204 – Known unknowns and unknown unknowns

Known unknowns and unknown unknowns: The map of what we don’t know

In February 2002, Donald Rumsfeld, then United States Secretary of Defense, delivered a response at a Pentagon press briefing that became one of the most quoted — and most mocked — statements in modern political life. Asked about the evidence linking Iraq to weapons of mass destruction, he said: “There are known knowns — there are things we know we know. We also know there are known unknowns — that is to say, we know there are some things we do not know. But there are also unknown unknowns — the ones we don’t know we don’t know.” The laughter that followed from the assembled journalists, and the sustained ridicule in the press over subsequent days, treated the statement as a masterpiece of bureaucratic evasion — a politician’s linguistic contortion designed to say nothing in as many words as possible.

The mockery was itself a demonstration of the very failure Rumsfeld was describing. His taxonomy was not evasion. It was, as several philosophers and epistemologists later pointed out, one of the most precise and practically important distinctions in the analysis of uncertainty — one with direct roots in the philosophy of knowledge and direct consequences for how any complex undertaking, from military planning to scientific research to personal decision-making, should be organized.¹ The people who laughed had encountered a map of the territory of ignorance and dismissed it because it failed to look like what they expected knowledge to look like. They were experiencing, in real time, the phenomenon the map was describing: the thing you don’t know you don’t know looks, from the inside, like nothing at all.

This article is about what lies in the quadrants beyond what we know — about the structure of ignorance, the specific dangers of its different forms, and what can be done, practically and honestly, to make the unknown a little more visible before it becomes consequential.

The four quadrants

Rumsfeld’s taxonomy distinguishes three categories. It is more useful, for the purposes of this series, to add a fourth and examine all of them together.²

The first quadrant — known knowns — contains what we know and know that we know: the facts, skills, and understandings that are explicitly available to conscious retrieval and that we would confidently assert if asked. The second quadrant — known unknowns — contains what we do not know but know that we do not know: the questions we are aware of, the gaps we have identified, the uncertainties we have named. These are the domains where we can in principle direct inquiry, because we have at least recognized that there is something to inquire about. The third quadrant — unknown unknowns — contains what we do not know and do not know that we do not know: the questions that have not occurred to us, the variables our model has no slot for, the risks that are invisible because our framework provides no category in which to place them. And the fourth quadrant — unknown knowns — contains what we know but do not know that we know: the tacit understandings, embodied competencies, and unarticulated assumptions that shape our behavior without ever becoming explicit. This fourth quadrant, largely invisible in Rumsfeld’s formulation, is in some ways the most interesting of all.

The first and second quadrants are the domain of ordinary intellectual life. We reason from what we know toward what we do not know, designing inquiries to convert known unknowns into known knowns. This is what science, scholarship, and systematic investigation do. The third and fourth quadrants are the domain of genuine epistemological difficulty — the places where the ordinary machinery of inquiry cannot easily reach, because in the third case the question has not been asked and in the fourth case the knowledge has not been articulated. Together, the third and fourth quadrants constitute the territory that every model omits and that every map leaves blank. And as the history of human catastrophe consistently demonstrates, it is in the blank spaces that the most consequential events tend to occur.

The unknown unknowns — what the model has no slot for

The third quadrant is the most dangerous because it is, by definition, invisible from inside the framework that is missing it. You cannot direct inquiry toward a question that has not occurred to you. You cannot take precautions against a risk that your model provides no category for. The unknown unknown reveals itself only after the fact — as a surprise, an anomaly, a disaster that nobody saw coming because the framework within which everyone was operating had no mechanism for anticipating it.

The financial crisis of 2008 is among the most extensively documented examples. The risk models used by financial institutions, rating agencies, and regulators to assess the safety of mortgage-backed securities were sophisticated, mathematically rigorous, and extensively tested against available historical data. What they did not model — what their frameworks provided no category for — was the possibility of simultaneous, nationwide declines in housing prices. In the decades of data that informed the models, American housing prices had never fallen simultaneously across all major markets. The models therefore treated correlated nationwide decline as either impossible or negligibly improbable. It was an unknown unknown: not a risk that had been assessed and found acceptable, but a risk that the model’s architecture could not represent. When the event occurred, it was not just surprising — it was structurally unpredictable from within the frameworks that had been deployed.³

The sociologist Charles Perrow identified a related pattern in what he called normal accidents: failures in complex, tightly coupled systems that arise from combinations of small faults that no designer anticipated because no individual fault was significant enough to worry about and because the combination had never occurred.⁴ The Three Mile Island nuclear accident, the various aviation disasters that preceded the development of modern safety culture, the Challenger space shuttle disaster — in each case, the failure mode was not in any individual component that had been identified as a risk. It was in an interaction between components that the system’s designers had not modeled, because the interaction had no precedent and their framework provided no category for it. The unknown unknown, in each case, was the emergent behavior of the system as a whole.

What makes the third quadrant so persistently dangerous is not that we are careless or unintelligent. It is that every framework, by its nature, has a boundary — a point beyond which it cannot see, because seeing requires a category and categories are what the framework provides. The model is always smaller than the territory. The quadrant of unknown unknowns is not a defect in particular models; it is a structural feature of what it means to use any model at all. The Conscious Look does not promise escape from this condition. What it promises is a somewhat more honest accounting of it.

The unknown knowns — what we know but have not said

The fourth quadrant — unknown knowns — is less frequently discussed but equally important. It contains the knowledge that is present in practice but absent from explicit formulation: the competencies we exercise without being able to articulate, the assumptions that organize our behavior without ever having been stated, the values that guide our choices without ever having been examined.

Tacit knowledge, as the philosopher Michael Polanyi described it, is the most fundamental form of this.⁵ The experienced surgeon who knows, through the feel of the instrument, that the tissue is behaving unusually, without being able to specify in advance what unusual feels like. The master teacher who reads the room and adjusts the approach before any student has said a word, without being able to describe afterward exactly what signals they were responding to. The experienced manager who senses that a project is in trouble from the tone of the weekly meeting, before any metric has turned negative. In each case, the knowledge is real, reliable, and consequential — and it is not explicitly known. It is practiced rather than stated, exercised rather than articulated.

The institutional version of unknown knowns is more problematic. Organizations develop characteristic assumptions about how the world works — about what the competition will do, what customers want, what risks are worth taking, what kinds of people succeed — that are never written down and rarely discussed, because they are treated as obvious. These assumptions organize every decision and every interpretation of evidence, but they are invisible to the people inside the organization precisely because they have never needed to be articulated. When the environment changes in a way that makes the assumptions wrong, the organization continues to apply them — not out of stubbornness, but out of simple unawareness that the assumptions exist and therefore could be questioned. The unknown known has become a trap.

The most consequential unknown knowns in public life are the ideological and moral assumptions embedded in the frameworks through which policy is designed and justified. When an economic model assumes that actors behave rationally, this is not merely a technical choice — it is an assumption about human nature that carries moral weight and that shapes which policy interventions seem promising and which seem absurd. When a public health framework assumes that information leads to behavior change, this is not merely an empirical hypothesis — it is an assumption about human psychology and human agency that has consistently been contradicted by evidence but that has proven extraordinarily resistant to revision because it is embedded in the framework as a background assumption rather than as a testable claim. The unknown known, in each case, is the assumption that is doing the work but has never been brought into the light.

The known unknowns — and why they are less dangerous than they feel

The second quadrant — known unknowns — is where most of the explicit anxiety about uncertainty lives, and it is worth observing that it is genuinely the least dangerous quadrant. A known unknown is a gap you are aware of. You can direct inquiry toward it, take precautions against it, design contingency plans for it, and at least communicate to others that you are operating in a domain of acknowledged uncertainty.

The physician who tells a patient that the diagnosis is uncertain — that the symptoms are consistent with several conditions and that further tests are needed — is operating in the territory of known unknowns. This is honest and useful, even if it is anxiety-provoking for the patient. The politician who acknowledges that the long-term economic consequences of a policy are uncertain — that the models produce a range of outcomes and that confidence in any specific prediction is limited — is doing the same. The scientist who publishes a finding with explicit acknowledgment of the study’s limitations is performing the same intellectual service.

What all of these share is the conversion of unknown unknowns into known unknowns through the act of naming the gap. This conversion — from unconscious incompetence to conscious incompetence, in the terminology of learning theory — is one of the most valuable things that careful thinking about any subject can accomplish. It does not fill the gap. It makes the gap visible, which is the necessary precondition for doing anything about it.

Making the unknown knowable — practical strategies

None of the strategies available for reducing the scope of unknown unknowns is reliable, complete, or sufficient. But several have proven consistently useful across domains.

The first is deliberate exposure to frameworks other than one’s own. The unknown unknown is unknown within a particular framework. Someone operating from a different framework — with different background assumptions, different categories, different prior experiences — may be able to see what the first framework cannot. This is the intellectual case for diversity of thought that goes considerably deeper than the social case: not diversity as representation but diversity as epistemological insurance, because the gaps in one framework are more likely to be visible from within a different one. The red team — the group explicitly tasked with finding the assumptions that the main group has not questioned — is an institutionalized version of this. So is peer review, which works not merely because experts can catch errors but because experts from different subfields can see the assumptions that practitioners in a single subfield have made invisible through familiarity.

The second is the pre-mortem — the thought experiment in which a team imagines, before acting, that the plan has failed catastrophically, and asks what went wrong. The standard post-mortem asks what went wrong after the failure has occurred. The pre-mortem asks the same question in advance, which has the effect of making certain unknown unknowns temporarily thinkable. By giving permission to imagine failure, the pre-mortem creates a momentary license to voice concerns that the culture of commitment and optimism around a plan would otherwise suppress. The psychologist Gary Klein, who developed the pre-mortem as a formal method, found that it consistently surfaced risks that had not been raised in conventional planning discussions.⁶

The third is the practice of asking, explicitly and systematically, what would have to be true for the current plan to fail. This is the diagnostic question of this series applied to the fourth quadrant: not what evidence would change our conclusion, but what conditions would need to obtain for the conclusion to be wrong. The exercise surfaces the hidden assumptions — the unknown knowns — that are doing the most work in any plan, because the conditions that would make the plan fail are often exactly the conditions that the plan’s assumptions have made it impossible to anticipate. When a team asks “what would have to be true for this to fail?” and finds itself answering “the market would have to behave in a way it has never behaved before,” it has named an assumption that was previously invisible: the assumption that historical market behavior is a reliable guide to future behavior under the current conditions.

The fourth is Chesterton’s fence, described more fully in article 706: do not remove an arrangement — do not discard a practice, dismantle an institution, or abandon a convention — until you understand why it exists. The practices that seem most arbitrary, most wasteful, most obviously due for elimination, are often the practices that encode the response to an unknown unknown that was discovered, at some cost, by an earlier generation. The bureaucratic procedure that seems pointless was probably created in response to a specific failure. The safety protocol that seems excessive was probably written in response to an accident. These practices are carriers of unknown knowns — of lessons that have been institutionalized without being explicitly taught — and eliminating them before understanding what they encode is the surest way to rediscover the original lesson at the original cost.

The blank spaces on the map

Medieval cartographers, confronted with the edges of their known world, filled the blank spaces with sea monsters. The monsters were not a claim about the actual contents of those regions — they were a notational convention meaning: here be the unknown, and the unknown is dangerous. It is an honest convention and, in the relevant sense, a correct one. The unknown unknown does not present itself as a monster. It presents itself as nothing at all — as the absence of a category, the silence where a question should be, the smooth continuation of a map that has simply run out of territory to represent.

There is a geometrical observation — widely attributed to Einstein but without traceable origin in his work, and worth keeping for its precision regardless of who first made it — that captures the relationship between knowledge and ignorance more exactly than any other formulation available. As the circle of knowledge expands, so does the circumference of darkness surrounding it. This is not merely a poetic observation. It is a mathematical one, and it requires careful reading if it is not to be misunderstood as a counsel of despair. The interior of the circle — the domain that has been thoroughly explored — is genuinely known, and known with increasing precision as inquiry proceeds. The basic mechanisms of chemistry, the laws of planetary motion, the structure of the genetic code: these are not becoming more uncertain as science advances. They are becoming more tightly constrained, more precisely verified, more reliably applicable. In Bayesian terms, the prior probability of the core findings has been updated by so much evidence that the residual uncertainty is small. Knowledge in the interior is real and cumulative. What grows as the circle expands is not ignorance about what is already known. It is the circumference — the frontier — the visible boundary between the explored and the unexplored.

The mechanism that drives this expansion is precisely what Richard Feynman described in his 1981 BBC interview, when he was asked why magnets repel each other.⁷ Feynman’s response was to decline the question — not out of ignorance but out of philosophical precision. Any answer, he pointed out, would be an explanation in terms of something else, something more familiar. And that something else would immediately invite a new why. Why do opposite charges attract? Because of the structure of the electromagnetic field. Why does the electromagnetic field have that structure? Because of the equations of quantum electrodynamics. Why do those equations hold? This is where Feynman paused — because at the bottom of the explanatory hierarchy there are things that simply are, with no more familiar framework available to reduce them to. The chain of why questions does not terminate because reality runs out of depth. It terminates because the questions eventually reach a level where no more familiar things are available to serve as an explanation. Every genuine answer generates a deeper why. The physicist who has mastered quantum field theory occupies a position from which more of physics is visible and from which the distance still to travel is more accurately appreciated. This is one of the reasons why genuine expertise so consistently produces humility: not because the expert knows less, but because from their position on the frontier they can see how much of the map remains undrawn.

There is a further observation that the Feynman chain makes available, and it is the most honest limit the series can acknowledge. The chain of why questions is in principle infinite. Reality has no obligation to stop being explicable at the point where human minds stop being capable of following the explanation. The constraint is not logical — there is no reason in principle why the chain should terminate. The constraint is cognitive: the architecture of the human mind was built by evolution for survival in a medium-sized world, at medium velocities, over medium timescales, and there is no guarantee that it is equipped to comprehend whatever lies at the deepest levels of physical reality. At some point in the descent, we will still ask why — and the answer may exist, may even be in some sense accessible to mathematics — but we will not be able to understand it in the way that Feynman meant by understanding: as a reduction to something more familiar, something felt rather than merely calculated. The quadrant of unknown unknowns does not shrink as knowledge expands. But this is not because progress is an illusion. It is because genuine progress in understanding the interior simultaneously reveals how much more exterior there is to explore — and because the exterior, unlike the interior, may eventually exceed not merely our current knowledge but our permanent cognitive reach. Knowing that the Earth orbits the Sun created the question of what moves the Earth, which created the question of what gravity is, which created the question of what spacetime is, which created questions that remain open today. Each answer is a real answer. Each answer generates more questions than it closes. This is not a failure of knowledge. It is its characteristic structure — and its characteristic humility.

The Conscious Look, applied to the quadrants of knowledge and ignorance, is the practice of asking where the map runs out — not where the known unknowns are, because those are already visible, but where the framework itself might be generating blind spots that are invisible from within it. This is harder than any of the specific strategies described above, because it requires questioning the instrument of inquiry rather than merely the conclusions it has produced. It is the difference between asking whether the answer is right and asking whether the question was the right one.

There is no method that guarantees success at this task. There is no framework for identifying the limits of frameworks that is itself free of limits. The best available response to this regress is not a technical solution but a disposition: the combination of genuine intellectual humility about the incompleteness of one’s current model and genuine curiosity about what lies beyond its edges. Not the performed humility that acknowledges uncertainty as a rhetorical gesture before proceeding with full confidence, but the operational humility that actually changes how a plan is designed, how a conclusion is qualified, and how a surprise is received when it arrives. The surprise, when it comes, is data. The model that can receive it as such — rather than explaining it away as noise, anomaly, or sabotage — is the model that is capable of learning. And a model that is capable of learning is, in the only sense that matters, a model that knows something about its own limits.

Notes

¹ The intellectual history of the known-unknown distinction significantly predates Rumsfeld’s 2002 formulation. Versions of the taxonomy appear in the Socratic tradition — Socrates’ claim to wisdom consisted precisely in knowing that he did not know, which is the conversion of an unknown unknown into a known unknown — and in the epistemological literature from at least the seventeenth century. In the twentieth century, the framework was developed in the context of organizational learning theory, particularly by the management scholars Chris Argyris and Donald Schon, whose concept of double-loop learning addresses the specific challenge of revising the assumptions that govern a model rather than merely the model’s conclusions. The most precise contemporary formulation prior to Rumsfeld is in the work of the philosopher Ann Kerwin, whose 1993 paper “None Too Solid: Medical Ignorance” introduced the quadrant framework explicitly. Rumsfeld’s formulation omitted the fourth quadrant — unknown knowns — which the philosopher Slavoj Zizek added in a widely circulated commentary on the original statement.

² The fourth quadrant — unknown knowns — was added to Rumsfeld’s original three-part taxonomy by several commentators, most prominently the philosopher Slavoj Zizek in a 2004 essay. The term captures the phenomenon that Polanyi had described decades earlier under the heading of tacit knowledge, and that Argyris and Schon had described in the organizational context under the heading of theories-in-use — the implicit theories that govern actual behavior, as distinct from the espoused theories that people consciously hold and articulate. The gap between theories-in-use and espoused theories is one of the most consistent findings in the study of organizational behavior and is directly related to the fourth quadrant’s status as unknown knowledge.

³ The failure of mortgage-backed securities risk models in the 2007-2008 financial crisis has been analyzed in detail in several subsequent accounts. The specific failure of correlation assumptions — the treatment of nationwide simultaneous housing price declines as negligibly probable — is described in Ricardo Rebonato’s Plight of the Fortune Tellers (2007), written before the crisis but anticipating exactly this failure mode, and analyzed in retrospect in Michael Lewis’s The Big Short (2010) and in the Financial Crisis Inquiry Commission Report (2011). The mathematical apparatus that produced the failure — in particular the Gaussian copula function used to model default correlations — is discussed accessibly in Felix Salmon’s 2009 Wired article, “The Formula That Killed Wall Street.”

⁴ Perrow, C. (1984). Normal Accidents: Living with High-Risk Technologies. Basic Books. Perrow’s central concept — interactive complexity combined with tight coupling — identifies the structural properties of systems that make normal accidents predictable as a class even when they are unpredictable in their specific instantiation. A system is interactively complex when its components interact in ways that are not fully anticipated in the design; it is tightly coupled when a failure in one component propagates rapidly to others before it can be isolated and corrected. Nuclear power plants, chemical processing facilities, aircraft, and — he later argued — financial systems all have both properties, which is why they are systematically prone to failures that emerge from unexpected combinations of individually minor faults.

⁵ Polanyi, M. (1966). The Tacit Dimension. Doubleday. Polanyi’s formulation — “we know more than we can tell” — is one of the most quoted in the philosophy of knowledge, and it captures the fourth quadrant more precisely than any alternative formulation. Polanyi’s examples range from the perceptual — the recognition of a face, which we can do reliably without being able to specify the features we are responding to — to the scientific — the judgment of a skilled experimenter about whether a result is reliable, which cannot be fully reduced to explicit criteria. The implications for the theory of knowledge are substantial: if much of what we know cannot be articulated, then the explicit knowledge that is available to formal transmission and verification is not all, and may not even be the most important part, of what we know.

⁶ Klein, G. (1998). Sources of Power: How People Make Decisions. MIT Press. The pre-mortem is described in chapter twelve. Klein’s development of recognition-primed decision-making — the observation that experienced decision-makers in naturalistic settings typically recognize situations as typical of familiar types and apply the response that has worked before, rather than evaluating options analytically — provides the cognitive background for understanding why the pre-mortem is necessary: the culture of commitment that surrounds any sufficiently ambitious plan systematically suppresses the concerns that the pre-mortem is designed to surface. The method works precisely because it gives explicit permission to voice what the ambient optimism of the planning process has been making it socially costly to say.

⁷ The BBC interview referred to here is the 1981 Horizon documentary “The Pleasure of Finding Things Out,” in which the interviewer Christopher Sykes asks Feynman to explain why magnets attract and repel each other. Feynman’s seven-minute response is one of the most instructive available accounts of what it means to explain something — of what an explanation is, what counts as a satisfying one, and why at the bottom of every explanatory chain there is always a brute fact that cannot be further reduced. The full interview is freely available online and repays repeated viewing. The specific connection to the circle-of-knowledge image is this: Feynman’s chain of why questions is exactly the mechanism by which the circumference expands. Each answer to a why question at the frontier generates a new why at a deeper level, pushing the frontier further out. The frontier does not disappear when the interior is understood. It relocates, to a position further from ordinary experience and closer to the limits of what the human cognitive architecture can follow. Article 408 of this series examines the Feynman interview in more detail, specifically in relation to the limits of scientific explanation and the distinction between explanation and description.