Epistemic humility is usually framed simply: we should acknowledge we might be wrong. But this framing, whilst useful, flattens something more interesting. There are distinct dimensions to not knowing, and recognising them separately may matter more than ever.
- The most familiar dimension is content ignorance: not knowing facts. I do not know the population of Laos or the melting point of tungsten. This is the kind of ignorance that search engines address, and we are reasonably comfortable admitting to it.
- But consider a second dimension: contextual ignorance. This is not knowing the background that gives meaning to what we think we know. We may know the word “justice” and use it confidently, yet remain unaware that its meaning has transformed so substantially across centuries that we are not really accessing the concept we think we are accessing. We possess the word but lack the context that once gave it depth. This form of ignorance is harder to detect because we feel knowledgeable. We have the vocabulary.
- A third dimension is methodological ignorance: not knowing about ways of knowing that we are not aware of. Different traditions, whether contemplative, scientific, indigenous, or artistic, have developed distinct methods for generating understanding. If we are trained only in one, we may not even recognise the others as legitimate epistemologies. A strict empiricist might dismiss phenomenological inquiry as mere introspection. A rationalist might overlook embodied or tacit knowledge. We do not know what we cannot see because we have never been given the eyes.
- Fourth is structural ignorance, the classic unknown unknowns. We do not know what questions to ask because we do not know the territory well enough to see the gaps. This is perhaps the most vertiginous form, because it cannot be remedied by effort alone. You cannot research what you do not know exists.
- Finally, there is perspectival ignorance: not knowing how our particular position, historical, cultural, embodied, shapes what we can perceive. The fish does not know it is in water. We see from somewhere, and that somewhere has a shape that excludes as much as it reveals.
Why does this taxonomy matter?
Because each dimension requires a different response. Content ignorance needs information. Contextual ignorance needs historical and cultural inquiry. Methodological ignorance needs exposure to other traditions of knowledge-making. Structural ignorance needs intellectual humility and openness to surprise. Perspectival ignorance needs dialogue with those who see from elsewhere.
In the age of AI, these distinctions become newly urgent. Large language models are extraordinarily good at addressing content ignorance. They can retrieve facts and synthesise information at scale. But they inherit and often amplify the other dimensions. They are trained on text that carries contextual assumptions invisibly. They encode particular methodological approaches whilst appearing neutral. They cannot know what their training data does not contain. And they have no perspective to acknowledge, which may be the most dangerous blindness of all, because it presents as objectivity.
If we mistake the first dimension for the whole of knowledge, we will assume AI has solved epistemology. It has not. It has solved lookup. The deeper dimensions of not knowing remain, now partially obscured by the fluency of machines that sound like they know everything.
Epistemic humility, properly understood, is not one thing. It is at least five, and probably more, each requiring its own practice. If this is true, we are only beginning.
Further reading:
↳ For the phenomenological dimension of this taxonomy — what AI does to human perception — see The Mirror, The Map and the Breath.
↳ How these dimensions of ignorance are embedded in technical architecture is the subject of Your Data Architecture Isn’t Technical.
↳ Why AI labs are structurally positioned to address only one of these dimensions is examined in The Homogeneity Trap.
Garden notes
- The four fogs — the permanent conditions these dimensions describe
- Why the frame cannot see itself — structural ignorance at the institutional level
- Your Data Architecture Isn’t Technical — companion essay: epistemology embedded in architectural choices