The Industrialization of Intelligence and Re-imagining Human Identity
For several years, much of the conversation around artificial intelligence has revolved around a deceptively simple question. When will AI become as intelligent as humans? There are variations of the same question everywhere. When will we reach AGI? When will AI replace programmers? When will it replace knowledge workers? When will it discover new science? When will it become superintelligent? These are fascinating questions. They are also, in some ways, the wrong questions. Not because they don’t matter. They do. But because they make AI appear to be a destination. A point on a timeline. A moment when something called “AGI” arrives and history changes. I suspect the reality will be less dramatic. The transformation may already have begun. It may not have a single starting date. There may be no morning when we wake up and discover that artificial general intelligence has arrived.
Instead, capability will continue to move into places where it previously did not exist. A machine will write software. Then manage software. A machine will assist a scientist. Then design an experiment. A machine will analyze a factory, then operate one. A machine will help a company make decisions, then become part of the company itself. None of these events, taken individually, necessarily looks like the beginning of a new civilization. Together, they may be.
This is why I have become less interested in predicting the exact arrival of AGI and more interested in understanding what happens when intelligence itself becomes increasingly engineered, scalable and deployable. That is a different question. And a much bigger one. The forecasts already tell us something important. They disagree. The OECD’s 2026 exploration of possible AI trajectories describes four broad possibilities through 2030, progress could stall, slow, continue along current trends, or accelerate substantially. The UK’s 2026 AI Scenarios exercise similarly develops five plausible and stretching futures rather than presenting one prediction.
Even the forecasting community that produced AI 2027 has since revised its timelines as new evidence arrived. Its newer AI Futures Model places its median superintelligence timeline substantially later than the original AI 2027 model. This is not a failure of forecasting. It is a reminder of something more fundamental. The future of AI is uncertain. And uncertainty does not mean that we know nothing. It means that we need a better way of thinking. Rather than asking which forecast will turn out to be correct, we should ask, what transformations would matter across several different futures?
If AI progress slows, some transformations may still occur. If progress continues, they may accelerate. If progress becomes extraordinary, entirely new ones may appear. The job, then, is not to predict one future. It is to understand the structure beneath many possible futures. That is what this work attempts to do. There is another reason. AI is often discussed as a technology sector. That makes sense if we are looking at companies, models and investment. But artificial intelligence is becoming difficult to contain within the boundaries of a sector.
It touches energy because computation requires electricity. It touches semiconductors because intelligence requires hardware. It touches manufacturing because increasingly capable AI can control machines. It touches science because researchers can use AI to search enormous spaces of possible discoveries. It touches finance because decisions can be automated. It touches education because teaching can become personalized. It touches defense because intelligence has always been strategic. It touches government because institutions themselves depend on information and decision-making. And eventually it touches something much harder to measure that is Human Identity.
If a machine can perform a task that once gave a person expertise, status or purpose, the consequence is not captured by productivity statistics. Something changes inside the person as well. This is where the AI conversation becomes much more interesting to me. Because the industrial and human dimensions are not separate. They are connected. A change in technology changes the economy. The economy changes the organization. The organization changes work. Work changes society. Society changes the way people understand themselves. And eventually those changes feed back into the technology itself. AI is therefore not simply a new tool entering an existing system. It is a new capability entering a system that will reorganize around it.
01. We Have Been Externalizing Ourselves.
There is a way of looking at human history that I find increasingly useful. Our civilization is the story of human beings gradually moving their capabilities outside themselves. At first, we had only our bodies, hands, feet, senses, memory, and brains. Then we began making things. A stone became a cutting tool. A stick became a spear. A wheel allowed the body to move things it could not move alone. Writing allowed a thought to survive its thinker. The telescope extended the eye. The microscope revealed a world the eye could never see. The steam engine amplified muscle. The electric motor amplified mechanical power. The telephone extended the human voice across distance. The computer extended calculation. The internet connected minds across the planet. None of these technologies merely added convenience. Each one moved a human capability into the external world.
That is what makes technology so powerful. It allows something that was previously confined to a biological organism to become part of the environment. Once externalized, that capability can be combined with other capabilities. The wheel can be combined with the engine and the engine with the factory. The computer can be combined with the network and the network with billions of people. Eventually the system becomes much more than the original invention. This is one reason technological change is so difficult to predict. We tend to look at the invention. History is shaped by the combinations that follow. For most of this history, however, one important boundary remained. Machines could extend our physical abilities. They could extend our memory. They could extend our calculations. They could extend our reach. But intelligence itself remained largely inside the human being.
A machine could calculate the trajectory of a spacecraft. A human had to decide where the spacecraft should go. A computer could search millions of records. A human had to decide which question was worth asking. A factory could produce thousands of components. A human decided what product to build. The machine was powerful. But the human supplied the purpose. This distinction has shaped the industrial world. It is also the distinction that AI is beginning to disturb. The significance of artificial intelligence is therefore not simply that machines are becoming more capable. Machines have been becoming more capable for centuries.
The more unusual development is that some of the capabilities we are engineering into machines are the same capabilities that humans have historically used to control the machines like planning, reasoning, interpreting information, generating alternatives, learning from experience, making decisions, and creating. This creates a kind of recursion. We built machines to extend ourselves. Now we are building machines that can help design, operate and improve other machines. And potentially help improve the systems that create them. That is a new territory.
We should be careful about overstating it. Today’s AI is not a human mind in a computer. It does not experience the world as we do. It does not necessarily understand in the way a person understands. And the question of consciousness remains entirely separate. But none of those caveats removes the practical significance. A technology does not have to reproduce the whole human condition to change human civilization. The steam engine did not need to become a horse. The computer did not need to become a mathematician. AI does not need to become a human being. It only needs to become capable enough at economically and socially important tasks to change the systems around it. That threshold is already worth taking seriously.
02. The Industrial Revolution Was Not About Machines.
We often describe the Industrial Revolution through machines, steam engines, factories, railways, and textile mills. But the deeper change was not the machines themselves. It was the discovery that human capability could be systematically amplified. Muscle could be replaced by mechanical power. Production could be standardized. Work could be divided into repeatable processes. Energy could be concentrated and directed. Production could be scaled far beyond the limits of an individual craftsman. The factory was therefore more than a building full of machines. It was a new way of organizing capability.
That is an important lesson for AI. The real transformation may not come from the model. It may come from the systems built around the model. An AI model sitting in a chat window is one thing. An AI connected to a company’s software, databases, financial systems, laboratories, machines and people is something else entirely. The capability of the individual model matters. But the architecture surrounding it may matter even more. This is why I think the phrase “AI revolution” can be misleading. It suggests a single invention. What we may actually be witnessing is the emergence of a new layer of infrastructure. A layer that can increasingly perform parts of the work of intelligence across almost every existing system.
03. From Muscle to Mind.
The first great industrial transformation made physical power abundant. The second made computation and information processing abundant. The emerging transformation may make certain forms of intelligence abundant. These three transitions are related. But they are not identical. The steam engine changed what a machine could physically do. The computer changed what a machine could calculate. AI is changing what a machine can do with information. That distinction sounds small. It isn’t. Calculation follows rules. Intelligence, at least in the practical sense in which we use the word, involves navigating uncertainty.
There may be many possible answers, many possible strategies, incomplete information, conflicting objectives, and changing circumstances. AI systems are increasingly useful precisely in these spaces. They can produce a first draft, compare alternatives, summarize large bodies of information, generate code, explore designs, suggest explanations, and find patterns. The machine is no longer simply calculating. It is participating in the process by which we arrive at an answer.
This is where the economic implications begin. Human expertise is expensive. Not because intelligence itself has a price tag, but because capable people take years to train, are limited in number and can only work on a finite number of problems at once. A good engineer is scarce. A good scientist is scarce. A good programmer is scarce. A good strategist is scarce. A good teacher is scarce. AI potentially changes the economics of this scarcity. Not by making humans unnecessary overnight. But by allowing one human to work with a much larger amount of cognitive capability.
That is the first stage. The human becomes amplified. But there is another stage. The machine begins doing the task without continuous human involvement. Then intelligence is no longer merely an augmentation of labour. It becomes a productive input in its own right. This is the idea I will return to throughout this work, The Industrialization of Intelligence and Transformation of Human Identity.
For the first time, a capability that has historically been embodied in individual human beings can be instantiated in a technological system and deployed repeatedly. That does not mean intelligence has become a commodity in the same sense as steel or electricity. It is more complicated than that. But it begins to acquire some properties of an industrial resource. It can be trained, copied, distributed, updated, and scaled. And potentially combined with other forms of capital. That last point may be the most important.
A machine with intelligence but no tools has limited economic reach. Give it software, databases and communication systems and its reach expands. Give it a laboratory and it can experiment. Give it a factory and it can manufacture. Give it a robot and it can act physically. Give it capital and the relationship changes again. Intelligence begins to connect with the rest of the production system. That is when the transformation becomes difficult to contain.
04. What Happens When Intelligence Becomes Cheap?
There is a thought experiment worth keeping in mind. Suppose useful intelligence becomes ten times cheaper, not ten times smarter, but ten times cheaper. What would change? A small business might be able to access expertise that previously required a large consulting firm. A scientist might investigate hundreds of hypotheses instead of ten. A doctor might have analytical support available continuously. An engineer might explore thousands of designs before choosing one. A student might have an always-available tutor. An entrepreneur might test an idea with a team of AI systems before hiring anyone.
The immediate result is productivity. But productivity is only the first-order effect. The second-order effect is experimentation. When the cost of thinking through an idea falls, more ideas get tested. Some will be bad. Most will probably be very bad. That doesn’t matter. The number of experiments increases. And when experimentation increases, the probability of discovering something valuable increases too. This is why AI may affect innovation more deeply than simple automation statistics suggest. It could change not only how efficiently we perform existing tasks, but how many things we are able to try. That is a different kind of economic growth. It is growth through expanded possibility.
There is a second thought experiment. What if intelligence becomes not ten times cheaper, but almost continuously available? There is no waiting for an expert and scheduling a meeting. There is no limit imposed by the working hours of one person. There is no need to choose between ten problems because there are only two experts available. You simply instantiate another system. This is where the industrial analogy becomes powerful.
Factories changed production because machines allowed physical work to scale beyond the number of human muscles available. AI could do something conceptually similar with certain forms of cognitive work. The scarce resource is no longer only the number of people capable of doing the work. The question becomes the amount of compute, energy, data, capital and infrastructure we are willing to devote to the problem. That is a deep change in the economics of intelligence. And we have barely begun to understand its consequences.
05. The Human Question Appears.
There is an irony here. The more successful AI becomes at making intelligence abundant, the less valuable intelligence may be as a source of individual status. For centuries, being knowledgeable mattered because knowledge was difficult to acquire. Being skilled mattered because skill took years. Being able to solve difficult problems gave people a place in society. If machines make many forms of expertise abundant, we may need to rethink what those things mean. This does not make human expertise worthless.
Quite the opposite will happen. Human judgement will become more important precisely because machines become better at generating possibilities. Someone still has to decide which possibility deserves to become reality. Someone has to accept responsibility. Someone has to care about the consequences. Someone has to decide what is worth optimizing. That distinction will become increasingly important. Because intelligence can answer a question. It cannot, by intelligence alone, determine whether the question was worth asking.
That is where this piece of work eventually has to go. But before we reach the human question, we need to understand the industrial one. Because if intelligence really is becoming an industrial resource, the consequences will first appear in the places where resources have always mattered most like companies, factories, laboratories, energy systems, markets, and states. The transformation of those systems may be the bridge between artificial intelligence and the transformation of civilization itself.
06. From Assistant to Agent.
The first generation of AI systems mostly waited for us. We asked a question. They answered. We gave them a document. They summarized it. We asked for code. They generated it. We asked for an image. They created one. It was impressive, but the relationship was still familiar. The human remained in charge of the process. The machine produced something. The human decided what happened next. That distinction is beginning to disappear.
The next generation of AI is increasingly being designed not simply to respond, but to pursue an objective. This is the transition from assistant to agent. It may turn out to be one of the most consequential changes in the entire AI story. Imagine asking an AI system to investigate whether a new product category is worth entering. An assistant might give you a market analysis. An agent could do something different. It might search public information, examine competitors, analyze pricing, identify suppliers, build a financial model, compare regulatory constraints, generate a product concept and propose a sequence of experiments. It could then monitor what happens and revise its plan.
The important change is not that it can do more tasks. It is that the tasks become connected by an objective. The system begins to operate over time. That introduces something that ordinary software has largely avoided, that is initiative. Not initiative in the human psychological sense. We should be careful with that word. The system does not necessarily have desires. But operationally, it can initiate actions that were not individually specified by its user.
That distinction matters. A traditional software program might be given, if, X happens, do Y. An agent may be given, achieve Z. The system determines some of the intermediate steps. That is a very different relationship between humans and machines. For most of the computer age, software has been deterministic enough that we could understand the chain between instruction and result.
Modern AI is different. We give it an objective, context and constraints. It generates a path. Sometimes the path is obvious. Sometimes it is surprising. As these systems become more capable, the space between intention and execution becomes increasingly occupied by the machine. That space is where management has traditionally lived. A manager does not tell an employee every keystroke. The manager establishes an objective and allows the employee to determine how to accomplish it.
AI agents begin to resemble this structure. This creates an interesting possibility. The first genuinely AI-native organizations will not look like companies with chatbots everywhere. They will look like companies with machine workers, not workers in the legal or moral sense. But systems that receive objectives, access tools, perform tasks, interact with other systems and produce outcomes. Once that becomes reliable, the economics of organizations could change very quickly.
07. The Cost of Coordination and Responsibility Gap.
A surprisingly large part of modern business exists because coordination is expensive. We create departments because specialized work needs to be organized. We create managers because people need direction. We create meetings because information is distributed. We create reporting systems because decision-makers cannot see everything. We create enterprise software because organizations have become too complicated for memory and conversation alone.
AI attacks several of these costs simultaneously. It can observe information across systems, summarize, monitor, identify anomalies, communicate, plan, and execute. The result may be a reduction in what we might call coordination friction. That could be enormously valuable. But there is a danger in interpreting this purely as efficiency. Coordination is not the same as understanding. A company can coordinate itself extremely efficiently while pursuing the wrong objective. An AI system can optimize a metric while damaging the thing the metric was supposed to represent. This is the old problem of optimization. AI does not eliminate it. It makes it faster, much faster.
There is another problem. If an AI system makes a decision, who made it? The person who deployed the system? The engineer who built it? The company that trained it? The manager who approved its use? The system itself? Today, we can often avoid this question because the human remains visibly involved. But as agents become more autonomous, responsibility can become distributed across a chain of people and machines.
That creates what I would call the responsibility gap. The system acted. The human authorized the system. The developer designed the system. The organization created the incentives. Who is responsible for the outcome? This will not remain a philosophical question. It will become a practical question for every major organization using autonomous systems. And eventually, a constitutional question for the society. Because agency without accountability is dangerous. The challenge of the agentic era is therefore not simply making AI capable of acting. It is creating systems in which capability, authority and responsibility remain connected. That may prove much harder.
08. The AI-Native Enterprise and the End of the Department.
The corporation was designed for a world in which human labour was the primary mechanism of production. That statement sounds obvious, but its implications are easy to overlook. The modern enterprise is built around people. Most organizational architecture exists because humans are the ones doing the work. Now imagine removing some of that constraint, not all of it, but enough to change the economics. A company could have a small human team supported by hundreds of specialized AI systems. One might research markets. Another could analyze customers. Another could develop software. Another could monitor operations. Another could handle procurement. Another could model finances. Another could generate marketing campaigns. Another could continuously test alternative strategies. The company would no longer simply employ people and use software. It would compose intelligence. That is a different idea.
For more than a century, organizations have been divided into functions like finance, marketing, operations, human resources, engineering, sales, and research. This structure made sense because people specialize. But what if an AI system can move across functions? A product problem might require market analysis, engineering, financial modeling and customer research. A human organization would create a team. An AI-native organization might create an agentic workflow. The boundaries between departments could become less important. The organization itself could become more fluid. Instead of asking, which department owns this problem? the system might ask, what capabilities are required to solve it? Then assemble those capabilities dynamically. This is potentially a deep organizational shift. The enterprise becomes less like a hierarchy and more like a dynamic network of capabilities. Humans may remain part of that network, but occupy a different role.
09. What Humans Might Actually Do?
This is where many discussions about AI become simplistic. If AI does more work, people do not necessarily do nothing. They may move upward in the hierarchy of decisions. From execution toward judgement. From production toward direction. From analysis toward choice. From following processes toward designing them. That transition is not guaranteed. It depends on how organizations are designed.
A badly designed AI organization could simply turn humans into supervisors of increasingly opaque automated systems. A better one could give people much greater leverage. One entrepreneur might be able to operate a company that previously required dozens of employees. One scientist might lead research programs that previously required a large team. One engineer might explore a design space that would previously have taken years.
The difference is enormous. AI could therefore produce a strange combination of fewer people required for execution, but much greater capability per person. This could create extraordinary entrepreneurial opportunities. It could also produce extraordinary concentration of wealth. Both possibilities need to be considered simultaneously.
10. The Small Company Problem.
The industrial era rewarded scale that created advantages. Large companies could afford factories, logistics networks, specialist departments, and expensive research. AI may weaken some of those advantages. If intelligence and coordination become cheaper, a small company can acquire capabilities that once required substantial organizational size. The result could be a new generation of extremely small but extremely capable enterprises. A ten-person company could compete with a thousand-person company in some industries.
That would be economically liberating. But there is a second possibility. The most powerful AI systems may require enormous amounts of capital, compute and energy. If that happens, the underlying intelligence becomes concentrated even while the applications become distributed. We could therefore see both trends at once, democratization at the application layer and concentration at the infrastructure layer. That tension will be one of the defining economic questions of the AI era.
11. When Science Gets a Second Brain.
Science has always been constrained by human attention. A scientist can think about only so many problems at once, read only so many papers, run only so many simulations, and design only so many experiments. The limitation is not necessarily that we lack ideas. It is that we cannot explore the possibility space quickly enough. AI changes that equation. Its greatest scientific contribution will not be replacing scientists. It will be allowing scientists to search more of the space of what might be true. Consider medical drug discovery. There are enormous numbers of possible molecules. Only a tiny fraction are useful. Historically, researchers have relied on theory, experience, intuition, computation and experimentation to navigate that space. AI can potentially search much larger regions. The same principle applies to materials, catalysts, proteins, battery chemistries, semiconductors, structural designs, and manufacturing processes. In each case, the number of possible configurations can be enormous. Human intuition is powerful. But it is not scalable. Computation is scalable. The combination could be transformative.
12. The Scientific Loop and Discovery Without Understanding.
The most interesting future may not be an AI that gives scientists answers. It may be a system that participates in the entire scientific loop. It proposes, tests, observes, learns, and proposes again. Connect that system to automated laboratories and something new appears. A machine-generated hypothesis can become a physical experiment without waiting for a human to manually translate the idea into a protocol. The result can return to the model. The next experiment can be selected automatically. Science begins to acquire a feedback loop operating at machine speed. This does not eliminate the scientist. It changes the scientist’s role. The researcher becomes increasingly responsible for defining questions, evaluating significance, designing constraints, and interpreting results at a higher level. The scientist becomes less of a manual explorer and more of an architect of exploration. That distinction could become important across almost every experimental field.
There is, however, a darker possibility. Suppose AI discovers something that works. But we don’t know why. This is not entirely new. Scientists have sometimes used empirical relationships before understanding their deeper mechanisms. But AI could make the gap much larger. A system might discover a material with remarkable properties through millions of computational and experimental iterations. The material works. The machine can reproduce the recipe. But the underlying pattern may be too complex for humans to intuitively understand.
We could therefore enter a strange scientific era in which discovery happens faster than explanation. That would challenge one of the deepest assumptions of modern science. We have generally expected explanation to accompany discovery. What happens when it doesn’t? Would we trust a machine-discovered medicine whose mechanism we cannot fully explain? Would we build infrastructure based on designs we cannot completely understand? Would we accept mathematical proofs that no human can meaningfully verify? These questions will become increasingly practical. The scientific revolution of AI may therefore be as much about epistemology as productivity. It could change what we consider knowledge.
13. Screen-less Intelligence and Physical Autonomy.
There is a moment in the development of any technology when it stops being confined to its original medium. Computers began as calculating machines. Then they became communication devices, then business infrastructure, and then cultural infrastructure. AI is now beginning a similar transition. It is leaving the screen. The connection between AI and robotics may ultimately matter more than the appearance of humanoid robots themselves. The important thing is not whether a machine has two arms and two legs. It is whether intelligence can perceive the physical world, make decisions within it, and act upon it. That is a much broader concept.
For years, AI could tell us how to repair a machine. Soon it may be able to inspect the machine. Then diagnose it, order the required part, and perform the repair. The progression is subtle. But once perception, reasoning and action become integrated, the distinction between digital and physical work begins to disappear. A factory becomes an information system with physical outputs. A warehouse becomes an autonomous decision-and-movement system. A farm becomes a distributed sensing and robotics system. A ship becomes a moving computational platform. A laboratory becomes a machine that can design and conduct experiments. The physical world becomes programmable in a new sense.
The internet was largely designed around people looking at screens. AI does not necessarily need that interface. Agents can communicate directly with software. Software can communicate with other software. Machines can communicate with agents. Sensors can continuously provide information. Decisions can be made without a human opening an application. This could make the next phase of the digital world much less visible. The most powerful AI system may be the one we never see. It will sit inside power grids, factories, vehicles, hospitals, supply chains, financial systems, and buildings. It will simply make decisions. This creates an interesting inversion. The first internet put computers in front of humans. The next phase may put intelligence around humans.
The ultimate significance of robotics is therefore not labour replacement. It is autonomy. A physical system that can perceive its environment, reason about it and act without continuous human control can operate in places humans cannot, at times humans cannot, and at scales humans cannot like deep oceans, space, disaster zones, mines, chemical plants, agriculture, and large-scale infrastructure. Eventually, entirely new environments that are economically impossible to operate manually. This is where AI becomes part of the physical metabolism of civilization. And once that happens, the question of intelligence can no longer be separated from energy, materials and manufacturing which brings us to another problem. The machine that makes intelligence abundant still has to obey physics.
14. The Physical Metabolism of Intelligence.
There is something almost misleading about the language we use. We call it “the cloud.” We speak about “digital intelligence.” We talk about software as if it floats somewhere outside the physical world. It doesn’t. Every AI inference is ultimately physical. Electricity moves, transistors switch, heat is produced, cooling systems operate, buildings consume water and energy, chips are manufactured from physical materials and networks require infrastructure. The intelligence may be abstract. Its metabolism is not.
This matters because the future of AI will depend partly on how quickly we can expand the physical systems beneath it. Compute requires semiconductors. Semiconductors require factories. Factories require energy and materials. Data centers require electricity and cooling. Electricity requires generation and transmission. Generation requires capital, land, materials and time. A forecast that looks only at model capability therefore misses half the system. There is an intelligence stack, but beneath the stack is a physical foundation with energy, materials, manufacturing, and infrastructure. Without them, intelligence cannot scale.
It is tempting to imagine that computing will simply become more efficient forever. History gives us good reasons to expect continued improvements. But efficiency can create its own problem. When something becomes cheaper, we tend to use more of it. This is one of the recurring patterns of technological progress. If AI becomes dramatically more efficient, that may not reduce total demand. It may allow AI to spread into many more applications. The question therefore becomes, how much intelligence will civilization want to consume?
It is a strange question. We don’t normally talk about consuming intelligence. But we should. Every organization may eventually decide how much computational intelligence it wants to deploy. Every machine may continuously reason. Every process may be optimized. Every sensor may feed an intelligent system. The amount of computation could therefore become embedded throughout the physical economy. AI could become something closer to electricity: a general-purpose capability consumed by almost everything.
There is an important counterforce. AI itself may help solve the infrastructure problem. It can optimize power grids, discover materials, design better batteries, improve semiconductor architectures, optimize cooling, improve manufacturing, and accelerate scientific research. In other words, AI can increase the efficiency of the very infrastructure required to produce more AI. That creates a feedback loop. More intelligence enables better infrastructure. Better infrastructure enables more intelligence. The question is whether this loop accelerates smoothly or runs into physical constraints. That may be one of the most important uncertainties of the next decade. Because the future of intelligence may ultimately be constrained not by algorithms but by physics.
15. Intelligence Sovereignty.
For most of history, strategic power depended on control over physical resources like land, food, metals, energy, oil, gas, and industrial capacity. In the digital era, another resource became increasingly important, that is semiconductors. AI adds something new. Intelligence itself becomes strategic infrastructure. That changes the geopolitical equation. A country with access to abundant AI capability may have advantages across almost every major sector like science, defense, manufacturing, finance, healthcare, energy, education, cybersecurity, and infrastructure. The technology is general-purpose. That is precisely what makes it strategically important.
It is tempting to compare countries by asking who has the best AI model. That is useful, but incomplete. A model sits at the top of a much larger system. You need advanced chips, fabrication, compute, energy, networks, researchers, engineers, capital, data, and manufacturing. And increasingly, you need robotics and autonomous systems capable of turning digital intelligence into physical capability. This creates a strategic stack. A country can be strong at one layer and weak at another. That weakness matters. A nation may produce excellent AI research but depend on foreign semiconductor manufacturing. Another may possess enormous energy resources but lack advanced computing infrastructure. Another may have exceptional manufacturing but lack frontier models. The countries that can connect these layers may acquire disproportionate power.
This leads to a concept that I believe will become increasingly important, that is intelligence sovereignty. It is the ability of a nation, organization or civilization to maintain sufficient independent access to the intelligence infrastructure on which its critical systems depend. It does not necessarily mean building everything domestically. No modern economy is completely self-sufficient. It means avoiding a level of dependence that becomes strategically dangerous. The analogy with energy is useful. A country does not necessarily need to produce every barrel of oil it consumes. But if its entire economy depends on an external supplier that can shut the supply off, the dependency becomes strategic. AI may create a similar form of dependency. If critical industries, government systems, scientific research or defense infrastructure depend on intelligence controlled elsewhere, the question is no longer simply commercial. It is sovereign.
This may eventually create a different map of technological power, not simply countries with the largest populations or countries with the highest GDP or with the best universities, but countries capable of integrating energy, compute, intelligence, manufacturing, science, and autonomy. That combination could become one of the defining sources of power in the second half of the twenty-first century. And it introduces an uncomfortable possibility. The AI race may not be primarily about creating the smartest machine. It may be about creating the most capable civilizational system around intelligence. That is a much larger competition which has barely begun.
16. Machine-Speed Civilization.
Human institutions were built for human beings. That sounds obvious, but it has consequences we rarely think about. A government moves at the speed of meetings, elections, legislation and courts. A company moves at the speed of management. A university moves at the speed of semesters, committees and academic careers. Science moves at the speed of experiments, peer review and human attention. These systems are slow for good reasons. They contain checks, allow disagreement, create accountability, and give people time to think. But artificial intelligence introduces something our institutions were never designed around, that is machine speed. A machine can read millions of documents in a period in which a person might read a few. It can run thousands of simulations, monitor systems continuously, generate alternatives, detect patterns, respond to changes, and increasingly take action. The difference between human and machine speed may therefore become an institutional problem.
It is not simply that AI can work faster. It is that our institutions may increasingly be unable to process the amount of information and the number of decisions that intelligent systems can generate. Imagine a regulator responsible for overseeing a financial system in which millions of AI-driven decisions are being made every second. The regulator cannot manually inspect them. A human committee cannot meaningfully review every case. The traditional model of oversight breaks down. The regulator itself will need intelligent systems. The same thing could happen in healthcare, infrastructure, defense, scientific research, and environmental monitoring. The institution begins using AI to supervise AI-mediated systems. This creates a strange loop. We may eventually have to govern machines with the help of machines. That is not necessarily undesirable. It may be unavoidable. But it changes the meaning of governance.
Technology can move exponentially while institutions often move incrementally. This creates what we might call institutional lag. The problem is not unique to AI. Previous technologies created similar tensions. The automobile changed cities faster than cities could redesign themselves. The internet changed communication faster than laws governing information could adapt. Social media changed public discourse faster than institutions understood its effects. AI could amplify this problem because it can change not only the environment institutions regulate, but the cognitive capacity of the institutions themselves.
A law may take years to write. An AI system may change substantially during that period. A policy designed for one capability level may become obsolete before implementation. This suggests that future governance may need to become more adaptive, not necessarily faster in every sense. Some decisions should remain deliberately slow. But institutions may need mechanisms for continuous learning. They may need to monitor technological change in much the same way that an intelligent system monitors its environment. The institution itself may need to become a learning system.
There is a deeper problem. Democracy depends partly on citizens being able to understand the decisions made in their name. But what happens when increasingly important decisions depend on systems that are too complex for any individual to fully understand? We already live with complexity. No single person understands the entire global financial system or the electrical grid. But AI introduces a different possibility. The system may not simply be complex because it contains many interacting parts. It may generate strategies that humans did not explicitly design. That makes accountability harder.
If an AI-assisted government system produces a decision, citizens will still ask a very human question, why? If the answer is too complicated to explain, trust begins to erode. This is why explainability should not be understood only as a technical property. It is also a political one. A democratic society may eventually need systems that are not merely capable of making decisions, but capable of making those decisions legible to the people who must live with them. That may become one of the central design problems of AI-era governance.
17. The Synthetic World.
The first internet connected people to information. The next phase may connect people to a world increasingly generated by machines. The distinction between real and synthetic is already becoming less obvious. A photograph can be generated. A voice can be cloned. A person can be simulated. A conversation can be automated. A video can be created without a camera. A piece of software can be written without a conventional programmer. The technical achievement is remarkable. The social consequence is more complicated.
For much of human history, we operated under an assumption that now feels almost invisible, if something existed, someone had probably experienced or created it. That assumption is disappearing. A picture no longer necessarily means that someone stood somewhere and took a picture. A voice no longer necessarily means that someone spoke. A message no longer necessarily means that a person wrote it. The surface remains. The origin becomes uncertain.
The internet created an enormous expansion in the amount of information humans could access. AI may create something even more dramatic. It can create information. And it can create it continuously. Millions of images, articles, videos, software components, personalized educational materials, and synthetic interactions. The limiting factor will no longer be our ability to produce content. It will be our ability to determine what deserves attention.
This changes the economics of information. When information is scarce, information has value. When information is abundant, selection has value. And when synthetic information becomes indistinguishable from human-created information, another resource becomes important, that is trust. We may eventually live in a world where almost anything can be fabricated convincingly. That does not mean that everything will be fake. It means we will have to become much more conscious of provenance. Who created this? Why? When? Under what circumstances? Was a human involved? Was the experience generated? Can I verify it? The answers may become more important than the content itself.
This could produce a surprising cultural shift. For centuries, technology has generally increased our ability to reproduce things. AI may eventually make reproduction almost trivial. At that point, originality becomes less scarce. But authenticity may become more scarce. A photograph taken by someone you love may matter precisely because it was taken by them. A handwritten note may matter because a particular hand wrote it. A live performance may matter because a particular person stood in a particular room at a particular moment. The value is not necessarily in the artifact. It is in the history behind it.
There is a possibility here that initially sounds paradoxical. As synthetic experiences become more abundant, physical reality may become more valuable such as a real journey, a real meal, a real conversation, a real person, a real landscape, and a real object. Not because synthetic experiences are necessarily inferior. Some may be extraordinary. But because scarcity creates value. If anyone can generate an image of a perfect sunset, seeing an actual sunset does not become meaningless. It becomes something else. It becomes an experience that happened. The AI era will therefore create two parallel economies. One based on infinite synthetic abundance. Another based on increasingly valuable human and physical authenticity. We should not assume that one will replace the other. They may coexist.
18. When Work Stops Defining Us.
For most people, the question of AI and employment begins with fear. Will I lose my job? Will my profession disappear? Will my children have work? These are legitimate questions. But they are not the deepest ones. Work has never been only about income. It structures time, creates social relationships, gives people status, provides a sense of contribution, and often becomes part of identity. When someone asks, “What do you do?”, they are rarely asking only how that person pays the bills. They are asking who that person is. That is why large-scale automation could be psychologically disruptive even if material living standards improve. A society can become richer while people feel less useful. That possibility deserves more attention.
Suppose AI makes organizations dramatically more productive. Goods and services become cheaper. Scientific discovery accelerates. Healthcare improves. Robots perform difficult physical work. Economic output rises. This sounds like a wonderful future. But productivity and purpose are not the same thing. A person does not necessarily feel fulfilled because society has become more productive. In fact, the opposite could happen. If machines perform most economically necessary tasks, some people may experience a loss of purpose precisely because they are no longer needed in the way previous generations were. This would be a very strange problem.
Humanity has spent thousands of years trying to escape unnecessary labour. If we succeed, we may discover that labour was carrying psychological and social functions we had never properly identified. The long-term transition is not from employment to unemployment. It is from earning a living to designing a life. That sounds attractive. But it is much harder than it sounds. Employment gives people structure. Without it, society would need other structures. Education might become lifelong rather than front-loaded. Communities might become more important. Creative and social contribution might matter more. People might pursue projects rather than careers.
The distinction between learning and working could weaken. The distinction between work and play could weaken too. Some people may spend their lives exploring science. Others may build communities, create art, care for children or elderly people, explore places that previous generations never had time to visit, or simply live quietly. The point is not that everyone will become an entrepreneur, artist or scientist. It is that we may have to become much better at answering a question that employment has partially answered for us, what is a worthwhile life?
There is another possibility. If AI creates extraordinary productivity but ownership remains concentrated, the benefits may not be distributed evenly. A world with abundant intelligence can still have scarce ownership. A small number of companies could control the most capable systems. A small number of investors could own the infrastructure. A small number of countries could control compute, energy and advanced manufacturing. Then AI could produce abundance without producing equality. This is why the economics of AI cannot be separated from questions of ownership. Who owns the machines? Who owns the models? Who owns the infrastructure? Who captures the productivity gains? Who decides how those gains are distributed? The answer to these questions may matter as much as the capabilities of the technology itself.
19. When Intelligence Becomes Abundant.
Every technological revolution changes the location of scarcity. When agriculture improved, food became less scarce in some societies and human labour could move elsewhere. Industrialization reduced the scarcity of physical power. Computing reduced the scarcity of calculation. The internet reduced the scarcity of information. AI may reduce the scarcity of certain forms of intelligence. But scarcity does not disappear. It moves.
This may be one of the most important ideas for understanding the AI economy. If intelligence becomes cheap, judgement becomes more valuable. If content becomes cheap, attention becomes more valuable. If information becomes cheap, trust becomes more valuable. If design becomes cheap, taste becomes more valuable. If execution becomes cheap, direction becomes more valuable. If simulation becomes cheap, real-world experience may become more valuable. If technical expertise becomes widespread, wisdom may become more valuable. These are not absolute rules. But they point toward something important.
AI may not eliminate scarcity. It may move scarcity toward qualities that are harder to automate. Imagine a world where almost anyone can generate a thousand excellent designs in an afternoon. The problem is no longer making designs. It is knowing which one matters. This is what taste does. Taste is difficult to define because it is partly cultural, contextual and personal. It is not simply technical competence. It is the ability to recognize something worth pursuing. AI may eventually become extremely good at generating possibilities. Humans may therefore become increasingly valuable as selectors. The same could happen in science, business, art, politics, architecture, and education. The ability to generate possibilities is only half of creativity. The other half is choosing.
Trust may become even more important. When machines can generate convincing information at enormous scale, simply receiving information is no longer enough. We need to know whether it is reliable. This creates opportunities for institutions, brands and individuals with credible provenance. Reputation becomes infrastructure. The paradox is interesting. AI may make information abundant while making credibility scarce. That could fundamentally change the economics of reputation.
20. The Human Identity Problem.
There is a deeper challenge and it concerns human identity. Human beings have always understood themselves partly through their capabilities. We are intelligent and we can reason, create, communicate, invent, and solve difficult problems. These abilities distinguish us from other species. Now imagine machines becoming better than us at many of them. Not necessarily all at once. But one by one with better memory, calculation, pattern recognition, language generation, coding, search, simulation, design, optimization, and eventually better scientific reasoning. What happens to our self-image?
For a long time, the answer to “What makes humans special?” was intelligence. Religion offered other answers. Philosophy offered others. But modern technological civilization placed enormous confidence in human rationality. We built institutions around the assumption that human intelligence was the highest general-purpose problem-solving capability available. AI challenges that assumption. It does not necessarily destroy human uniqueness. But it forces us to look for a deeper foundation. Humanity is not valuable because humans are the most intelligent beings. We already know this intuitively. Yet much of modern society behaves as if this capability determines worth.
AI may expose that contradiction. A machine can potentially know more than a person without being wiser. It can process more information without caring about the consequences. It can optimize a system without knowing whether the objective is worth pursuing. This distinction is easy to state but difficult to institutionalize. Our civilization has become extraordinarily good at asking, can we? Technology keeps answering yes. AI may make that answer even more powerful. But the next great question is, should we? That question cannot be solved by intelligence alone. It requires values, judgement, responsibility, and ultimately, some conception of what constitutes a good future.
21. The AI Cascade.
The most important consequences of AI may not be direct. They may be cascading. A capability appears in one layer. That changes another layer. The second change creates a third. The third feeds back into the first. This makes the system nonlinear. Consider a simple example, AI improves software development, software becomes cheaper to build, more companies can automate processes, and automation increases productivity. Higher productivity creates more demand for compute. More compute creates demand for energy and chips. Investment flows into infrastructure. Infrastructure expands. AI becomes cheaper and more capable. More AI enters industry. Industrial productivity rises. Scientific research accelerates. New technologies emerge. Those technologies improve AI infrastructure again. The loop continues.
This is why it may be dangerous to forecast AI by looking only at model benchmarks. The important variable is not simply, how capable is the model? It is, how does that capability propagate with different orders of change through the system? First-order change is easy to see. AI writes software. Second-order change is harder. Software becomes cheaper. Third-order change is harder still. More people can build companies. Fourth-order change is how more companies increase competition. Fifth-order change is how competition changes capital allocation, and eventually, how industries reorganize.
This is where forecasting becomes difficult. The further we move from the original technological capability, the more interactions appear. That is also where the biggest opportunities may exist. The company that benefits most from AI may not be the company that builds the AI. It may be the company that recognizes an industry structure that becomes possible only because AI exists.
22. The Futures We Are Not Preparing For.
Forecasting usually begins with visible trends. What is happening today? What is accelerating? What technologies are improving? These are useful questions. But they can create a blind spot. The future is not only an extrapolation of what is visible. Sometimes the most important change is the one that sits outside the current frame.
What if AI does not primarily replace jobs, but changes the number of organizations that can exist? What if the most important AI application is not a chatbot but an autonomous laboratory? What if robotics becomes the dominant interface between intelligence and the physical economy? What if energy, rather than algorithms, becomes the main constraint? What if scientific discovery accelerates so quickly that regulatory institutions cannot keep up? What if AI makes some countries dramatically more productive than others? What if the largest source of inequality is not access to AI, but ownership of autonomous capital? What if humans increasingly choose not to compete with machines and instead reorganize society around entirely different measures of contribution? These are not predictions. They are stress tests and that is what good foresight should do.
The greatest mistake in long-term forecasting is assuming continuity. The future often contains discontinuities. A technology becomes cheap, regulation changes, scientific discovery unlocks a new capability, supply constraint disappears, new interface changes adoption, geopolitical conflict changes access to infrastructure, and company discovers a new business model. Suddenly the trajectory looks different. AI is particularly vulnerable to this kind of discontinuity because it is connected to so many other technologies. A breakthrough in energy storage, robotics, chip manufacturing, and automated science could affect AI. AI could then accelerate each of those fields in return. The system is coupled. And coupled systems are difficult to predict using simple linear models.
23. Civilizational Intelligence.
The most ambitious possibility is also the easiest to misunderstand. What happens if AI does not merely make individuals more intelligent? What if it makes civilization itself more intelligent? A civilization already has something resembling a collective mind. It senses through billions of people and billions of instruments. It stores knowledge in books, databases and institutions. It communicates through networks. It makes decisions through governments, markets and organizations. It learns through science. But the system is fragmented. Information is distributed. Decisions are slow. Knowledge is often inaccessible to the people who need it. Different institutions rarely share a complete picture.
AI could change this. It could become a connective layer across the system. Imagine an energy network that continuously predicts demand, identifies failures and optimizes generation. A healthcare system that continuously learns from millions of cases. A scientific network that continuously searches for relationships across disciplines. A transportation system that continuously adapts to demand. A government that can simulate the consequences of policies before implementing them. A company that continuously learns from every interaction. Each system becomes more intelligent. But something more interesting happens when the systems connect.
Intelligence becomes distributed across the infrastructure of society. At that point, it becomes difficult to say where the “AI” ends. It is no longer one model. It is a network of models, machines, institutions, people and feedback loops. The intelligence belongs to the system.
From Artificial Intelligence to collective intelligence may be the deeper destination, not machines replacing humans, not humans controlling machines, but a hybrid system in which human and machine intelligence continuously interact. Humans provide values, experience, judgement and meaning. Machines provide scale, memory, computation, simulation and increasingly autonomous action. Neither is sufficient alone. Together they create something different.
This raises an uncomfortable question. If intelligence becomes distributed across a civilization, who controls it? The answer may no longer be a person. It may be institutions, protocols, markets, governments, companies, networks, or some combination of all of them. That is why AI governance cannot be treated as a narrow technology-policy problem. We are beginning to design the architecture of a new layer of collective decision-making.
24. Beyond 2030.
2030 is close enough to feel concrete. That is useful. But it is also dangerous. A date can create the illusion of precision. We begin asking whether a particular capability will exist in 2029 or 2031, as if history respects calendars. It doesn’t. Technological transitions overlap. Some arrive early, some arrive late, and some never arrive. Others appear from somewhere nobody expected. So the purpose of AI 2030+ is not to create another countdown. It is to build a way of thinking about the transition. The period after 2030 may be defined less by a single breakthrough than by the interaction of many breakthroughs in AI, Robotics, Biotechnology, Energy, Materials, Semiconductors, Quantum technologies, Space infrastructure, Synthetic biology, and Autonomous systems. The boundaries between these fields may become difficult to maintain. AI could become the connective tissue between them.
This may be the most important transition we have not yet fully named. Artificial intelligence is converging with the physical world, biology, energy, manufacturing, networks, science, and human cognition. The result is not simply better AI. It is a new technological ecosystem. And ecosystems behave differently from individual technologies. They adapt, interact, produce unexpected combinations, and create feedback loops. That is why the future after 2030 could become qualitatively different from the future before it. Not because some magical threshold is crossed. But because enough intelligent systems may exist simultaneously that they begin to change one another.
The future is not only what we can see. It is also what our models prevent us from seeing. This may be the hardest part of foresight. When we build a model of the future, we necessarily leave things outside it. We focus on capability, economics, employment, geopolitics, safety, and technology. But civilization does not experience these things separately. They interact. A technological change becomes an economic change. An economic change becomes a political change. A political change becomes a social change. A social change changes human behavior. Human behavior changes markets. Markets change technology. The loop closes. This is why oversight matters.
It is the attempt to look at the things that fall between disciplines. The biggest blind spot in current AI thinking is that we treat artificial intelligence as a technology problem. It is not only that. It is also an industrial, energy, institutional, geopolitical, cultural, psychological, philosophical, and eventually, a civilizational problem. The challenge is not to predict every consequence. That is impossible. The challenge is to become better at noticing the consequences that our existing categories hide.
25. What Remains Human?
There is a temptation to end with work about artificial intelligence with a prediction. I don’t think that would be appropriate. The future is too uncertain. And the deeper lesson of this exploration is that uncertainty is not something we can eliminate. We have to learn to think inside it. AI may become extraordinarily capable. It may transform industry, accelerate science, change the structure of companies, alter geopolitics, and make many forms of intelligence abundant. It may change work, education, how we create, and how we govern. Some of these changes are already underway. Others remain speculative. But underneath all of them is one question. What happens to humanity when the thing we have always considered one of our defining characteristics as intelligence, is no longer exclusively ours?
I don’t think the answer is that humans become irrelevant. That would be too simple. Human value was never actually determined by intelligence alone. We value people we love even when they cannot perform a task. We value courage even when it is inefficient. We value kindness even when it produces no measurable output. We value beauty even when it solves no problem. We value curiosity because it opens something inside us. We value freedom because we want to choose our own direction. These things were never primarily about intelligence. AI will force us to remember that.
For most of history, our challenge was to become more capable. We built tools. We built machines. We built institutions. We accumulated knowledge. We became more productive. And now we are building something that may amplify capability beyond anything previous generations could have imagined. But capability creates a new responsibility. The more powerful our tools become, the more important the question of purpose becomes.
A machine can help us build almost anything. That does not tell us what deserves to be built. A machine can help us optimize almost anything. That does not tell us what is worth optimizing. A machine can help us explore almost any possibility. That does not tell us which possibilities we should pursue. Those remain human questions. At least for now. And that is the paradox at the heart of the AI transition.
We are building artificial intelligence in order to expand what intelligence can do. But in doing so, we may discover that intelligence was never the final destination. It was the instrument. The deeper question was always, what will we do with it? That is why 2030 is not the end of the story. It is barely the beginning of the difficult part. The future will not simply be a world with more intelligent machines. It may be a world in which intelligence itself becomes part of the infrastructure of civilization. And when intelligence becomes infrastructure, the question is no longer whether we can create it. The question is whether we can live wisely with what we create. That, ultimately, is the unfinished work. And it should remain unfinished.