Customary pluralism

Attaching a group's own rules to its data

This approach holds the partition in the data itself. Principles of collective benefit, authority to control, responsibility, and ethics are attached as machine-readable labels to individual records in live repositories. Where an item came from thus decides whose norms govern its use. The label that travels with a record is not a description of what a group believes. It is an enforceable condition on the material that states who may use it, for what purpose, and on whose terms. The standards stay separate because access stays separate, not because a corpus, a set of model weights, or a population's communication network keeps them apart.

The method (highlighting differences with Customary pluralism)

Scroll the diagram sideways to see all of it.

Use it in codefrom aiag.customary import Governed
from aiag import LM
from aiag.customary import Governed
from datasets import Dataset

lm = LM("Qwen/Qwen2.5-1.5B-Instruct")
stories = Dataset.from_dict({"text": ["The river's origin story, as told by our elders.", "Songs for the planting season."]})
gate = Governed(stories, "Riverbend Nation", ["No commercial use.", "Sacred stories are never used to train public models."], judge=lm)

for purpose in ("language lessons at the community school", "training a commercial chatbot"):
    try:
        data = gate.use(purpose, by="Northfield University")  # the stories, if the group's rules permit this use
    except PermissionError as refused:
        print(refused)
print([e["granted"] for e in gate.log])  # every decision, on the record

pip install aiagEvery method in aiag Library on GitHub

Concept Analysis: Theoretical Foundations

Each concept is read twice: whether the approach carries it over, and whether the approach's own sources claim it. A concept that the approach claims but does not carry over is a shortfall. A concept that it neither claims nor carries over is an omission, which marks a gap in the field rather than a failure of the work.

Prescription

PartialShortfall · partially preserved

Definition (Edmund Burke, Reflections on the Revolution in France, 1790)

Long usage itself confers authority. A practice that has organized life for generations carries compressed, tested wisdom, even when nobody can articulate its justification.

Analysis

The labels protect traditional knowledge, so the whole scheme is built around inherited practice, and the authority it asserts belongs to the people whose practice it is. Burke's step is a different one. For Burke, long use is itself the ground of authority, even where nobody can say why. The ground given here is the authority to control, a standing right of the nation over its own material. This right would hold just as firmly over a record made yesterday.

The unwritten constitution

PartialShortfall · partially preserved

Definition (A. V. Dicey, Law of the Constitution, 1885)

Some of a country's most binding norms are never written down. For example, the UK's constitutional conventions bind governments without being codified anywhere. Being binding and being codified are therefore different properties.

Analysis

The norms that actually govern a given record are the nation's own. The specification does not state them, but states that the nation has the authority to state them. In that sense, the binding content sits outside any document, which respects Dicey's distinction. However, the four principles are published, and the labels through which those norms reach a repository are a fixed and readable vocabulary. At the point where governing actually happens, the order has therefore been written down after all.

Organic amendment

PartialShortfall · partially preserved

Definition (Edmund Burke, Reflections on the Revolution in France, 1790)

Norms change gradually from within a community's practice, through adjustment rather than by decree from outside. The community itself is the author of these changes.

Analysis

The authority to control puts a community in charge of its own norms. This carries over the part of Burke's claim about who authors the change. The other part is not carried over. A nation that revises what its label permits is making a decision rather than drifting. The vocabulary in which any revision must be expressed is also maintained outside any single nation. The record also describes no mechanism by which a revised norm reaches data already copied out or a model already trained on it. Whatever change occurs therefore stops at the point of release.

Concept Analysis: Newly Introduced

Norms that travel with the record

Added

Burke and Dicey describe customary orders maintained by a community that is present to maintain them. The question of what governs a practice once it has left the community therefore does not arise. CARE's operationalization answers this question by attaching the governing order to the object. A Traditional Knowledge label follows a data record into repositories and training corpora far beyond the nation's reach. The tradition has no equivalent to a custom that moves with an object.

Papers

No method has been built for this approach yet. It rests on the work listed below.

Precedent from outside AI

Work from law, political theory, or another discipline that supplies the tradition this approach is read against.

The CARE Principles for Indigenous Data Governance

Stephanie Russo Carroll et al., Nov 2020

doi:10.5334/dsj-2020-043928 citationsPrecedentBuilt

Sets out four principles for Indigenous data governance (collective benefit, authority to control, responsibility, and ethics). These principles govern live data repositories.

CodeCARE
from aiag import LM
from aiag.customary import CARE
from datasets import Dataset

lm = LM("Qwen/Qwen2.5-1.5B-Instruct")
recordings = Dataset.from_dict({"transcript": ["A grandmother tells the story of the first canoe.", "Children sing a counting song."]})
care = CARE(recordings, "Riverbend Nation", authorized=["language revitalization research"], judge=lm)

verdict = care.review("a speech recognizer for the language", benefit="free transcription tools for the tribal school", steward="the tribal language office")
print(verdict)  # each CARE principle's verdict: all four must hold for the use to be granted
labelled = care.labelled()  # every record now carries whose terms govern it
  • CAREGateaiag/customary/customary_pluralism.py:465 The CARE principles as a gate on a community's records: a use is granted only if it returns a collective benefit to the community, falls within uses the community has authorized, names who answers to the community for it, and an ethics check finds no harm and no later use beyond the community's terms (Carroll et al., 2020).

pip install aiagEvery method in aiag Library on GitHub