From Educated to Capable

For most of human history, intelligence was scarce.

A knowledgeable teacher was scarce. A good tutor was scarcer still. Manuscripts were expensive, universities were geographically concentrated, and specialized expertise was gatekept. A child born in the wrong village, family, or country might live an entire life without encountering someone capable of teaching her calculus, genetics, computer science, philosophy, or Mandarin.

Modern education systems were constructed in response to this scarcity. Their great achievement was to industrialize access to knowledge. The classroom allowed one teacher to instruct thirty students; the printing press allowed textbooks to reach millions; examinations provided a scalable mechanism for determining what students had learned.

This architecture transformed humanity. But it was built for a world in which intelligence was expensive. We are now entering a world in which it may become abundant.

We are no longer an economy that produces and implements static information, but one that can produce living, responsive cognition. Artificial intelligence can already explain difficult concepts, translate languages, write software, analyze data, generate images, critique arguments, conduct research and provide individualized instruction. Increasingly agentic systems can move beyond answering questions toward completing extended sequences of cognitive work. The trajectory suggests something historically unprecedented: many forms of expertise that were once scarce may become available, on demand, to billions of people. In other words, ours is no longer a knowledge economy, but a cognition economy.

Education policy must therefore confront a deeper question:

What should education accomplish when intelligence itself is no longer scarce?

The answer is not as simple as adding an AI tutor to our education system, because AI requires human beings to target a different set of skills than the ones being taught now.

The architecture we inherited

Much of modern education rests, implicitly, on a simple sequence: instruction → mastery → credential → employment.

Students receive instruction. Institutions measure whether they have mastered a body of knowledge. Credentials certify that mastery. Employers use those credentials as signals of competence.

This model made considerable sense in the industrial and information economies of the twentieth century. A competent accountant needed to perform calculations. A programmer needed to write code. An analyst needed to research and synthesize information. A lawyer needed to draft documents. A student demonstrated readiness for these occupations partly by proving that he or she could independently perform the cognitive tasks the occupation required.

Artificial intelligence destabilizes every link in this chain.

Knowledge transmission is becoming extraordinarily inexpensive. Cognitive production is becoming cheaper as machines perform increasingly sophisticated writing, calculation, research, coding and analysis. And credentials based primarily on examinations become less informative when machines can perform many of the tasks those examinations were designed to test.

The danger is that education responds to this transformation too narrowly. We could digitize the textbook, personalize the worksheet, automate the tutor and congratulate ourselves for modernizing education—while leaving its underlying objective unchanged.

That would be analogous to attaching an electric motor to a horse carriage and calling it the future of transportation.

The deeper opportunity is to redesign the system around the capabilities humans need when intelligence is abundant.

From knowledge acquisition to human agency

The central objective of education in the age of artificial intelligence should be human agency.

This definition retains knowledge and mastery. A person cannot reason well about a world they do not understand. Foundational literacy, numeracy, scientific reasoning, historical knowledge and cultural understanding remain essential.

But knowledge becomes a foundation rather than the destination.

Consider two students in 2040.

Our existing assessment systems would probably favor the first, but the economy and society of 2040 may depend increasingly on the capabilities of the second.

This does not imply that independent cognition is obsolete. Quite the opposite. The more capable our machines become, the more important it may be that humans retain sufficient understanding to question them.

A pilot does not become safer because the autopilot becomes better if she loses the ability to recognize when the aircraft is heading toward a mountain.

Education therefore needs to measure two forms of capability simultaneously — and the educational challenge of the coming decades is to maximize both.

The paradox of abundant intelligence

Artificial intelligence creates an unusual paradox.

A machine can optimize an objective. It cannot relieve society of the responsibility to choose the objective. This distinction matters profoundly for education.

The twentieth-century classroom often rewarded students for producing correct answers to questions selected by someone else. The examination hall represents this model in its purest form: sit alone, use no outside assistance, answer predefined questions, and reproduce the correct knowledge under time pressure.

But much of adult life operates in precisely the opposite direction. The difficult problems are frequently not given to us. They must first be recognized and prioritized.

There is no examination paper announcing that a neighborhood needs a better transportation system, that a scientific assumption is wrong, that a business opportunity exists, that an institution is failing, or that a community faces an emerging danger.

Someone must recognize the problem before anyone can solve it.

Education in an age of abundant intelligence should therefore cultivate not merely problem-solving, but problem-finding: curiosity, observation, moral imagination and the capacity to distinguish consequential problems from trivial ones.

Measuring capability rather than schooling

This transition also exposes a weakness in the way governments measure educational progress. We are exceptionally good at administering metrics to measure how well an education system or program is doing.

These measures matter. Access remains profoundly unequal, and hundreds of millions of children still lack foundational literacy and numeracy.

But enrollment is an input, not an outcome. The increasing number of years spent sitting in classrooms does not translate into increasingly capable students.

Recent research is already moving to address this distinction. OpenAI's Learning Outcomes Measurement Suite, developed with researchers at the University of Tartu and Stanford's SCALE Initiative, explicitly argues that short-term test scores alone cannot capture the effects of AI-assisted learning. Its proposed longitudinal framework examines dimensions including critical thinking, creativity, memory, persistence, motivation and metacognition.

Meanwhile, evidence from the workplace illustrates why broader capability matters. Anthropic's 2026 Economic Index finds increasingly agentic forms of AI use in which people delegate substantial portions of cognitive tasks to machines. Its survey research also found that experienced workers frequently identify judgment, contextual knowledge, situational reasoning, trust and interpersonal capability as difficult to delegate—although Anthropic appropriately cautions that its survey of Claude users is not representative of the overall workforce.

A 2026 experiment involving more than 1,000 students offers another useful clue. Researchers found that access to ChatGPT improved the quality and coherence of students' work, while explicit causal-reasoning training increased the uniqueness of ideas; students receiving both interventions demonstrated both effects.

The implication is not that AI replaces thinking, but that machine capability and human capability can be complementary when education deliberately cultivates both.

The unit of educational success should therefore gradually move beyond classroom hours toward capability acquired.

A new metric: Time to Capability

One way to make this transformation measurable is through a concept we might call Time to Capability, or TTC. For any defined capability, ask:

How long does it take the median person, beginning without the relevant expertise, to become capable of accomplishing a meaningful real-world objective safely and reliably?

Foreign language acquisition has typically required hundreds of hours of standardized coursework. Now, AI tutors offering unlimited conversation, instant correction and adaptive exercises might dramatically reduce that time.

Or consider software development. What once required years of specialized training may increasingly become accessible to people who understand computational concepts and can effectively direct coding agents. Similar transformations could occur in entrepreneurship, engineering, scientific research, design, financial literacy and vocational training.

The abundance produced by education could therefore be understood partly as time compression.

This reframing has important consequences for development. Imagine a young woman in a rural community who wants to install a solar microgrid, diagnose a crop disease, build an online business or learn biotechnology.

The relevant question is no longer merely whether she can enroll in a university. It is: How far is she, in time and resources, from capability?

If the answer falls from four years to four months, or from four months to four weeks, something profound has occurred even if no traditional education statistic captures it.

Distance from capability may become as important a development indicator as distance from a school.

From the education ladder to the capability network

The traditional education system resembles a ladder. Primary school leads to secondary school. Secondary school leads to university. University leads to employment. Advancement occurs through predetermined stages, usually organized by age and measured in years.

AI makes another architecture possible: a capability network.

Learning becomes less like climbing a staircase and more like accessing electricity: available when needed, adapted to demand and increasingly independent of geography. This changes the meaning of lifelong learning. It ceases to be a slogan about adult education and becomes an infrastructure layer of society.

What remains scarce when intelligence becomes abundant?

Perhaps the most important question for educational policy is not what AI makes abundant, but what remains scarce.

An AI may help a person achieve almost any articulated objective. Education must help people confront the harder question: Which objectives are worth pursuing?

This introduces dimensions that industrial education systems have often treated as peripheral: ethics, civic responsibility, self-knowledge, relationships, physical health, resilience, meaning and service to others.

These are not ornamental additions to education after the "real" subjects have been taught. In a world of powerful machines, they may increasingly constitute the core.

The danger of cognitive dependency

None of this justifies technological triumphalism. An education system built around AI could fail spectacularly.

And personalization itself contains a paradox. Education is partly about discovering what we already like. But it is also about encountering what we would never have chosen.

A perfectly personalized intellectual environment risks becoming an epistemic cocoon.

Great teachers do not merely satisfy curiosity. Sometimes they create it. A child may not request astronomy before seeing Saturn through a telescope, philosophy before encountering an impossible question, or poetry before hearing a line that changes how language feels.

The role of educational AI must therefore not be merely to predict what a learner wants next. It should expand the frontier of what the learner can imagine wanting.

The objective is not frictionless education. Some friction is educational. Struggle builds persistence. Confusion can precede understanding. Debate exposes weak reasoning. Working with difficult people teaches cooperation. Physical experimentation teaches realities that simulation cannot fully reproduce.

Toward an Abundance Index for education

An Education Domain designed for the age of artificial intelligence should consequently measure more than enrollment, attainment or academic mastery. It should ask whether every person has access to the conditions necessary to become progressively more capable.

This question also gives educational abundance a powerful definition.

Educational abundance arrives only with the last of these.

A new social contract for education

The most consequential promise of artificial intelligence may ultimately be distributive.

For centuries, the quality of a child's education has been strongly influenced by accidents of birth: geography, parental income, language, disability, school quality and access to talented adults.

Personal tutoring illustrates the inequality particularly clearly. The educational experience historically available to princes, aristocrats and wealthy families—a knowledgeable adult responding continuously to the needs of one learner—has never been economically scalable.

AI could change this. But access to a model is not equivalent to educational abundance.

A child requires connectivity, devices, language support, safe environments, motivation and opportunities to act on what she learns. Education remains social and embodied. Teachers, families, peers, communities and institutions still matter.

Intelligence is not the destination

The twentieth century built institutions capable of distributing knowledge at unprecedented scale. The twenty-first may distribute intelligence.

But intelligence itself is not the destination.

A civilization could possess extraordinary machine intelligence while its citizens become passive, dependent and directionless. Equally, abundant intelligence could enable billions of people to become more creative, scientifically capable, entrepreneurial and empowered than any previous generation.

Technology alone will not determine which future emerges. Education will.

For the first time, humanity may be approaching a world in which almost every child can have access to something resembling a polymath tutor: patient, multilingual, adaptive and available at any hour. That is an extraordinary achievement.

But the greatest educational opportunity is not to create machines that can answer every child's questions.

It is to create a generation of humans capable of asking better ones.

The defining measure of education in the age of artificial intelligence should therefore no longer be simply what a person knows. It should be what a person can understand, what they can become, what they can create with the intelligence available to them—and whether they possess the judgment to decide what is worth creating at all.