cankun.me

The Shape of Not Knowing

Jul 21, 2026
A lone traveler stands on a rocky edge while a narrow path disappears into luminous mist
A lone traveler stands on a rocky edge while a narrow path disappears into luminous mist · View full resolution

In 2002, answering a question about evidence linking Iraq, weapons of mass destruction, and terrorist organizations, Donald Rumsfeld sorted uncertainty into three kinds: the things we know we know, the things we know we do not know, and the things we do not know we do not know. The briefing transcript is worth reading in context; an elegant distinction can still be used inside a disastrous argument. But the distinction outlived the argument because it gives ignorance a shape.

Complete it into a two-by-two matrix and a fourth category appears:

AwarenessI know itI do not know it
I am aware of itI know that I knowknown knownI know that I do not knowknown unknown
I am not aware of itI do not know that I knowunknown knownI do not know that I do not knowunknown unknown
The Rumsfeld Matrix separates whether something is known from whether that state is itself recognized. See also this applied walkthrough.

The top row is easy to discuss. A known known can be stated. A known unknown can at least be asked. Ordinary question answering is built for these two cells: give the system a proposition or a gap, and ask it to confirm or fill it.

The bottom row is where conversation gets interesting. It contains what I cannot simply type into a prompt. I may possess a distinction without having words for it. I may be missing an entire region of possibility and therefore lack even the question that would reveal it. These two failures look equally blank from the outside, but they require opposite kinds of help.

A map is not an inventory

There is a paradox in putting unknown unknowns on a map. The moment I can name one, it is no longer wholly unknown. At best I can mark a frontier: there is probably something in this region that my present vocabulary and model do not expose.

This is why the matrix should not be treated as a complete inventory of mental contents. It is an orientation device. It says where the next conversational move should point, not everything waiting to be found there.

The distinction matters because a bad map of ignorance creates a peculiar kind of overconfidence. A list of known unknowns feels comprehensive precisely because everything outside the list is absent. The map becomes useful only when one of its symbols means not “miscellaneous unanswered questions,” but “the frame itself may be missing something.”

That makes the fourth cell, the unknown known, equally important. It prevents us from treating whatever was not stated as whatever was not known.

We know more than we can tell

Michael Polanyi built The Tacit Dimension on one blunt sentence: “we can know more than we can tell.” The line is often reduced to know-how: riding a bicycle, recognizing a face, sensing that an argument is wrong before locating the broken step. But its force is broader. Articulation is not a transparent export operation that copies already explicit knowledge from the mind into language. Some knowledge exists first as skilled discrimination, attention, timing, or recognition.

You can see this in aesthetic judgment. Someone says a transition feels cheap. Ask what easing curve they want and they cannot answer. Show them linear motion, ease-out, spring motion, and a staggered entrance, each attached to a tiny scene, and the answer may be immediate. Nothing was retrieved by recall. A vocabulary was offered, and recognition selected from it.

This is one species of unknown known: not a secret proposition hidden intact in the mind, but a capacity that becomes legible when the right contrast appears. The correct conversational move is therefore not another blank question. “Tell me what you want” merely repeats the demand the person cannot satisfy. The useful move is to lend words: offer a small set of named distinctions, make each concrete enough to feel, and let the person point, reject, or revise.

Tacit knowledge does not exhaust the unknown-known cell. Unexamined habits, inherited categories, bodily warnings, and assumptions a person has never needed to state can live there too. Polanyi is not a definition of the quadrant. He explains why the quadrant cannot be collapsed into ordinary missing information.

Meaning arrives in use

An earlier line of thought comes from Ludwig Wittgenstein. I place him after Polanyi here because the essay follows the problem, not the calendar: Philosophical Investigations appeared before The Tacit Dimension. If a word's meaning were a private object stored behind it, then finding the right word would mean matching a label to that hidden object. Wittgenstein's later philosophy instead turns attention toward use. The Stanford Encyclopedia of Philosophy summarizes the movement in the Philosophical Investigations from meaning as representation toward meaning as use: look at the activities in which a word does work, the contrasts it draws, and the judgments it makes possible.

This is why an example matters more than a synonym. “Spring easing” becomes useful when it is placed in a miniature language-game: a panel overshoots its resting point, returns, and feels as if it has mass. The scene is not decoration around the definition. It is part of how the expression acquires grip.

Wittgenstein's discussion of aspect-seeing goes one step further. In the duck-rabbit illusion, the marks on the page do not change when perception changes from duck to rabbit. What changes is the aspect under which they are seen.

One ambiguous black-ink animal head that can be seen as either a duck facing left or a rabbit facing right
Duck or rabbit? Read the two projections on the left as an open bill and the animal faces left; read the same lines as long ears and it faces right. The marks remain fixed while the aspect changes. Original illustration generated for this essay.

Later commentary connects this idea of seeing an aspect with seeing a meaning. That is close to what a productive conversation often does. It does not merely add another fact. It makes an existing situation newly nameable, visible, or testable.

I do not mean that Polanyi, Wittgenstein, and the Rumsfeld matrix form a hidden unified theory. They do not. The connection is an engineering one. The matrix identifies where a conversation is stuck. Tacit knowing explains why asking for explicit recall may fail. Meaning-in-use and aspect-seeing suggest how examples, contrasts, and probes can change what the user is able to recognize and say.

The two dark cells need opposite help

Suppose I know what I like but cannot describe it. The answer is already constrained by my experience. I need candidates rich enough to recognize. This is the unknown known, and the move is: lend words.

Suppose instead that I am entering a field whose important options I have never encountered. No candidate is waiting inside me to be recognized. Asking me which architecture, research design, or career path I prefer only makes me choose among the few names I happened to bring. This is the unknown unknown, and the move is: give a tour. Show me the nearby option space, what each region is for, what it costs, and which direction you think fits.

The distinction is practical:

  • For an unknown known, recognition beats recall.
  • For an unknown unknown, orientation must come before preference.

The other two cells also imply moves. Confirm and compress what is already settled; do not let it occupy the whole conversation. Investigate a named gap by identifying what evidence, reasoning, or experiment would resolve it; when that probe is cheap enough to run, run it rather than asking the user to speculate in its place. A four-cell map should produce four behaviors, not merely four labels.

And the behaviors cannot be a closed list. Taste articulation, capability discovery, and open decisions are useful examples, but they are not the boundaries of the method. Another topic may require exposing an assumption, changing scale, offering a counterexample, drawing a distinction, or running a premortem. The common action is aspect-seeing: make one relevant part of the subject more nameable, visible, or testable.

A conversation should leave a trace

Once the matrix becomes a discussion method, it needs state, and the state must stay cheap to redraw: a fixed skeleton, one phrase per entry, a card rather than a document. A map that outgrows a glance stops being redrawn, and a map that is not redrawn stops being state. Within the card, I use four marks:

  • ✓ settled
  • ? a named question
  • ! a recognition candidate
  • ~ an unexplored frontier

These are not confidence scores. They describe what kind of work remains. A ? can be answered or explicitly deferred. A ! can be recognized, rejected, or refined. A ~ cannot be “solved” in the exhaustive sense; it can be explored into something more concrete or honestly retained as residual darkness.

That last possibility matters. A conversation that converts every mark to ✓ has probably confused closure with omniscience. Before closing, it should attempt one frame-break: name a plausible missing aspect that, if real, would change the conclusion. If it matters, keep going. If it does not, preserve it as a boundary rather than manufacture an answer.

The final artifact should not be the map itself. The map is a receipt: it shows how the discussion moved. What the user leaves with should be a brief: the vocabulary they recognized, the choices they settled, the gaps they deferred, and the darkness that remains, written so it can be pasted into a future conversation without reconstructing the entire path.

Turning the idea into a skill

I encoded this method as an open-source agent skill called known-unknowns.

It gives an agent a simple discipline. Open by drafting the whole map at once, from the topic and whatever domain knowledge applies, with every inference tagged as a guess. This sounds presumptuous, and that is the point: a wrong map draws corrections faster than a blank interview draws answers. Then work one cell per turn, entering wherever the map is darkest but still useful. Contribute before asking: lend vocabulary, tour an option space, run the probe a named gap allows, test an assumption, or choose another move that creates an aspect-seeing. Ask exactly one question. Redraw the map on the next turn, so anything the user recognized or corrected becomes part of the visible state.

The constraint is deliberately conversational. A blank interview offloads all the work onto the person who came because they did not yet have the words or the map. An unbroken lecture fills space without letting recognition steer. One contribution followed by one question creates a narrower loop: the agent supplies structure, the user supplies correction, and the map changes.

There is no promise that this discovers every unknown unknown. Such a promise would contradict the point of the matrix. The skill does something smaller and more useful: it refuses to treat silence as emptiness, and it gives a discussion a repeatable way to approach what cannot yet be asked.

← Back to Writing