When Intelligence Automates the Work of Intelligence
1. The Horse Moment: When Technology Changes the Rules
Horses did not become worse when automobiles arrived. They did not lose their strength, endurance, or ability to carry loads. What changed was the technology available to perform their work. Once automobiles became practical and economical, many of the activities that had made horses valuable could be performed more efficiently by machines.
The horse had not failed. The world had changed around it.
Artificial intelligence presents a comparable challenge to human beings. A graphic designer may retain every skill acquired over twenty years, yet a client may choose AI if it produces an equally good illustration in minutes at a fraction of the cost. A typist may remain excellent at typing, but typing itself may no longer command the same price. An editor may still recognise a poorly constructed argument, but the amount of paid editing required may decline as AI improves.
The unsettling possibility is that people can become economically less valuable without becoming less capable. Their skills remain intact, but the market no longer rewards those skills as it once did.
Yet there is a crucial difference between horses and humans. Horses could not understand the arrival of automobiles, learn to operate them, or decide how transportation should be organised. Humans can do all three. We build the technology, own the infrastructure, establish the rules, and determine how its benefits are distributed.
The question is whether we can adapt our economic institutions as quickly as we adapt our tools.
2. Schumpeter Meets the Luddites
Joseph Schumpeter described capitalism as a process of creative destruction. New technologies and business models disrupt existing industries, destroy established sources of income, and create new opportunities. Economic progress does not preserve every existing occupation; it continually reorganises the activities through which people earn a living.
The Luddites, the nineteenth-century textile workers who resisted mechanisation, are often remembered as enemies of progress. Their resistance, however, also reflected concerns about livelihoods, wages, working conditions, and the distribution of the gains from industrial change. Their experience reminds us that a technological advance can benefit an economy while imposing severe costs on particular groups.
AI brings these two perspectives into direct conflict. Schumpeter explains why businesses adopt a technology that improves productivity and reduces costs. The Luddites remind us to ask what happens to the people whose work becomes less necessary.
Consider an author who once paid separately for illustration, editing, research, and other production services. With AI, the author may be able to perform much of this work independently. The author benefits from lower costs and greater control. The specialists who previously supplied those services may lose customers, even if their work remains excellent.
Neither outcome is imaginary. Both can occur simultaneously.
The mistake is to assume that because technological change increases productive capacity, its benefits will automatically reach everyone. Creative destruction describes a process of economic change; it does not guarantee a fair distribution of its rewards.
3. Darwin at Work: Intelligence Creates Its Successor
Human intelligence evolved within the natural world. It enabled us to recognise patterns, anticipate consequences, cooperate, construct tools, and modify our surroundings. Over time, those capabilities produced increasingly sophisticated technologies, from language and writing to mathematics, industrial machinery, and computers.
AI can be understood as another development in this history. Human intelligence has created systems that can perform increasingly complex forms of information processing, including tasks once associated primarily with human cognition. In this sense, AI is not necessarily an alien force entering nature. It is also a product of nature’s history, mediated through human ingenuity and technology.
Darwinian evolution and technological progress are not identical. Biological evolution operates through mechanisms such as variation, inheritance, and natural selection. AI development is shaped by human design, training, investment, and deployment. Nature does not consciously prefer progress, and evolution does not always produce greater complexity. Nevertheless, the analogy raises a useful question: what happens when a capability developed through evolution begins producing systems that alter the conditions under which that capability is economically valuable?
Human beings developed intelligence to adapt to their environment. We then used intelligence to construct machines that transformed that environment. Now, some of those machines can perform tasks that previously distinguished human workers from other forms of machinery.
Evolution does not guarantee that every organism benefits from the changes it produces. Similarly, technological evolution does not guarantee that every person will benefit from the systems humans create.
The irony is that intelligence may be extraordinarily successful at transforming the world while making some of the activities through which human intelligence earns a living less scarce.
4. The Picasso Principle: Same Tools, Different Outcomes
Give two people the same pen and canvas, and they will not necessarily produce the same painting. Give two people the same AI model, and they will not necessarily produce the same illustration. Access to a tool does not automatically confer the ability to use it exceptionally well.
Consider two AI-generated graphics. One is visually attractive, with harmonious colours, elegant shapes, and balanced composition. The other uses visual structure to explain a difficult idea, reveal relationships between concepts, and help the reader understand something that was previously unclear. Both may be beautiful, but their purposes differ. One is primarily decorative; the other is informative.
An informative graphic requires more than attractive rendering. Its creator must determine what the central idea is, which relationships matter, how those relationships should be represented, and what information should be omitted. The visual structure must serve the argument. Otherwise, the result may look sophisticated while communicating very little.
This distinction matters because AI is making execution increasingly accessible. More people can generate illustrations without years of training in drawing or graphic design. But the ability to conceive a useful visual argument, recognise a weak representation, and guide the model towards a better one does not automatically follow from access to the same technology.
A good graphic designer may therefore remain valuable not simply because the designer can draw, but because years of experience provide a better understanding of composition, visual hierarchy, perspective, and communication. That expertise can be applied to directing AI as well as producing images manually.
The same principle applies to writing, education, research, and other intellectual activities. The same book can be read by two students, the same teacher can teach two classrooms, and the same explanation can be available to everyone. Their outcomes can still differ.
AI can democratise access to the tools of creation without making every creator equally capable.
5. The Idea Person and the Executionist
Before AI, producing a finished article could require an author to conduct research, organise notes, develop an argument, write the text, edit it, create illustrations, and prepare the final presentation. Even when the author supplied the original idea, execution consumed much of the available time.
AI changes this relationship. An author can now explore an argument through conversation, use AI to organise research, develop supporting passages, refine the prose, and generate illustrations. The author can concentrate more of the available time on the conceptual work that makes the article distinctive.
My own experience illustrates the change. Previously, writing an article by hand, including its illustration, limited my practical output to approximately one article per week. With AI, producing as many as two articles a day becomes possible under suitable conditions. The increase does not mean that I suddenly possess twelve times as many original ideas. It means that the time required to turn an idea into a finished article has fallen dramatically.
The distinction is between generating an idea and executing it. AI can make the second process much cheaper without eliminating the need for the first.
This changes the economics of expertise. A specialist who sells execution alone may face intense pressure when AI can perform the same task at lower cost. A specialist who can conceive the right result, specify it clearly, judge its quality, and coordinate the entire production process may retain an advantage. The best graphic designer may become an excellent director of AI-generated graphics, even when the designer no longer creates every element by hand.
The future may therefore reward people who combine conceptual originality with the ability to direct increasingly capable tools. But this advantage is not guaranteed to last. If AI becomes better at developing concepts, planning work, and evaluating results, the boundary between the idea person and the executionist will shift again.
6. A Cognitive Fingerprint Cannot Be Reduced to a Prompt
My approach to AI-generated graphics is not simply to write a prompt and request an image. I construct a map or system that embeds the instructions. The relationships between the ideas, the constraints, and the intended visual outcome shape the prompts that emerge from that system.
The distinction is important. A prompt is an instruction. A system determines how the instructions are developed, connected, evaluated, and revised. The visible graphic is the output of this process, not the whole process itself.
This method reflects a broader characteristic of thinking. People differ in the connections they recognise, the questions they ask, the structures they construct, and the standards they apply when evaluating a result. Their methods of working are partly expressions of their distinctive cognitive patterns.
We can call this a cognitive fingerprint—not a literal neurological measurement, but a useful description of the characteristic way a person transforms information into ideas and outputs.
Suppose someone obtains my books, prompts, graphics, and entire ChatGPT conversation history. That material may reveal much about my approach. A capable person could learn from it and reproduce some of my results. Yet access to the records does not guarantee that the person will independently generate the same new connections or recognise the same opportunities.
The distinction is familiar outside AI. Two students can study the same textbook under the same teacher and achieve different results. Information is transferable, but the ability to interpret and apply it develops through experience, judgment, and practice.
My method is the way I think. The prompts are traces of that thinking; they are not necessarily its source.
This does not make the method impossible to learn or imitate. Nor does it establish that an individual has exclusive rights over a way of thinking. It means that copying the visible instructions is not always equivalent to acquiring the capability that produced them.
As AI makes prompts and production workflows easier to reproduce, the more durable advantage may lie in the capacity to keep generating useful new systems rather than in the secrecy of any single instruction.
7. The Next Transition: When AI Supplies the Instructions
At present, many people gain an advantage by knowing how to direct AI. They provide the concept, specify the desired outcome, evaluate the result, and request improvements. AI performs much of the execution, but human judgment guides the process.
It is reasonable to expect that this arrangement will change as AI systems become more capable of planning, generating alternatives, identifying errors, and improving their own outputs. The precise pace is uncertain. There is no reliable basis for assuming that the transition will take exactly five years. Nevertheless, the direction presents a serious question for anyone whose advantage depends on being better at directing AI than other users.
The first stage of AI adoption may reward people who know how to use the tools effectively. A later stage may reward a different set of capabilities if AI begins supplying much more of the planning and direction itself.
Imagine a graphic designer who initially gains an advantage by understanding composition and translating a client’s needs into sophisticated prompts. As AI improves, it may become increasingly capable of suggesting layouts, explaining trade-offs, and refining images according to a brief. The designer’s expertise may remain useful, but the portion of the task that requires that expertise could shrink.
The same possibility applies to research, editing, programming, and other knowledge-intensive activities. A skill can remain intellectually impressive while becoming less scarce in the market.
This is why the long-term question cannot be answered simply by telling people to learn prompting. Prompting is a current interface to a changing technology, not a permanent guarantee of economic advantage.
The more important challenge is to understand how human agency, judgment, and economic participation can be preserved when AI becomes capable of supplying an increasing share of both execution and direction.
8. A New Role for AI: Finding the Paths We Miss
Humanity has spent generations working on some of mathematics’ deepest questions. In 2000, the Clay Mathematics Institute announced seven Millennium Prize Problems. More than a quarter-century later, six remain unsolved; the Poincaré Conjecture has been solved. Their history does not prove that AI is necessary, but it raises a reasonable question: when brilliant minds have tried for so long, should we only repeat established approaches—or also invite new forms of intelligence to help us see what we may be missing?
AI could therefore become more than a faster problem-solver. It could help expose conceptual blind spots, connect distant areas of mathematics, and suggest routes that human intuition has not explored. The breakthrough may come not from doing the same thing faster, but from approaching the problem differently.
This does not mean AI will automatically solve these problems. It can reproduce our assumptions, pursue unproductive paths, or produce plausible but incorrect arguments. Mathematical claims still require rigorous verification. The point is that AI gives researchers another way to generate conjectures, explore representations, and test approaches that might otherwise remain unnoticed.
Adaptability may be the decisive advantage. When a problem repeatedly defeats our best efforts, progress may require not just more effort, but a different way of thinking. AI should not be judged only by how well it imitates human reasoning; one of its greatest contributions may be helping us discover where our reasoning has become stuck.
9. The Economic Reckoning: Abundance Without Purchasing Power
Imagine an economy capable of producing excellent books, illustrations, software, educational material, and many other services at a fraction of their previous cost. Its productive capacity could rise dramatically. Yet if a large share of people lose the labour income on which they depend, they may struggle to purchase the goods and services that the economy can produce.
Production capacity and purchasing power are different things.
In a market economy, people generally gain access to goods and services through income, ownership, transfers, or other economic arrangements. If AI reduces the need for human labour, the distribution of income may become increasingly dependent on who owns the productive systems and receives the returns they generate.
This outcome is not inevitable. AI may create new occupations, lower prices, expand demand, and enable new forms of work. Human services may also retain value because people care about relationships, accountability, trust, and the experience of dealing with another person. The effects will differ across industries and over time.
But it would be equally mistaken to assume that every displaced worker will automatically find an equally rewarding job. If labour income declines faster than new sources of income emerge, society could face economic insecurity even while aggregate productivity increases.
Possible responses include broader ownership of productive capital, social dividends, universal basic income, shorter working weeks, and expanded public services. Each would involve practical difficulties and distributional trade-offs. None can be treated as a complete solution without examining how it would be funded, administered, and sustained.
The central question is not whether technology should preserve every existing occupation. It is whether an economy can distribute the gains from production when human labour becomes less necessary to generate them.
The horse could not negotiate the terms of its replacement. Humans can influence the terms of technological change through ownership, institutions, and collective decisions. The challenge is to turn that capacity into meaningful economic security.
10. Humans Cannot Become Cars, but They Can Learn to Drive Them
Horses could not compete with automobiles at transportation, but humans could learn to operate automobiles and build entire industries around them. The transition did not require people to become better horses. It required them to change their roles.
AI offers humans a similar opportunity. Instead of competing with a model at every task, people can learn to use it to extend their capabilities. An author can direct AI to explore arguments and produce illustrations. A designer can guide it towards a visual concept. A teacher can use it to generate explanations and examples while retaining responsibility for what students need to learn.
The objective is not to defeat the machine at the task it performs best. It is to accomplish something worthwhile with the capabilities the machine provides.
Yet the analogy has a limit. Learning to drive a car creates a new skill, but it does not guarantee that everyone whose previous livelihood depended on horses will find an equally secure role in the automobile economy. Nor does learning to use AI guarantee that every person will retain a marketable advantage as the technology improves.
If AI eventually performs not only execution but also much of the planning, analysis, and judgment, simply operating the system may not be enough. Humans will need to make decisions about goals, values, accountability, ownership, and the purposes for which the technology is deployed. Whether those activities generate sufficient paid employment for everyone is a separate question.
Humans are not horses. We can understand the transition, influence its direction, and redesign the institutions around it. But our ability to adapt does not absolve society of responsibility for those who bear the costs.
11. The Final Question: What Is the Economy For
Technological progress has often been justified by the promise of greater productivity, lower costs, and improved living standards. AI may accelerate all three. But productivity alone does not tell us whether the benefits will be broadly shared or whether people will retain economic security as their work changes.
If machines can perform an increasing proportion of the work required to produce goods and services, society may eventually need to loosen the connection between employment and access to economic output. That does not mean work will disappear, or that every human contribution will become worthless. It means that employment may no longer be a sufficient mechanism for distributing the benefits of production if the demand for human labour falls substantially.
The task is not to reject AI in order to preserve every existing job. Nor is it to assume that the market will solve every problem created by technological change. We must consider how ownership, income, education, and social institutions should evolve alongside the technology.
The horse moment for humans is not simply the fear that machines will replace us. It is the possibility that we will build systems capable of producing extraordinary abundance without adequately deciding how that abundance should benefit the people who created them.
We have an advantage horses never possessed: we can anticipate the consequences, debate the choices, and change the rules. Whether we use that advantage wisely remains an open question.
The real challenge is not to make humans compete with AI at every task. It is to build an economy in which humans can flourish even when the market needs less of their labour.