Situated Knowledge
Situated knowledge is the form of understanding that develops only through a practitioner's sustained, first-person engagement with the resistance of specific material reality. Unlike propositional knowledge (facts that can be stated) or generic skills (capacities that transfer across domains), situated knowledge is bound to the history of practice that produced it. It is the senior engineer's ability to feel that a codebase is wrong before she can say why; the surgeon's hand that knows, before the mind does, that this tissue is not healthy; the intelligence analyst's judgment that this signal is anomalous in a way the training data cannot specify. Suchman's framework identifies situated knowledge as the irreducibly human contribution to any system that must operate in open worlds — and as the specific capacity that AI threatens to erode by performing the activities through which it develops.
In the AI Story
Situated knowledge differs in kind, not in degree, from propositional knowledge. Propositional knowledge — the database schema, the statute, the diagnostic criteria — can be written down, transmitted, stored, and processed. Situated knowledge is a relationship between a practitioner and a domain, forged through years of engagement and maintained through continued practice. It resists extraction because its content is inseparable from the history that produced it. The senior engineer cannot hand her situated knowledge to a junior engineer; she can only work alongside the junior engineer on real problems, creating the conditions under which the junior engineer might begin to develop her own.
The philosopher Michael Polanyi's formulation — 'we know more than we can tell' — captures an adjacent truth but not the full one. Suchman's framework pushes beyond Polanyi: the issue is not merely that practitioners possess knowledge they cannot articulate. The issue is that the knowledge exists as a relationship, as a repertoire of responses available in the moment of situated action, that cannot be relocated from its developmental history. It is not hidden inside the practitioner waiting to be extracted; it lives in her engaged practice.
The concept directly addresses The Orange Pill's geological metaphor for how understanding accumulates. Segal's intuition — that every hour debugging deposits a thin layer of understanding that aggregates into something solid — Suchman's framework formalizes as a specific developmental mechanism. Each encounter with resistance in specific circumstances is a moment of situated action. The practitioner improvises. The improvisation deposits understanding. Over time, the accumulation produces a practitioner who can navigate unfamiliar territory because she has navigated enough familiar territory to recognize the patterns that persist across cases.
AI's disruption is specific. It handles the situational labor through which situated knowledge accumulates. The AI encounters the dependencies, resolves the edge cases, produces the output. The practitioner receives the output without having passed through the developmental friction. She possesses the result without the residue. This is not a loss of skill in the generic sense; it is the specific failure of specific situated knowledge to develop in specific practitioners who would otherwise have developed it. And because that knowledge is what allows practitioners to evaluate AI outputs against reality, its absence is the precise institutional risk Suchman's work illuminates.
Origin
The concept runs throughout Suchman's corpus, from Plans and Situated Actions (1987) to her recent work on military AI and algorithmic targeting. It draws on Garfinkel's ethnomethodology, Donna Haraway's feminist epistemology (particularly the 1988 essay 'Situated Knowledges'), and the broader STS tradition's attention to the material and embodied conditions of knowledge production.
The sharpest recent articulation comes in Suchman's 2025 AI Now Institute interview, where she extends the framework to intelligence analysis and targeting systems: the situated judgment that human analysts develop through years of working specific theaters with specific actors is precisely what automated targeting systems bypass — at speeds incompatible with its exercise.
Key Ideas
Knowledge as relationship. Situated knowledge is not stored in the practitioner's head but lives in the relationship between practitioner and domain, constitutively tied to the history that produced it.
Resistance to transfer. The knowledge cannot be extracted and installed elsewhere. It can only be acquired through equivalent engagement with equivalent material reality.
Built through improvisation. Every moment of situated action — every encounter with the gap between plan and reality — deposits a layer. The accumulation produces the practitioner.
Mentoring as co-practice. Situated knowledge transfers only through shared engagement with real problems. Consultation transfers propositions; co-practice transfers dispositions.
The evaluation capacity. Situated knowledge is what allows practitioners to evaluate AI outputs against reality. Its erosion is the institutional risk that output metrics cannot detect.
Appears in the Orange Pill Cycle
Further reading
- Donna Haraway, 'Situated Knowledges' (Feminist Studies, 1988)
- Michael Polanyi, The Tacit Dimension (Doubleday, 1966)
- Jean Lave, Cognition in Practice (Cambridge University Press, 1988)
- Lucy Suchman, Human-Machine Reconfigurations (Cambridge University Press, 2007)