My father repaired engines for most of his adult life. He never asked what an engine wanted, or what it was entitled to, or whether it would eventually surpass him. 
These questions would have struck him as faintly absurd — the sort of thing a man says when he has spent too long away from the workshop and not long enough under a chassis. What he understood, in the particular way that people understand things through their hands rather than their heads, was something rather more precise: that every tool has a threshold, and that the moment a tool crosses it — the moment it becomes capable of something its maker did not fully anticipate — the relationship between the person and the object is permanently altered.
I have been thinking about that threshold for weeks now. Not in relation to engines, but in relation to the systems we are building with such confidence and such urgency in the opening years of the twenty-first century. I have been reading the historical record — not searching for reassurance, which is what most people look for when they read history, but searching for the pattern. Because the pattern exists. It has always existed. And it does not especially care whether the people living through it have noticed it.
What follows is an attempt to trace that pattern from the Bronze Age to the present, and to say honestly what it suggests about where we are going.
In an AI-modified image, a 1983 Soviet stamp commemorating 1200 years since the birth of the Persian mathematician and astronomer Al-Khwarizmi (c. 780–850 AD), from whom the term algorithm comes.
A Tool Is Never Just a Tool
Before the specifics, a point of principle. Every technology that has reshaped the global order has done so through the same mechanism: asymmetric advantage. The tool does not need to be universally available to be historically decisive. It needs only to be available to one actor — a state, an empire, a trading bloc — before its rivals. Even a brief monopoly, measured in years rather than decades, is sufficient to restructure the world.
The historian Peter Turchin, through his discipline of cliodynamics — the mathematical modelling of historical dynamics — has documented this with considerable rigour. Advances in military technology, he argues, exercise a transformative influence on civilisational structure that extends far beyond the battlefield. They reconfigure economies, accelerate ideological change, and redraw the maps through which power is distributed. The economist Michael Kremer adds a further dimension: larger populations produce proportionally more inventors, which accelerates the pace of innovation — a self-amplifying loop that, once initiated, is extraordinarily difficult to interrupt.
The pattern Turchin and others have documented runs as follows: invention at a single node; a period of strategic monopoly; gradual proliferation to rival actors; equilibration as all major players adopt the technology; and then permanent reconfiguration of the global order at a higher level of complexity. It has happened five times in recorded history in ways sufficiently decisive to be unmistakeable. It appears to be happening a sixth time now.
Bronze and Iron: The First Asymmetry
The story begins, as so many stories do, with metal.
Around 3000 BCE in Mesopotamia, the development of bronze weapons and armour created the first documented strategic inequality. Bronze is harder and more durable than stone or copper. The civilisations that controlled sources of tin — the rare component that makes bronze possible — and the trade routes along which tin moved became militarily superior in ways that no amount of organisational ingenuity could easily compensate. This was not merely an advantage in battle; it was an economic transformation. The control of mining resources and commercial routes became as strategically significant as the size of an army. War and economy had fused.
What makes the Bronze Age especially instructive is its ending. The Late Bronze Age Collapse, around 1200 BCE, brought widespread disruption across the Eastern Mediterranean. The collapse was systemic — involving climate shifts, population migrations, economic disruptions, and military defeats that operated simultaneously and reinforced one another. What this catastrophe reveals is how thoroughly interconnected the Bronze Age civilisations had become, and how fragile complex technological and economic systems are when subjected to compound pressure. Interdependence, which had been the source of strength, became the mechanism of collapse.
Then came iron. And iron was different in a crucial respect: it was geographically distributed and relatively cheap to produce. The transition from bronze to iron democratised armament. Peoples who had previously been marginal — Philistines, Anatolian tribes, Gauls — could now field armies equipped with weapons that matched those of the great empires. The strategic monopoly of bronze dissolved. The lesson of this dissolution is one that has repeated itself so often since that it ought by now to be a truism, but apparently requires constant rediscovery: when a technology becomes widely accessible, the former monopoly disappears. The actor who invested earliest in scaling the technology retains an advantage — but not an insuperable one.
Gunpowder and the Geometry of Power
The introduction of gunpowder into European warfare across the thirteenth to sixteenth centuries did something that no previous military technology had quite managed: it made the castle obsolete. Walls that had guaranteed security for generations proved useless before artillery. The economic foundation of feudal nobility — the castle as a strategic asset — was abruptly invalidated.
This was not merely a military event. The transition from aristocratic cavalry and elite infantry to mass armies of disciplined professionals redefined the relationship between military power and political structure in ways that reverberate to the present day. Only states with centralised resources — meaning, in practice, only nascent monarchies with functioning tax systems and administrative apparatus — could manufacture and maintain artillery. Decentralised feudal arrangements could not. The technology did not simply change warfare; it selected, with considerable ruthlessness, which form of political organisation would survive.
The economic consequences extended further still. The production of firearms drove the earliest state manufacturing enterprises — prototypes of the industrial factories that would transform the world three centuries later. There is a direct line, if one is willing to follow it, from the cannon foundry to the spinning jenny.
Gutenberg and the Democratisation of Cognition
In 1440, Johannes Gutenberg's movable-type press did not merely speed up the copying of books. It transformed the economics of knowledge. Documents that had previously required months or years to reproduce could now be produced in hours. The cost of disseminating information fell by orders of magnitude. The effects were explosive and, in retrospect, almost comically predictable: literacy rates rose, scientific exchange accelerated, and the conditions for both the Renaissance and the Enlightenment were quietly assembled.
Francis Bacon, writing in the early seventeenth century, identified printing, gunpowder, and the compass as the three inventions that had most fundamentally changed the face and condition of the world. He was correct, though he could not have known that he was describing a pattern rather than three isolated miracles.
Economic historians have since documented a causal connection that deserves to be more widely understood: the printing press created the conditions for the Industrial Revolution. The dissemination of technical knowledge, the standardisation of measurements, the ability to communicate innovations across distance and time — these were the preconditions for the progressive accumulation of know-how that industrialisation required. Gutenberg did not cause the steam engine. But without Gutenberg, the steam engine would not have been possible when it was.
The comparison to artificial intelligence is one that serious researchers have been making with increasing frequency. A Brookings Institution report puts it with admirable directness: the printing press liberated ideas and information from captivity; AI accelerates their analysis to produce conclusions that extend the boundaries of knowledge. The distinction, however, matters enormously. The printing press was a distribution technology. It moved existing knowledge more efficiently. Artificial intelligence is something categorically different: it can interpret and synthesise knowledge actively. A well-designed AI system can analyse millions of documents in minutes — a task that would require teams of human specialists working for years. It is not merely amplifying cognition. It is, in the relevant sense, performing it.
The Steam Engine and the Logic of Compound Advantage
Britain acquired an unprecedented economic and military advantage through industrialisation — and then, in the nineteenth century, began slowly to lose it. Germany and the United States adopted industrialisation later but scaled it more efficiently, gradually eroding the British lead. By the end of the nineteenth century, British hegemony was already in structural decline, though few in Britain wished to see it clearly.
The lesson here is subtle but important. The first adopter of a transformative technology gains a temporary advantage that can be enormous. But the long-term gain belongs to whoever scales most efficiently. Britain invented the factory system. America perfected it. This asymmetry between invention and optimisation recurs throughout the historical record and has direct implications for the present moment.
There is a further lesson embedded in the industrial story that tends to receive less attention: the technologies that create the greatest strategic advantages also create the most disruptive social consequences. The Industrial Revolution generated extraordinary wealth. It also generated child labour, urban squalor, and a century of class conflict. The technology did not choose these outcomes. The institutional and political frameworks — or rather, their absence — did. Tools of civilisational scale require governance of civilisational scale. This is not a moral observation. It is a historical one.
The Twentieth Century: When Technology Became Total
The First World War demonstrated, with catastrophic efficiency, what happens when new technologies arrive before the doctrines to govern them. Tanks, aircraft, and poison gas transformed military strategy in ways that the generals of 1914 were entirely unprepared to manage. The human cost of that unpreparedness is measured in the millions.
The Second World War refined the principle. The United States and Britain, geographically insulated from continental Europe, developed strategic bombers to project power at a distance. The divergence in technological investment between the major powers shaped the strategic outcomes of the conflict in ways that are, in retrospect, almost deterministic. Technology did not decide the war by itself. But it established the ceiling of what was possible.
Then came the atomic bomb, and with it the most extreme strategic monopoly in history. For several years, the United States alone possessed the capacity for mass destruction at a global scale. The monopoly lasted until 1949, when the Soviet Union conducted its first nuclear test — but even that brief period of exclusivity permanently redefined the international order. What followed was not a return to normal competition. It was a new logical structure: mutually assured destruction, in which the weapons became deterrents rather than instruments of gain. The technology had become too powerful to use directly. It could only be threatened.
This moment — the nuclear threshold — seems to me of unusual relevance to the present. There is a credible scenario, which I will return to below, in which sufficiently advanced AI systems come to occupy precisely this position.
The Internet and the Architecture of Surveillance
The digital revolution of the 1990s and 2000s decentralised information and communication in ways that initially appeared liberatory. And they were, in many respects. But what was also happening, less visibly, was the construction of new and unprecedented asymmetric advantages for the states and companies that controlled digital infrastructure.
Access to data at massive scale created new forms of strategic advantage: the predictability of population behaviour, informational manipulation, and surveillance of adversaries. By the time these implications became widely understood — by the mid-2010s, at the earliest — the architecture was already built, the habits already formed, and the dependencies already structural. China and the United States had constructed separate digital ecosystems, prefiguring a technological bifurcation whose consequences are still unfolding.
The internet proved, in the end, not to be a technology that eliminated power asymmetries. It reconfigured them. The question for artificial intelligence is whether the same reconfiguration will occur — or whether AI is sufficiently different in kind to produce a genuinely novel outcome.
What Artificial Intelligence Actually Is
This is the point at which precision becomes important, because the word "artificial intelligence" is currently doing an enormous amount of rhetorical work for a very wide range of actual phenomena.
What makes the current generation of AI systems structurally different from every previous technology is not their speed or their scale, though both are remarkable. It is that they automate cognitive functions — analysis, synthesis, abstraction, linguistic generation — rather than physical labour or information storage and distribution. Previous revolutions mechanised the body, or amplified the reach of the mind. This one is automating the operations of the mind itself.
Academic researchers at arXiv have argued, in a 2025 paper titled "Closer to Language than Steam," that AI resembles language more than it resembles the steam engine. Writing was the static storage of human thought; AI can actively manipulate and generate new thinking dynamically and interactively. This is not a poetic distinction. It is a structural one, and it has direct implications for the kind of strategic advantage that advanced AI can confer.
The US Department of Defence has adopted this framing with considerable candour. Its various AI initiatives — including the creation of the Chief Digital and Artificial Intelligence Office — are oriented explicitly around the concept of "decision superiority on the battlefield." This is precisely the same strategic principle that governed the transition from stone to bronze, from lance to firearm, from cavalry to tank. The language is modern. The underlying logic is ancient.
The Current Geopolitical Race
The Atlantic Council characterised the US–China AI competition in November 2025 as the most significant technological race since the nuclear era — but this time the weapons are algorithms rather than atoms. The Chinese government regards AI as decisive for what it calls "comprehensive national power." The framing is explicit, unsentimental, and historically literate.
The convergence data is striking. At the beginning of 2024, the performance gap between the best American AI models and the best Chinese ones stood at approximately 9.3%. By February 2025, that gap had narrowed to 1.7%. This is exactly the pattern of technological proliferation that the historical record predicts: initial monopoly, rapid convergence as rivals invest, approaching equilibration. The only question is whether the first mover can capitalise on its advantage before the window closes.
By 2026, the United States had crystallised an explicit export strategy: to satisfy global demand for AI by exporting its entire AI technology stack. This follows the logic of British industrial hegemony through the control of standards — whoever sets the technological standards controls the architecture of power. India launched its own national AI model in 2026. The dynamics of nuclear proliferation are being reproduced with notable fidelity: every significant regional power feels the pressure to develop an independent capability. Dependence on another actor's AI infrastructure is, increasingly, understood as a form of strategic vulnerability.
The Stargate initiative, announced in 2025, commits five hundred billion dollars to AI infrastructure over five years. The principle animating this investment is stated directly by its proponents: nations must control AI before AI controls them.
Four Scenarios for 2025–2040
I am not, by inclination, a futurist. Prediction is a form of hubris that history punishes with particular consistency. What the historical record permits is not prediction but the identification of structural scenarios — the set of futures that are consistent with documented patterns.
The first scenario — cognitive hegemony. One actor, or a small group of actors, maintains a significant AI advantage for five to ten years. The advantage is self-amplifying: more computational power produces better models, which generates more economic productivity, which funds more AI research. The actor that holds this loop holds a disproportionate economic and military advantage. The analogy is Britain in the Industrial Revolution — not a victory by force, but by compounded efficiency. The expected outcome is a profound restructuring of global alliances around access to AI technology.
The second scenario — rapid proliferation. Advanced AI becomes widely accessible, as iron superseded bronze. Open-source models, falling computational costs, and the diffusion of technical knowledge democratise access to AI capability. New actors — medium-sized states, non-state actors — acquire competitive capabilities and disrupt the established order temporarily, before a new equilibrium forms. This scenario is destabilising in the short term but potentially stabilising over a longer horizon.
The third scenario — global bifurcation. Two separate and incompatible AI ecosystems — the American model and the Chinese model — solidify. States are compelled to choose an ecosystem, creating blocks of technological influence that mirror political alliances. Interoperability decreases. Technological dependencies become instruments of geopolitical coercion. The Cold War, transposed from nuclear to digital. This is, it must be said, the scenario that current trajectories most clearly suggest.
The fourth scenario — AI as deterrent. Advanced military AI becomes, like the nuclear weapon, too dangerous to use directly. Autonomous systems capable of rapid military decisions create instability analogous to the early nuclear era — the fear of a first-strike creates pressure towards pre-emptive action. The technology becomes a deterrent rather than an instrument of gain. This scenario demands, with considerable urgency, a global governance framework analogous to the nuclear non-proliferation treaties. Whether the international community is capable of constructing such a framework — given that it has not managed to do so for the less complex problems of the past three decades — is a question I find myself unable to answer with optimism.
The critical variables that will determine which scenario, or which combination of scenarios, actually unfolds are: the speed of proliferation; whether effective international governance norms emerge before a major crisis; the depth to which AI replaces human decision-making in critical domains; and — a factor that receives insufficient attention — the energy infrastructure required to power AI computation. The data centres that run large language models consume electricity at a scale that makes access to energy sources a direct strategic variable. The geopolitics of oil has been replaced, partially, by the geopolitics of chips and electricity.
What Is Actually New
I want to be careful here, because the historical analogies can be read as reassuring — as suggesting that this is simply another turn of a familiar wheel, and that things will work themselves out as they always have. This reading is available, but I think it is wrong, or at least incomplete.
There is something genuinely unprecedented about a technology that automates cognition itself. Every previous transformative tool — the bronze sword, the cannon, the steam engine, the computer — amplified human capacity in some domain while leaving the domain of judgement, analysis, and decision intact. AI does not leave that domain intact. A state with advanced AI can make better decisions, more quickly, across all domains simultaneously — economic, military, diplomatic, scientific — in a way that no previous asymmetric advantage permitted.
This is why the recovery gap matters so much more than in previous technological transitions. If you fall behind in steel production, you can catch up. The knowledge is available. The infrastructure can be built. If you fall behind in cognitive infrastructure — in the capacity to process information, identify patterns, and generate strategic options faster than your rivals — the gap compounds rather than closes. The advantage, once established, is harder to reverse than any previous strategic asymmetry in recorded history.
The historians and researchers who have studied the contemporary AI race consistently arrive at a conclusion that is simultaneously obvious and profoundly difficult to act upon: the period of monopolistic technical advantage is invariably accompanied by instability and conflict. The transition from any one technological era to the next has been disruptive — but eventually governable, as new norms, standards, and institutions have emerged. The question is whether governance can evolve in parallel with the technology this time — or whether, as in 1914, the doctrine will lag behind the capability until the lag becomes catastrophic.
An Honest Conclusion
My father would have found most of this discussion puzzling, I think. Not because the concepts are difficult, but because he would have wanted to know what we were going to do about it. He had very little patience for analysis that did not lead to action. If the engine was broken, you fixed it. If you did not know how to fix it, you found someone who did, or you learned. The situation was never, in his estimation, improved by talking about it at greater length.
I am not sure he was wrong. But I think the present situation is one in which the action required is not immediately obvious, which is precisely why the analysis matters. We are, as a civilisation, in the position of a person who has built a new kind of engine and is only now — with the engine already running — asking what it will do to the vehicle it is in. The honest answer, based on everything the historical record tells us, is: something irreversible. Something that will restructure the world in ways that will take decades to fully comprehend. Something that has already begun.
The pattern that runs from the Bronze Age to the present is consistent in its broad outlines: the tool arrives, the advantage is temporary, the proliferation follows, the order is reconfigured. What is not determined — what has never been determined — is whether the reconfiguration is managed or catastrophic. Every case in history where governance frameworks emerged in advance of the crisis was better than every case where they did not. That, at least, is unambiguous.
The algorithm is the new weapon. It has been invented. The monopoly is already eroding. The only question that remains open is whether we will manage the reconfiguration, or merely survive it.
Petru Cojocaru is the founder of Absolut Web Expert, a digital agency based in Botoșani, Romania, specialising in web presence for public institutions and private companies. He writes analytical and essayistic content on AI, digital civilisation, and technological strategy at cojocarupetru.info.
Selected sources: Peter Turchin, Cliodynamics (Big Think, 2021) — Brookings Institution, AI and the printing press analogy — RAND Corporation, The Information Age and the Printing Press (1998) — RAND, Building a U.S. National Strategy for the AI Era (2025) — Atlantic Council, Eight Ways AI Will Shape Geopolitics in 2026 (January 2026) — arXiv, Closer to Language than Steam (2025) — arXiv, Three Lenses on the AI Revolution (2025) — Goldman Sachs, The Generative World Order (2023) — JPMorgan Chase, The Geopolitics of AI (2025) — Oxford Academic, Technology, War and the State, International Affairs vol. 95 (2019)