Schools keep searching for skills that machines cannot replace. That strategy makes machine capability the reference point for human development. Each advance in AI then forces schools to revise what they consider distinctively human.
A stronger basis for human advantage is personal knowledge: what people develop through experience, judgment, relationships, context, and purpose, then apply where the problem, pathway, or value to be created has not been fully specified.
Researchers at Stanford’s Digital Economy Lab, using payroll data covering millions of U.S. workers through June 2026, found no evidence of widespread AI-driven job displacement across the economy. They did find a sharp difference by age and occupational exposure. Among workers ages 22 to 25 in occupations highly exposed to AI, employment stood 19% below where it would have been had it kept pace with employment among young people in less-exposed occupations. Experienced workers in the same fields showed no comparable gap. The divergence has widened since 2025 and appears mainly in reduced hiring rather than increased firing (Brynjolfsson, Chandar, & Chen, 2026).
The researchers caution that these are descriptive findings and do not establish causation. Even so, the pattern appears precisely at the transition formal education is supposed to prepare people for: entry into professional work.
For generations, that transition followed a recognizable structure. People studied a field, earned a credential, entered near the bottom of an occupation, and worked their way to retirement. Junior accountants reconciled accounts before making consequential financial judgments. Young lawyers researched precedent and drafted documents before advising clients. Analysts assembled reports before shaping strategy. Programmers wrote routine code before designing systems.
Those entry-level roles did more than provide employment. They placed novices inside communities of practice. Lave and Wenger (1991) described learning as movement from peripheral toward fuller participation in such communities. Much professional expertise develops this way, through participation in work alongside people who already know how to do it. But now that old contract has expired.
AI is increasingly capable of performing many of the codified tasks assigned to early-career workers. The International Labour Organization estimates that one in four jobs worldwide has some exposure to generative AI. Clerical occupations remain the most exposed, while exposure is growing in highly digitized professional and technical fields. The ILO expects transformation of occupations to be more common than outright elimination because most jobs still contain tasks requiring human participation (Gmyrek et al., 2025).
Transformation can still reduce opportunities for beginners. An occupation can survive while employing fewer novices. A firm can retain experienced accountants, programmers, analysts, consultants, and managers while automating enough junior work to hire fewer people beneath them.
For education, the problem is twofold. Graduates may find fewer conventional points of entry into professional work, while organizations may weaken the process through which novices become experienced practitioners. More fundamentally, the career ladder itself may no longer describe how working life is organized.
The career ladder (as we know it) has toppled
The career ladder assumes an organizational model built around stable divisions of labor. Jobs occupy boxes, boxes belong to departments, authority follows reporting lines, and experience accumulates through increasingly senior positions.
Education developed alongside this model. Subjects became disciplines, disciplines became degrees, degrees mapped onto occupations, and occupations offered recognizable pathways upward. The correspondence was never exact, but it was stable enough to organize institutions and expectations.
AI makes it increasingly possible to separate tasks from the jobs that once contained them. Research, drafting, analysis, coding, scheduling, documentation, monitoring, coordination, and routine decision support can increasingly be handled independently. Some tasks can be automated; others can be handled by fewer people working with machines. Teams can form around problems and projects without reproducing every layer of the hierarchy that previously organized the same work.
Managerial work can be separated into tasks as well. Algorithmic systems already allocate tasks, monitor workers, evaluate performance, and exercise forms of organizational control once associated primarily with human supervisors (Kellogg, Valentine, & Christin, 2020). More capable systems extend these functions into synthesis, planning, coordination, and decision support. A manager using such systems may be able to oversee work that once required several layers of administrative support.
Partial automation can therefore reshape hierarchy without eliminating occupations. Firms can widen spans of responsibility, remove layers of coordination, and assemble work through changing combinations of employees, specialists, contractors, and machines. For workers, creating value may depend increasingly on moving among problems, projects, technologies, domains, and relationships rather than advancing through a predictable sequence of positions.
Despite knowing better, schools remain heavily invested in preparing people for the boxes.
The wrong human advantage
Much of the educational response to AI has focused on supposedly durable human capacities: creativity, empathy, critical thinking, collaboration, leadership, communication, and so forth.
These capacities matter in their own right. They are a weak basis for an AI strategy when their value depends on machines being unable to perform them. Machine capabilities are changing too quickly for curricula to rest on their current limits. A capability treated as distinctively human today may be replicated or approximated by machines tomorrow, requiring schools to draw the boundary again.
A human advantage defined by machine limitations belongs to the machine.
The mistake is to treat human advantage as a residual category: whatever remains after automation. That leaves the structure of work intact and asks where people can still fit inside it.
The more useful question is what new value a person can create with what is known. Answering it requires a more careful account of knowledge itself.
From information to new value
AI systems are extraordinarily capable data and information systems. They can process vast bodies of recorded material, identify patterns, recombine representations, produce analyses, generate alternatives, and increasingly act on the information they process. Their outputs can resemble products of human knowledge with remarkable fidelity.
I use knowledge more narrowly. Michael Polanyi argued that human knowing exceeds what a person can state explicitly. Skill, recognition, judgment, practice, and experience contribute to what people know even when they cannot translate that knowledge into a complete set of rules (Polanyi, 1966).
David Autor later described the automation problem this creates as Polanyi’s paradox: people routinely perform tasks whose governing rules they cannot fully articulate, historically making many forms of work difficult to automate (Autor, 2015).
Ikujiro Nonaka placed the relationship between tacit and explicit knowledge at the center of organizational knowledge creation. In his account, new knowledge develops through their interaction. Individuals create knowledge through experience and interpretation, while organizations provide structures through which it can be articulated, shared, and extended (Nonaka, 1994).
AI now reaches into forms of work that once appeared resistant to automation because their rules were difficult to specify. That makes the distinction between information and personal knowledge more consequential.
A system can process a vast literature about teaching without having taught a particular child. It can model an organization from its documents without having spent years learning why formal procedures diverge from actual practice, which relationships hold the institution together, or why an apparently sensible decision will fail in that setting.
Human knowledge develops through participation. Experience accumulates through memory, relationships, culture, judgment, mistakes, commitments, interests, and purpose. Some of this can be articulated; some becomes visible only when a situation calls for it. I use personal knowledge to describe this integration of information with experience and judgment in a particular person.
AI also changes the economics of information because it changes what is scarce. The classical economic logic associated with Smith still applies: scarcity matters. What changes is the scarce input. Retrieval, synthesis, comparison, explanation, translation, and recombination can increasingly be performed at very low marginal cost. Codified information therefore carries less scarcity value on its own, while judgment, context, trust, experience, access, ownership, and the capacity to decide what should be done with information become relatively more important (Smith, 1776/1976).
Formal education developed when access to information was far more constrained. Schools and universities were important gateways to books, experts, laboratories, archives, credentials, and specialized bodies of knowledge. Much of schooling consequently became organized around acquiring, retaining, organizing, and demonstrating command of information.
As information becomes abundant, the educational case for organizing learning primarily around its acquisition and reproduction weakens. The educational question becomes what a person can make of what is known.
A useful progression runs from data to information, from information to knowledge, and from knowledge to innovation. Data acquire significance through organization and interpretation. Information becomes personal knowledge as a person integrates it with experience, context, judgment, and purpose. Innovation occurs through application: knowledge brought to bear on a situation in a way that creates new value.
That value can be commercial, scientific, civic, artistic, educational, technological, or social: a scientific insight, a business, a better public service, a work of art, a community institution, a teaching practice, a technical solution, or a reframing of a problem that allows others to act.
Innovation begins when someone sees a possibility that the available information does not specify on its own. The accepted problem may have been framed incorrectly. Knowledge from one domain may prove useful in another. A connection between people may matter more than another analysis. A technically efficient solution may fail because it ignores culture, trust, history, or human motivation.
This is the human advantage schools should develop: the capacity to apply personal knowledge where the problem, pathway, or value to be created has not already been defined.
The knowmad revival
I began developing the concept of the knowmad in the late 2000s. By 2008, I was using the term for a “nomadic knowledge worker”: a person able to work across changing social, organizational, and technological contexts (Moravec, 2008). I developed the concept further in the coauthored Aprendizaje Invisible (Cobo & Moravec, 2011) and in “Knowmad Society: The ‘New’ Work and Education” (Moravec, 2013), emphasizing personal knowledge, contextual mobility, networks, innovation, and changing forms of work.
Physical mobility was never the defining feature. Knowmadic mobility is contextual: the capacity to carry knowledge, judgment, relationships, and ways of working into new situations and recombine them there. Knowmads connect knowledge, people, and ideas across contexts to create value where established roles or procedures provide little guidance.
The conditions the concept was meant to address are becoming more pronounced as AI takes on more codified work and organizational structures become less predictable. Knowmadic capacity includes learning across contexts, framing problems, exercising judgment, connecting people and domains, working fluently with technologies, and recognizing opportunities to contribute. Its value becomes especially apparent when no predefined occupational script explains what comes next.
A knowmad can extend that capacity with AI, using machines to search, model, calculate, translate, simulate, automate routine tasks, and test possibilities at a scale no individual could manage alone.
Knowmadic capacity matters most where established roles and procedures provide less guidance: in noticing what matters, framing problems, connecting contexts, exercising judgment, and deciding what is worth creating.
Educating for each person
Schools do not need to predict the occupations of 2035 or identify a permanent catalogue of uniquely human skills. They need to know their students well enough to help them develop what is distinctive in their knowledge, interests, judgment, and capacity to contribute.
Common curricula, schedules, subjects, assessments, and credentials solve real problems of scale and public access. Shared knowledge matters. Disciplines matter. Standards can protect learners from arbitrary expectations and unequal access to powerful knowledge. Standardization becomes distorting when common outcomes come to define the purpose of education.
Every learner brings prior experience, interests, relationships, abilities, cultural knowledge, questions, and ways of seeing. Personal knowledge requires substance, so schools should bring learners into serious contact with mathematics, science, history, language, art, philosophy, technology, craft, communities, and people whose knowledge differs from their own.
This does not eliminate direct instruction, practice, or repetition. Novices cannot exercise informed judgment in a domain they barely understand. Disciplinary knowledge, technical fluency, and repeated practice provide the material from which personal knowledge develops. The difference lies in refusing to treat mastery of that material as the endpoint.
Students also need to use what they learn. They need sustained opportunities to work on problems whose solutions are unknown, undertake projects with consequences beyond a grade, encounter expertise outside the classroom, make judgments under uncertainty, and discover connections that were never written into the syllabus.
If routine entry-level work no longer provides the same apprenticeship into professional practice, schools and employers will need to create other ways for novices to acquire judgment. That means sustained participation in consequential work: supervised practicums, simulations in which decisions have visible consequences, work alongside experienced practitioners, and repeated opportunities to make decisions, receive feedback, and try again. Tacit knowledge cannot simply be delivered as content. It develops through experience, observation, correction, and increasingly responsible participation.
Teachers are central to this work. They can recognize an emerging interest, expose a weak assumption, introduce a student to a field or person, demand greater intellectual rigor, and help learners interpret what experience has taught them.
This kind of education is labor intensive because it requires sustained attention to where each learner is developing and what might help that person go further. Standardized systems have often treated such attention as a problem to be engineered away. As AI takes on more standardized cognitive production, it becomes more valuable.
None of this makes the problem of scale disappear. It changes what schools should try to scale. A system cannot mass-produce intimate relationships or personal judgment, but it can organize smaller advisory structures, shared project infrastructures, community partnerships, common assessment protocols, and technologies that reduce administrative work so that more human attention is available where it matters. Scale is an institutional design problem, not a reason to define education around what is easiest to standardize.
Table 1. What changes when schools take personal knowledge seriously.
| Educational dimension | The prevailing wager | A knowmadic wager | What changes in practice |
|---|---|---|---|
| Purpose | Prepare students to demonstrate specified knowledge and capabilities. | Develop people who can apply personal knowledge to create value in changing contexts. | Students are expected to use what they know to make, change, solve, contribute, or create. |
| Curriculum | Determine in advance what students should know and sequence it into courses. | Build strong disciplinary foundations while helping students connect knowledge across contexts. | Sustained work crosses disciplinary boundaries when the problem requires it. |
| Problems | Give students problems whose form, relevant information, and expected solution are largely known. | Develop the ability to notice, frame, and investigate problems that are not yet well specified. | Students increasingly define questions, constraints, methods, and criteria for success. |
| Knowledge | Treat mastery of established information as the principal evidence of learning. | Integrate explicit information with experience, judgment, relationships, and context. | Practice, field experience, dialogue, reflection, making, and participation contribute to what students know. |
| Application | Use exercises primarily to demonstrate mastery. | Apply personal knowledge where the outcome matters beyond demonstrating mastery. | Students work with real organizations, communities, audiences, problems, and consequences. |
| Teacher | Deliver, explain, sequence, and assess prescribed learning. | Understand learners well enough to challenge, connect, guide, and extend their development. | Advising, mentoring, feedback, introductions, and individualized intellectual challenge become central teaching work. |
| Learner | Progress through a common sequence toward predetermined outcomes. | Build a distinctive body of personal knowledge and learn to apply it across contexts. | Students exercise increasing authority over questions, projects, collaborators, tools, and directions. |
| Relationships | Organize learning primarily around relationships inside the institution. | Treat networks and communities as part of knowledge development. | Learners work with practitioners, experts, communities, mentors, and collaborators beyond the classroom. |
| AI | Use AI to improve efficiency or performance on existing educational tasks. | Use AI as a powerful data and information resource within human knowledge work. | Students use machines for search, synthesis, simulation, analysis, and production while remaining responsible for judgment, purpose, and application. |
| Assessment | Ask whether the learner met a predetermined standard. | Examine what the learner can do with knowledge when the route is not predetermined. | Common measures are complemented by investigations, portfolios, prototypes, performances, consequential projects, and defenses. |
| Success | Produce graduates qualified for identifiable next steps. | Develop people capable of creating useful next steps when established ones are unavailable. | Evidence of judgment, initiative, contextual transfer, contribution, and value creation matters alongside credentials. |
Personal knowledge does not make assessment impossible at scale. It changes what should be standardized. Schools can hold students to common expectations for evidence, reasoning, disciplinary command, judgment, and quality while allowing the problems, contexts, and products through which those capacities are demonstrated to vary. Moderated scoring, common rubrics, external review, portfolio sampling, and public defenses can provide comparability without requiring every learner to perform the same task. Equity still requires common standards; it does not require identical demonstrations of learning.
The harder equity problem lies upstream of assessment. Students do not enter school with equal access to professional networks, mentors, technologies, cultural capital, or consequential opportunities. If schools make contextual learning central, they also assume responsibility for providing those opportunities rather than treating them as advantages students are expected to bring from home. Personalization without such public provision would reproduce precisely the inequalities it is meant to overcome.
The limits of knowmadic capacity
Knowmadic capacity does not protect people from economic disruption. A resourceful, knowledgeable, creative, well-connected person can still face concentrated ownership, weak labor protections, discrimination, inaccessible technology, or an economy that distributes the gains from automation badly. Creating value does not guarantee receiving a fair share of it.
Education can develop capacity; it cannot determine how power, security, and opportunity are distributed. A person’s ability to act under uncertainty depends partly on the political and institutional conditions surrounding that action. Portable rights, public infrastructure, social protection, bargaining power, access to continued learning, open systems, and authority over consequential technologies shape whether fluid forms of work expand agency or intensify precarity.
Adaptability can easily become a way of transferring risk. Workers are asked to reskill when technology changes. Students are told to prepare for occupations institutions expect to exist. When those expectations fail, individuals are asked to adapt again.
Knowmadic learning cannot mean continual accommodation to changing labor markets. Its purpose is to develop the capacity to create meaningful possibilities when established structures provide less direction. For educators, that changes the governing questions: What do you want to bring into the world? What do you need to learn, build, or connect to make it possible?
A better wager
The Stanford findings may prove temporary rather than marking a durable transformation of employment. AI may generate new categories of entry-level work, and employers may discover that reducing junior positions damages their future supply of expertise. The shape of the labor market remains uncertain.
Schools already have enough reason to reconsider their wager. Explicit information is becoming abundant and cheap to manipulate. More codified cognitive tasks are becoming available for automation. The connection between a degree, an entry-level position, and a predictable career is becoming less reliable.
Deep knowledge remains indispensable. Students need science, mathematics, history, language, craft, technical understanding, and access to the accumulated intellectual achievements of others. They also need the experiences through which that knowledge becomes personal: opportunities to connect it with lived situations, test it against reality, carry it between contexts, combine it with other forms of understanding, and apply it to create something of value.
Educating for “what machines can’t do” makes human development dependent on a moving technical boundary. A better approach is to help each person develop deep personal knowledge and the capacity to use it where the answer, the pathway, and sometimes the problem itself remain open.
References
Autor, D. H. (2015). Why are there still so many jobs? The history and future of workplace automation. Journal of Economic Perspectives, 29(3), 3–30. https://doi.org/10.1257/jep.29.3.3
Brynjolfsson, E., Chandar, B., & Chen, R. (2026). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence (revised August 2026). Stanford Digital Economy Lab.
Cobo, C., & Moravec, J. W. (2011). Aprendizaje invisible: Hacia una nueva ecología de la educación. Laboratori de Mitjans Interactius / Publicacions i Edicions de la Universitat de Barcelona.
Gmyrek, P., Berg, J., Kamiński, K., Konopczyński, F., Ładna, A., Nafradi, B., Rosłaniec, K., & Troszyński, M. (2025). Generative AI and jobs: A refined global index of occupational exposure (ILO Working Paper No. 140). International Labour Organization. https://doi.org/10.54394/HETP0387
Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366–410. https://doi.org/10.5465/annals.2018.0174
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge University Press. https://doi.org/10.1017/CBO9780511815355
Moravec, J. W. (2008, November 20). Knowmads in Society 3.0. Education Futures. https://educationfutures.com/post/knowmads-in-society-30/
Moravec, J. W. (2013). Knowmad society: The “new” work and education. On the Horizon, 21(2), 79–83. https://doi.org/10.1108/10748121311322978
Nonaka, I. (1994). A dynamic theory of organizational knowledge creation. Organization Science, 5(1), 14–37. https://doi.org/10.1287/orsc.5.1.14
Polanyi, M. (1966). The tacit dimension. Doubleday.
Smith, A. (1976). An inquiry into the nature and causes of the wealth of nations (R. H. Campbell & A. S. Skinner, Eds.). Clarendon Press. (Original work published 1776)
