotics contributed to antimicrobial resistance; and global digital systems introduced concerns over surveillance, monopoly power and employment displacement.

The AI opinion split

Those patterns are the backdrop; the live argument is over how they will play out for AI. Public debate is usually reduced to optimists versus sceptics, but the economic, social and political divisions are more complex. The different camps are not mutually exclusive: someone may consider AI useful but overvalued, productive but socially unequal, or transformative but politically dangerous.

The bubble sceptics, represented most forcefully by Ed Zitron, argue that investment in chips, data centres and AI companies has moved far ahead of proven revenue. A milder version, associated with analysts such as Jim Covello, accepts that AI has useful applications but doubts whether these can generate returns sufficient to justify present spending and valuations. As with railways and the dot-com boom, the technology could survive and transform society even while investors and pioneering companies lose heavily.

The measured-growth sceptics, most notably Daron Acemoglu, do not necessarily predict a crash. They expect AI to produce genuine but comparatively modest economic gains because only some workplace tasks can be automated reliably and profitably. Improvements demonstrated in selected occupations may not translate quickly into economy-wide productivity growth, particularly when adoption requires new skills, processes and infrastructure.

The productivity optimists, including Erik Brynjolfsson, Ethan Mollick and Andrew Ng, regard AI primarily as a tool that will augment workers, lower the cost of cognitive tasks and improve organisational efficiency. They argue that the benefits will emerge gradually as businesses redesign work around the technology.

The more ambitious transformational optimists, such as Dario Amodei, Demis Hassabis and Sam Altman, expect AI to do more than improve existing processes: they believe it could accelerate scientific discovery, reduce the cost of expertise and create major new sources of economic growth.

Other disagreements concern not how much value AI will create, but who will receive it. Labour and inequality critics — including Acemoglu, David Autor and Simon Johnson — warn that higher productivity does not automatically produce higher wages or better employment. AI could augment employees, but it could also replace, deskill or intensively monitor them, allowing shareholders and dominant companies to capture most of the gains. Critics such as Timnit Gebru, Kate Crawford and Meredith Whittaker extend this argument to hidden labour, discrimination, privacy, environmental costs, surveillance and the concentration of technological power.

This leads to the political divide. Market-oriented advocates favour rapid development and relatively limited intervention, arguing that competition and innovation will spread the benefits. Democratic-governance advocates instead believe that competition law, labour protections, privacy rights, independent auditing and public investment are needed to ensure that AI serves society rather than a small number of corporations. A further geopolitical perspective treats AI, semiconductors and computing capacity as strategic infrastructure, making their control a question of national security, economic sovereignty and international power.

The central disagreement is therefore not simply whether AI will succeed. It is whether the value created will justify the investment, how its benefits and costs will be distributed, and who will control the economic and political power that follows.

History, repeating

The reason this belongs in a series called History Repeating Itself is that we have been here before — and the past offers cold comfort to both camps. Some of the patterns in Appendix A are already visible: the early disruption and the vast, debt-fuelled infrastructure build-out. Others I expect to follow — the pioneers who lose while society keeps the infrastructure, and a division of the spoils that favours capital over labour.

Reading the current boom against that history, here is what I expect:

• AI will be one of the larger revolutions, and it will become embedded in every economic, social and political sphere.Few significant decisions will be taken without consulting an AI; countries and organisations will bring AI "advice" into the room at the point of decision. But it will be consulted rather than blindly obeyed; formal human accountability will remain, even where much of the underlying process is automated. Systems of several AI engines working together to solve problems will become the norm, and AI itself will come to be treated like electricity or the internet: a basic necessity rather than a luxury.
• The energy and footprint needed per unit of AI work will fall sharply, as advances in technology and improvements in how systems are trained yield solutions that require far less power. Total demand may well keep rising for years — greater efficiency tends to increase overall use rather than curb it — but the energy and financial cost of comparable AI inference should continue to fall, even if more capable reasoning systems consume more computation per request.
• Many countries will adopt or build open-weight models to protect their own interests and reduce dependence on foreign firms and governments they cannot control.A nation that runs critical systems on someone else’s closed model is exposed to data-access risks, opaque model changes, policy restrictions, abrupt service withdrawal and potentially undetectable bias — risks serious enough to make sovereign AI a matter of national security.
• The AI bubble is real and will burst, taking a number of the pioneering companies with it and wiping out trillions in investor wealth. It will shake the world economy — but, as after the railways and the dot-com crash, a new generation of owners will buy the wreckage cheaply, build on the infrastructure the pioneers paid for, and carry AI into its next stage.
• The market will probably settle around a handful of dominant providers, creating pressure for utility-style regulation, interoperability rules or competition-law intervention.After the shake-out, a few survivors will control the large "frontier" models, because the cost of building them is enormous and favours scale. History suggests what follows: railways, telephone networks and electricity all collapsed into a few giants, and once they became too essential to daily life to fail, governments stepped in — regulating them like utilities or breaking them up under competition law. Expect the same for the dominant AI providers once societies decide they are too important to leave unchecked.
• The models themselves will become cheap; the real money will move to the things around them.As open-weight models spread and prices keep falling, access to a capable model will become a low-margin commodity, much like bandwidth or electricity. The durable profits will accrue instead to whoever owns the scarce complements — the chips and compute, the energy to run them, the proprietary data to train on, and the direct relationship with the customer. As in a gold rush, it is often the sellers of picks, shovels and land who prosper most, not the prospectors.
• It will cause large-scale displacement and losses in particular categories of work at first, as tasks move to AI under human oversight. Many of the roles that remain will become supervisory — reviewing the output of AI, a kind of quality control, and handling the exceptions the models cannot.
• It will also create new work.Some of it will centre on supplying, testing and validating the real-world knowledge AI systems lack — including the collection of physical-world data, simulation, evaluation, exception handling and supervision. One can imagine people employed to raise AIs from "infancy" into a deeper understanding of reality, so that a large share of today's digital workers become AI-workers. There will also be roles that today have not yet been imagined. Manual, dexterous trades will hold their value longest, since physical dexterity is something AI will be slowest to match, even where an AI assistant helps diagnose the problem and suggest a fix.

• The productivity payoff will disappoint before it arrives.For the first few years, measured gains will fall short of the hype — feeding "AI was oversold" headlines — before a later surge. This resembles what economists call the productivity J-curve: early organisational investment can depress or obscure measured productivity before redesigned processes begin producing visible gains, as occurred with electrification and information technology. The benefits show up only once organisations rebuild their processes around the technology instead of bolting it onto the way they already work. Expect a frustrating gap between what AI can obviously do and what appears in the economic statistics.

• Society will end up more productive, able to do more at lower cost. The catch is distribution: a far greater share of the gains will flow to investors than to employees, whose bargaining power in many internationally tradable occupations is already constrained by globalisation, remote hiring and access to a worldwide labour pool.

• As labour's share of the gains shrinks, redistribution will move from the fringe to the mainstream.If output rises while wages stagnate, pressure to share the proceeds by other means will grow. Ideas dismissed today as radical — universal basic income, taxes on automation or on compute, even sovereign "compute dividends" paid to citizens — will become serious political platforms, much as the upheavals of industrialisation eventually produced trade unions, labour laws and the welfare state.
• The gap between the wealthiest companies and individuals and everyone else will widen sharply— and this, in my view, is the real danger. Concentrated wealth buys bargaining power over governments, influence over elections and control of the media narrative. Combined with the vast amount these entities already know about each of us, it lets them micro-target their messaging and steer attention away from the problems that matter towards whatever they would prefer us to look at. That is the biggest threat on this list, and the one most likely to end in social upheaval.

None of this is a reason to dismiss AI, any more than the Railway Mania was a reason to reject railways. The technology is real, and in time it will earn its keep. But the lesson of the last thousand years is that the machine is rarely the hard part; the hard part is who ends up owning it and who is left carrying the cost.

There is, though, one way in which AI may not repeat history: its pace. The revolutions in Appendix A unfolded over decades or generations, which gave societies time — however painful — to adapt, retrain and legislate. AI is moving far faster. The methods behind today's models matured through the 2010s, but the public turning point was ChatGPT's debut in November 2022; less than four years later the technology has directly or indirectly touched billions of lives. If its economic and political consequences arrive at that same speed, we will have far less time to soften the blow or steer the outcome.

So if AI merely repeats that history, the decisive contest will not be technological at all — it will be political. The difference this time is that we may have only years, not decades, to get the answer right, which is why it deserves our attention now, while the outcome is still open.

Appendix A

The committee was given the following prompt:

Identify and rank up to six events from the past 1,000 years — innovations, technologies or developments — that have had the most profound and lasting economic impact on global society. Focus on developments that significantly transformed productivity, trade, employment, wealth creation, economic growth or the structure of society; consider worldwide rather than regional influence; rank from greatest to lesser impact; and provide one well-structured paragraph per item covering the development, its early costs and disruptions, how it ultimately transformed economies, and approximate dates. Use neutral, factual language and base conclusions on broadly accepted historical and economic evidence.

Six Most Economically Transformative Developments of the Last 1,000 Years

Shortlisted from the AI committee's rated list. Selection was driven primarily by the votes each item received, then by scale, duration and worldwide reach of the effects on productivity, trade, employment, wealth creation and social organisation. This is an informed historical assessment, not a precise statistical measurement.

How the six were chosen: the three five-vote items and the two four-vote items qualified automatically. The final place was a tie between two two-vote items — modern public health and the joint-stock company. It was broken in favour of public health, on the grounds that its worldwide reach, duration and documented economic returns are larger and better evidenced, and that the joint-stock company's influence is already embedded in the industrial, railway and digital entries below. Central banking and fiat currency received no committee vote and was excluded.

1. The Industrial Revolution and Mechanised Production (c. 1760–1900) — 5 votes

Beginning in Britain in the late eighteenth century and spreading worldwide, the Industrial Revolution combined mechanised manufacturing, steam power, coal, machine tools, factories, railways and steamships, shifting economic activity from agriculture and handicrafts towards industry. Its early decades were harsh: mechanisation displaced skilled artisans, prompting the Luddite machine-breaking of 1811–1816 (A Farewell to Alms

2. Electrification, Motorisation and Mass Production (c. 1880–1960) — 5 votes

The spread of electrical power, electric motors, internal-combustion engines, telecommunications and assembly-line production from the late nineteenth century created a second industrial transformation across factories, farms, transport and homes. Adoption was slow and costly: it demanded enormous investment in generating stations, grids, roads and redesigned plants; older equipment and occupations lost value; assembly-line work could be repetitive and tightly controlled; motor vehicles brought mass casualties; and dependence on coal and oil produced severe pollution and long-term climate damage. Once the infrastructure and organisational changes were in place, however, the gains were substantial — a study of US manufacturing from 1890 to 1940 found that electrification produced labour-productivity gains that were "rapid and long-lasting," accompanied by capital investment and new production methods (Powering Up Productivity

3. The Digital Revolution and Global Information Networks (c. 1945–present) — 5 votes

Electronic computers after the Second World War, followed by semiconductors, personal computers, the commercial internet, mobile communications, cloud computing and artificial intelligence, turned information into a resource that can be processed, copied and transmitted globally at almost no marginal cost. The transition carried familiar costs: early machines were confined to governments and large firms; many early corporate investments showed no immediate payoff — captured in Robert Solow's 1987 quip that the computer age was visible everywhere except in the productivity statistics; automation displaced clerical and production workers; and the dot-com crash of 2000–2002 erased more than US$5 trillion in market value and bankrupted thousands of firms (

4. Oceanic Globalisation and the Columbian Exchange (from 1492) — 4 votes

Sustained voyages between Europe and the Americas from 1492 connected previously separate ecosystems and economies, creating large-scale exchanges of crops, animals, people, pathogens, precious metals and commercial institutions across the Americas, Europe, Africa and Asia (Journal of Economic Perspectives

5. Movable-Type Printing and Mass Knowledge Distribution (c. 1440–1700) — 4 votes

Movable type appeared in China in the eleventh century, but Johannes Gutenberg's European press of the 1440s combined reusable metal type, suitable ink and mechanical pressing into a commercially scalable way to reproduce books, documents and news. Printing initially demanded substantial capital, displaced scribes, spread errors and inflammatory claims faster, intensified religious and political conflict, and drew censorship from authorities; Gutenberg himself lost control of his printing equipment in a financial dispute with his backer Johann Fust. Over time it sharply reduced the cost of preserving and circulating information: book prices fell by roughly two-thirds between 1450 and 1500, and cities that adopted the press grew about 60% faster than comparable cities between 1500 and 1600, as business manuals, science and literacy spread (

6. Modern Public Health, Sanitation and Medical Prevention (c. 1850–present) — 2 votes (tie-break winner)

From the nineteenth century, clean-water systems, sewerage, germ theory, vaccination, antibiotics and organised public-health programmes reduced the vast human and economic toll of infectious disease and premature death. Implementation was costly and contested: cities needed extensive infrastructure, early treatments were sometimes ineffective or unsafe, vaccination met resistance, and access remained unequal; later antibiotic misuse bred resistance, while longer lives placed new demands on pensions and healthcare. The economic dividend, however, was enormous. Lower mortality and better health let people attend school, work more consistently and invest over longer lives; a landmark modelling study estimates that vaccination against 14 diseases prevented roughly 154 million deaths between 1974 and 2024 and accounted for about 40% of the global decline in infant mortality (The Lancet