Situated Knowledges
In Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective, Haraway proposed a radical reframing of objectivity. Against the god trick — the fantasy of a view from nowhere that she identified as politically interested rather than neutral — she argued that the strongest knowledge is produced not by transcending perspective but by acknowledging its partiality and combining partial perspectives in ways that make their specificities visible. The framework has become, somewhat improbably, one of the most cited texts in contemporary AI ethics, because it provides the vocabulary for analyzing the specific partiality of the machine's gaze: what its training data includes, what it excludes, whose knowledge it carries, and whose it erases.
In the AI Story
The essay was published in Feminist Studies as Haraway's response to the late-1980s debate between feminist empiricism and postmodern skepticism. She refused both poles: the empiricist claim that better method could produce unbiased knowledge, and the postmodern claim that all knowledge was equally situated and therefore equally arbitrary. Her alternative was that knowledge was situated and could be evaluated — but the evaluation required practices of accountability, of tracing the specific conditions under which any given claim was produced.
Applied to large language models, the framework exposes what the industry's rhetoric of universality conceals. Claude does not see the world as it is. It sees the world as its training data can show it — and the training data overrepresents the English-language internet, urban and educated populations, propositional rather than embodied knowledge, recent rather than deep history, powerful rather than marginalized voices. These specificities are not technical limitations to be corrected. They constitute the machine's situated perspective.
Computer vision researchers have applied Haraway's framework directly, identifying how traditional systems treat image sets as objective recordings of reality, detached from the cameras and photographers who take them and treat model performance as objective truth — performing exactly the double god trick she described, with what the researchers call horrible consequences for bias and injustice.
The concept also connects to the fishbowl metaphor that Edo Segal uses in The Orange Pill. Everyone occupies a fishbowl of familiar assumptions they have stopped noticing. The powerful think theirs is bigger, and sometimes it is. It is still a fishbowl. Haraway's situated knowledges framework is a rigorous account of why the fishbowl is not a metaphor for delusion but a structural feature of any actual knower, including the machine.
Origin
Haraway drafted the essay in the late 1980s while teaching at UC Santa Cruz, responding to a Sandra Harding essay in the same feminist science studies debate. The argument built on her work in Primate Visions, where she had shown how the science of primatology was shaped by the specific gendered, racialized, and colonial positions of the scientists who produced it — not as a critique to be overcome but as a structural feature of the knowledge itself.
Key Ideas
Partiality is a feature, not a bug. All knowers are somewhere; no knower is everywhere. The pretense of universality is what must be dismantled.
The god trick is political. The fantasy of objectivity from nowhere serves specific interests — it naturalizes dominant perspectives as neutral.
Stronger objectivity through accountability. Acknowledging the conditions of knowledge production strengthens rather than weakens the claims that can be made.
Combining partial perspectives. The richest knowledge comes from the productive collision of multiple situated views, not from any single view's aspiration to totality.
The machine is situated too. AI systems have specific perspectives determined by training data, design choices, and institutional contexts — not universal views of human knowledge.
Appears in the Orange Pill Cycle
Further reading
- Donna Haraway, "Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective," Feminist Studies 14:3 (1988)
- Sandra Harding, Whose Science? Whose Knowledge? (Cornell University Press, 1991)
- Lorraine Code, What Can She Know? Feminist Theory and the Construction of Knowledge (Cornell, 1991)
- Timnit Gebru et al., "Datasheets for Datasets" (2018)
- Kate Crawford, Atlas of AI (Yale University Press, 2021)