What a person becomes able to do
Domains describe what there is to understand. Capabilities describe who a person becomes. They are developed across all ten domains, each most directly in a few.
- A · Historical fact
- B · Empirical evidence
- C · Philosophical proposition
- D · Future scenario
The distinction matters. A curriculum of domains alone produces people who know about many things. A curriculum aimed at capabilities produces people who can judge, create, care and act. The capability approach of Sen and Nussbaum makes a parallel move in development economics: judge progress by what people are actually able to do and be [28].
The objective is not to train people to compete with AI at every task. It is to form people who keep their agency in a world where AI may exceed their abilities in many cognitive domains.
Independent judgment
Forming one's own considered view and acting on it, rather than deferring by default to authority, crowds or machines.
In the age of AI. The more capable the system, the more tempting deference becomes. Independent judgment is not refusal to use AI; it is keeping the final assessment, and the responsibility, human.
Critical reasoning
Analysing arguments, evidence and assumptions; recognising fallacies, including one's own.
In the age of AI. Fluent machine text can be wrong in ways that sound right. Critical reasoning is the first line of verification.
Epistemic humility
Calibrating confidence to evidence, recognising the limits of one's knowledge and changing one's mind for good reasons.
In the age of AI. Applies twice: to our own beliefs and to machine outputs, which often express uniform confidence regardless of reliability.
Creativity
Bringing something new and valuable into being, through play, discipline and the willingness to fail.
In the age of AI. Recombination is cheap. Choosing what is worth making, and making it one's own, is not.
Ethical decision-making
Reasoning about right action under uncertainty and disagreement, and taking responsibility for the result.
In the age of AI. Decisions made with or by AI still have human authors. Someone must answer for them.
Communication
Expressing ideas clearly and listening well, across languages, disciplines and cultures.
In the age of AI. Instructing machines is a form of communication; so is explaining their outputs to people affected by them.
Collaboration
Working with others towards shared ends: trust, coordination, disagreement handled well.
In the age of AI. Teams will include people and agents. Collaboration now includes delegating to, coordinating and checking non-human contributors.
Physical and emotional self-development
Caring for one's body, attention and emotional life; building the resilience to live well.
In the age of AI. If work no longer structures the day, self-regulation, health and purpose must be cultivated deliberately; Marienthal is the warning.
Practical agency
The capacity to make things happen in the world: start, organise, build, repair and persist.
In the age of AI. The risk of automation is not only job loss but learned helplessness. Agency is the habit of acting rather than waiting.
Metacognition and lifelong learning
Knowing how one learns, monitoring one's understanding and continuing to learn throughout life.
In the age of AI. AI tutors can accelerate learning or replace it. Metacognition is what tells the difference.
Transversal literacy
AI fluency
A transversal literacy that runs through every domain and capability: the knowledge, skills and dispositions to work with increasingly capable AI systems while retaining human agency.
- Capabilities and limitations. What current systems can and cannot do, and how fast that is changing (time horizons, benchmarks, their caveats).
- Working with AI. Using models and agents as tools, tutors and collaborators across domains.
- Delegation. Deciding what to hand over, specifying it well, and what to keep.
- Evaluating outputs. Checking claims, code and analyses against sources, tests and domain knowledge.
- Supervising agents. Setting goals, monitoring long-running autonomous work and intervening in time.
- Failure modes. Hallucination, reward hacking, automation bias, brittleness outside training conditions.
- Independent judgment. Keeping one's own skills in practice so that oversight remains possible.
- Human-AI collaboration. Designing workflows in which people and machines each do what they do best.
- Technological governance. Understanding who builds, owns and regulates AI, and how citizens can take part.
Research on human-automation interaction gives reasons to take this literacy seriously. People over-rely on automated aids, experts included, and training alone only partly corrects it [50]. Highly automated systems leave supervisors out of practice exactly when they are needed [49]. Early AI-literacy research defined competencies for evaluating and collaborating with AI before today's agents existed [51]. METR's 2026 results show models weakest where feedback is slow or ambiguous [43], which is where human oversight is hardest.
The capability matrix
Where each capability is cultivated most directly. All capabilities are exercised in all domains; the marks show emphasis, not exclusivity.
| Capability | LOGOS | RATIO | MATHEMATICA | PHYSICA | BIOS | PRAXIS | ARS | CIVITAS | ETHOS | EPISTEME |
|---|---|---|---|---|---|---|---|---|---|---|
| Independent judgment | — | — | — | — | — | ● | — | — | ● | ● |
| Critical reasoning | ● | ● | — | — | — | — | — | — | — | ● |
| Epistemic humility | — | — | ● | — | — | — | — | — | ● | ● |
| Creativity | — | — | ● | — | — | ● | ● | — | — | — |
| Ethical decision-making | — | — | — | — | ● | — | — | ● | ● | — |
| Communication | ● | — | — | — | — | — | ● | ● | — | — |
| Collaboration | ● | — | — | — | — | ● | — | ● | — | — |
| Physical and emotional self-development | — | — | — | — | ● | ● | — | — | ● | — |
| Practical agency | — | — | — | ● | — | ● | — | ● | — | — |
| Metacognition and lifelong learning | — | ● | — | — | ● | — | — | — | — | ● |
| AI fluency | — | ● | — | — | — | ● | — | ● | ● | ● |