She knows the policy.
She helped write it.
The organization has agreed that consequential AI-supported decisions must be independently checked. The wording is clear. The examples are sensible. She has explained the principle to colleagues often enough that the sentences arrive without effort.
Then a senior executive forwards an AI-generated analysis ten minutes before a meeting and writes, “Looks good. Can we move?”
The sources are not attached.
One claim feels wrong.
Everyone else is already discussing implementation.
The policy is still in her head.
What is missing is not knowledge of the rule.
The difficult part is making the rule available as action inside status, urgency and social momentum.
That difference is one reason the Institute distinguishes knowledge from capacity.
Definition
Knowledge refers broadly to what a person knows: concepts, facts, relationships, procedures, explanations and understandings that can be represented, recalled or used.
Capacity, as used in this Knowledge Library, refers to an ability becoming sufficiently available, reliable and adaptable to be used in the conditions where it matters.
The distinction is operational rather than absolute.
Knowledge is part of many capacities.
Capacity often depends on knowledge.
The point is not that knowledge is passive and capacity is noble.
The point is that knowing something and being able to use it are different claims and should be measured differently.
Education frameworks already resist reducing competence to knowledge alone
The OECD Learning Compass 2030 provides a useful public example.
Its competence framing includes knowledge, skills, attitudes and values rather than treating knowledge as the entire educational outcome. The OECD describes knowledge as including theoretical concepts as well as practical understanding developed through experience, and skills as the ability and capacity to carry out processes and use knowledge responsibly toward goals.
This is important for terminology.
The OECD does not use “capacity” in exactly the specialized way the Institute does on this page. In some policy frameworks, capacity and skill may overlap linguistically.
What the framework does support is the broader proposition that competent action cannot be reduced to possessing information.
Knowledge, skill, attitude, value, agency and context interact.
Knowledge is not merely facts
Public discussions often treat “knowledge” as memorized information and “capacity” as something deeper.
That is too crude.
Knowledge can be rich, procedural, embodied in practice and organized through expertise.
A surgeon's knowledge is not a list of anatomical facts. A historian's knowledge includes patterns of evidence, source criticism and contextual relations. A designer's knowledge includes materials, precedents, constraints and ways of seeing. A skilled therapist's conceptual knowledge may include years of case-based distinctions.
Knowledge changes perception.
Experts often notice different features of a situation because they know what those features mean.
Capacity without relevant knowledge can become confident improvisation.
The distinction should therefore protect knowledge, not demote it.
Why knowing still fails to become action
Several things can happen between understanding and use.
Retrieval fails
A principle that was easy to recognize during training may not be retrieved when the real situation looks different.
The cue is missed
The person knows what to do if they recognize the category, but they do not recognize that this situation belongs to it.
Competing goals take over
They know verification matters, but speed, belonging, status, avoidance or convenience becomes more salient in the moment.
State changes access
Stress, fatigue, emotion or time pressure can alter what information receives attention and what responses are easy to initiate.
Social conditions make use costly
A person may know the correct procedure and lack the authority to insist on it.
Support was doing more work than expected
The learner performed well because prompts, examples or a facilitator structured the task. When those supports disappear, so does the performance.
These are different failure modes.
Calling all of them “lack of knowledge” hides the design problem.
Learning and performance cannot be collapsed
Soderstrom and Bjork's review gives this distinction a strong empirical foundation.
Learning concerns relatively durable changes that support later retention and transfer. Performance is what can be observed during practice or instruction.
The two can dissociate.
A learner may perform well with cues that make the task easy while retaining little later. Another may struggle during practice in ways that produce better long-term retrieval.
This matters because tests of knowledge are often administered under conditions that make the relevant cue obvious.
“Which principle applies here?”
Real life more often asks:
“What is happening?”
No category label is supplied.
Capacity includes the ability to detect that the principle is relevant before applying what is known.
Transfer is the distance between the lesson and the world
Blume and colleagues' meta-analysis of training transfer shows that learned behavior does not simply travel from a programme into work by default.
Transfer varies with learner characteristics, training design and the work environment.
The finding is particularly important for open skills that require adaptation.
Knowing a conflict framework in a workshop is not the same as using it with a defensive colleague.
Knowing a bias definition is not the same as noticing bias in one's own favored interpretation.
Knowing an AI policy is not the same as interrupting a prestigious project when evidence is weak.
The transfer problem is where knowledge claims often become capacity claims without sufficient evidence.
Knowledge can be inert without being useless
Learning researchers have long used terms such as “inert knowledge” for information that a learner possesses but does not spontaneously apply in relevant situations.
The existence of inert knowledge should not be interpreted as proof that instruction failed entirely.
Stored understanding may become useful with better cues, practice, context or experience.
The important question is what outcome was intended.
If the goal is examination performance, recall under exam conditions may be enough.
If the goal is ethical action, clinical judgment, safe system use or relational competence, the bar is different.
The desired knowledge has to become callable by the world.
AI changes both the value and the role of knowledge
When information is easy to retrieve, it is tempting to conclude that internal knowledge matters less.
Sometimes it does.
A person does not need to memorize every detail that a reliable external system can provide more accurately and quickly.
But external access changes the role of internal knowledge rather than eliminating it.
To evaluate an answer, a person needs enough conceptual structure to know what to ask, what to inspect and what does not fit.
To learn from generated material, they need existing knowledge to integrate new information.
To verify a claim, they need to distinguish authoritative evidence from decorative citation.
To use an explanation responsibly, they need context the model may not possess.
The OECD's AI Capability Indicators are relevant here because AI systems increasingly perform tasks involving language, problem solving, knowledge and memory. As those capabilities expand, humans may spend less time retrieving or producing some forms of information and more time selecting, evaluating and integrating it.
Those higher-level capacities are still knowledge-dependent.
Abundant answers increase the value of question formation
When answers were expensive, access itself carried value.
When answers become cheap, the quality of the question becomes more visible.
A generated answer can be excellent and irrelevant because the problem was framed badly.
It can be factually accurate and ethically incomplete because the important stakeholder was omitted.
It can be well sourced and still fail because the user does not understand the domain well enough to interpret the evidence.
Question formation is not independent of knowledge.
Good questions often come from knowing where the uncertainty actually is.
This is one reason the AI age does not make education a simple shift from “knowledge” to “skills.”
Rich capacity often emerges from their integration.
Knowing the right words can create an illusion of capacity
Human beings are especially vulnerable to domains in which vocabulary is easier to acquire than practice.
A person can learn to say “boundaries” without being able to set one.
They can say “psychological safety” while punishing disagreement.
They can say “human oversight” while approving outputs too quickly to inspect.
They can say “critical thinking” while treating skepticism toward disliked claims as evidence of rigor and skepticism toward favored claims as unnecessary.
Language can mark genuine conceptual development.
It can also create a performance of understanding.
Capacity becomes visible when words are placed under consequence.
The body can know something the person cannot yet explain, but that is not infallibility
Not all knowledge is easily verbalized.
Experts can develop tacit pattern recognition through repeated exposure. A craftsperson may feel that a material is about to fail before articulating the physical cue. A clinician may notice that a case “doesn't fit” before identifying which detail produced the impression.
This does not mean intuition is magical or always correct.
Tacit recognition is shaped by experience, feedback, bias and the quality of the learning environment.
The capacity-versus-knowledge distinction should therefore not become a hierarchy in which felt experience automatically outranks explicit evidence.
Sometimes the body is signaling a useful discrepancy.
Sometimes it is signaling an old expectation.
Capacity includes knowing that both possibilities exist.
Measurement should match the claim
If we claim a person knows something, a knowledge test may be appropriate.
If we claim they can apply it, we need an application task.
If we claim they can recognize when it applies, they need cases in which the relevant category is not announced.
If we claim they can use it under pressure, assessment should eventually include proportionate pressure.
If we claim it transfers, we need a changed context.
If we claim it persists, we need delay.
If we claim it supports ethical action, self-reported agreement with ethical statements is not enough.
The stronger the capacity claim, the more demanding the evidence should become.
Knowledge is often the fastest thing to scale
Institutions love knowledge interventions partly because they are easy to distribute.
Send the policy.
Publish the guide.
Deliver the webinar.
Add the module.
Knowledge transmission can reach thousands of people quickly.
Capacity development is slower because it requires opportunity to act, feedback, variation and often changes in the surrounding environment.
This does not make knowledge programmes pointless.
It makes them honest when they are called knowledge programmes.
A two-hour seminar can transform understanding.
It is less credible to claim that it has transformed a complex relational or ethical capacity without evidence beyond the room.
Knowledge can be external while understanding remains internal
AI also makes it easier to confuse access to knowledge with possession of an adequate mental model.
A person can retrieve a definition at any moment and still lack enough organized understanding to notice when the retrieved answer is being misapplied. External access is often valuable precisely because internal memory is limited. But access and integration answer different questions.
For high-consequence work, the useful design question is not how much information a person can memorize. It is what conceptual structure they need internally in order to interpret what the external system returns, recognize exceptions, compare evidence and know when another source or expert is required.
This creates a layered knowledge environment: some material can remain externally available, some needs to be quickly retrievable, and some needs to be deeply organized because it shapes perception itself. A mature capacity does not maximize memorization. It allocates knowledge according to function.
What the evidence does not show
Current evidence does not show that:
- knowledge is merely memorized fact;
- knowledge and capacity are scientifically agreed opposites;
- a person can develop complex capacity without relevant knowledge;
- experiential or embodied knowing is automatically more accurate than explicit knowledge;
- poor real-world performance always means the person lacks understanding;
- strong test performance proves transfer;
- AI access makes domain knowledge obsolete;
- generic critical-thinking skill can replace subject knowledge;
- every piece of knowledge must be internally memorized to support capacity;
- a brief practice converts understanding into durable capacity by default.
The distinction is useful when it sharpens claims, not when it creates a false hierarchy.
A practical test: what changes when the note is closed?
Imagine a person learns a principle from a page.
They can explain it while the page is open.
Close the page.
Can they retrieve it?
Wait a week.
Can they still retrieve it?
Change the example.
Can they recognize it?
Add competing information.
Can they select what matters?
Add social pressure.
Can they act?
Remove the facilitator.
Can they begin on their own?
These are not universally appropriate tests. They illustrate the progression from knowledge availability toward capacity claims.
Back in the meeting
She still knows the policy.
This time she does something with it.
“Before we move, I want the source for this claim. It may be right, but we haven't checked it.”
There is a small silence.
The executive is impatient.
Nothing in her knowledge has changed in the last thirty seconds.
What changed is that the knowledge crossed the distance into behavior.
Capacity is not a mysterious substance added on top.
It is knowledge, perception, regulation, judgment, context and action becoming coordinated enough to survive the moment in which they are needed.
That is why understanding is precious.
And why it is not always the end of development.
References
- OECD. Future of Education and Skills 2030/2040 and OECD Learning Compass 2030. https://www.oecd.org/en/about/projects/future-of-education-and-skills-2030.html
- OECD. What Students Learn Matters: What does research say? Knowledge, skills, attitudes and values in the Learning Compass. 2020. https://www.oecd.org/en/publications/what-students-learn-matters_d86d4d9a-en/full-report/what-does-research-say_816c97c8.html
- Soderstrom NC, Bjork RA. Learning Versus Performance: An Integrative Review. Perspectives on Psychological Science. 2015;10(2):176-199. https://doi.org/10.1177/1745691615569000
- Blume BD, Ford JK, Baldwin TT, Huang JL. Transfer of Training: A Meta-Analytic Review. Journal of Management. 2010;36(4):1065-1105. https://doi.org/10.1177/0149206309352880
- OECD. Introducing the OECD AI Capability Indicators. OECD Publishing. 3 June 2025. https://doi.org/10.1787/be745f04-en
- Risko EF, Gilbert SJ. Cognitive Offloading. Trends in Cognitive Sciences. 2016;20(9):676-688. https://doi.org/10.1016/j.tics.2016.07.002