Indigenous Knowledge and AI Limits

In the AI Story

The Zuni people of New Mexico maintain astronomical knowledge developed through centuries of careful observation. This knowledge is not recorded in written star charts or mathematical models. It is encoded in the orientation of ceremonial structures, in the timing of ritual cycles coordinated with solstice and equinox events, in oral narratives that integrate astronomical observation with agricultural practice, water management, and social organization. A Zuni elder explaining when to plant does not cite an astronomical calculation abstracted from context. The answer is embedded in a narrative that connects the position of specific star clusters to soil moisture patterns to the behavior of local species to the obligations of communal labor—a holistic understanding in which separating the astronomy from the ecology from the social practice would destroy the knowledge's practical value.

Srinivasan's research revealed that when external institutions attempt to capture this knowledge in databases—recording which stars mark which planting times, cataloging which plants grow in which conditions—the extraction process systematically strips away the knowledge's organizing logic. The database has slots for 'astronomical observation,' 'planting schedule,' 'crop type,' 'soil condition'—categories that reflect Western scientific taxonomy. The relationships between these elements, the narrative structure that holds them in meaningful connection, the cultural protocols that determine who may speak this knowledge and in what contexts—these dimensions are lost because the database's architecture has no place for them. What remains is data. Accurate data, potentially useful data, but data severed from the epistemological framework that made it indigenous knowledge rather than a collection of observations.

The epistemological difference is not merely organizational. It is ontological. Western science assumes that knowledge is a representation of an independently existing reality that can be captured in propositions, tested through controlled experiments, and accumulated through publication. Indigenous epistemologies often assume that knowledge is constituted through relationship—between knower and known, between community members, between generations, between humans and more-than-human beings. The knowledge is not a representation to be extracted but a set of practices to be performed, relationships to be maintained, protocols to be honored. Extracting it means changing its nature. An AI system trained on extracted indigenous knowledge does not amplify indigenous knowledge. It amplifies a degraded translation.

The implications extend beyond indigenous contexts to any knowledge that is relational, tacit, or practice-embedded. Patricia Benner's expert nurses know things they cannot articulate in propositional form. Matthew Crawford's master mechanics diagnose through embodied perception that resists codification. Alasdair MacIntyre's virtuous practitioners develop judgment through immersion in traditions that cannot be compressed into rules. These are all forms of knowledge that AI cannot amplify without transforming—and transformation into what the amplifier can process may destroy what made the knowledge valuable. The machine does not merely fail to hear certain frequencies. The frequencies themselves may be constituted in ways that resist the signal-processing the machine performs.

Origin

Srinivasan's engagement with indigenous knowledge began with his doctoral research at MIT (2003-2006), collaborating with Zuni Pueblo on the design of a cultural heritage database. The Zuni community's concern was not technical but epistemological: could a database respect the cultural protocols governing which knowledge could be shared publicly and which must remain restricted? The research produced Srinivasan's realization that technology design is never culturally neutral—that every database structure, every interface convention, every assumption about access embeds cultural values that may conflict with the values of the communities the system is meant to serve. This recognition shaped his subsequent fieldwork in Oaxaca, Bolivia, India, and his development of participatory design methodologies that position indigenous communities as authorities over their own knowledge and its technological representation.

Key Ideas

Relational knowledge resists extraction. When meaning is constituted by connections between elements rather than by the elements themselves, extracting the elements destroys the knowledge—leaving data points severed from the relationships that gave them significance.

Oral traditions and the limits of text. Knowledge transmitted through storytelling, embedded in performance, governed by protocols of who may speak to whom—this knowledge cannot be captured in written form without violating the transmission protocols that constitute its legitimacy.

Holistic knowing vs. disciplinary silos. Indigenous systems that integrate astronomy-ecology-agriculture-social practice into unified understanding cannot be decomposed into Western disciplinary categories without losing the integrative wisdom that is the knowledge's distinctive contribution.

Cultural protocols as knowledge architecture. Restrictions on who may know what, when, and in what contexts are not obstacles to knowledge transmission but its organizing principles—ensuring appropriate use, preventing depletion, maintaining the relationships that sustain the knowledge across generations.

Preservation through non-amplification. Some knowledge is better served by protection from the amplifier than by inclusion in it—requiring institutional support for knowledge forms that resist digitization and recognition that silence in the training data is not absence but a different form of presence.