Multiple centers of decision authority
Each community runs its own procedure under shared ground rules. A mechanism is intended to settle matters that cross between the separate centers.
Scroll the diagram sideways to see all of it.
aiag/procedural/polycentric_governance.pyimport aiag
# The 4 approaches of Polycentric governance, with the classes that carry them out
for approach in aiag.approaches(locus="proc", conflict="partition"):
print(approach)pip install aiagaiag/procedural/polycentric_governance.py Every method in aiag Library on GitHub
Limitation
Each community's procedure works locally, but nothing settles questions that span communities. On those questions, the system has no answer. In addition, since users can move to whichever center suits them, centers compete for members by loosening their rules. This leads to a race toward the most permissive standard.
A separate standard is fit to each group, rather than pooling everyone's feedback into one standard. A response meant for group A is therefore judged by A's standard and never by a population average. The groups can arise in two ways, and the difference matters. They can be given, as real, bounded communities whose norms are inferred from what they already accept and engage with, rather than from labels or a written charter. Or they can be discovered, as latent clusters found within pooled feedback. Such clusters answer to no institution that anyone can join, contest, or leave. The argument for this approach is that the authority to say what "aligned" means belongs with many contextual, participatory centers rather than with one lab. Each per-group standard is meant to be one such center. One paper here belongs only partly to this regime. It sets out a personalization family (several reward models, built by representation learning or clustering) alongside an aggregation family that combines those models back into a single pipeline under utilitarian and Leximin rules. Only the personalization half places it in this regime.
Each group trains on its own members' feedback where that feedback is already stored. It then sends out only the result of training (model updates, raw parameters, or compact selectors that stand in for preferences). No group's data leaves the group, and no pooled corpus is ever assembled. This approach differs from the approach that fits a separate standard for each group in what is partitioned. Here, it is the pipeline (the data and the computation), and not necessarily the number of standards at the end. This distinction has consequences in both directions. One paper trains a personalized policy for each client, and so also meets that approach's test. It is placed here because its defining feature is the federated architecture. Another paper keeps data local but converges, by averaging, on a single shared preference predictor. Its plurality lies entirely in the process (who holds the data, who computes, and who can withhold or fork) and not in the outcome. This process-level plurality is a real lever that a contributor to a pooled corpus does not have, which is why these papers belong in a procedural regime. However, the single-model case has the weakest claim to partition among these papers.
Each online community writes its own procedures (e.g., votes, juries, term limits, and appeal steps) as executable policy attached to actions on the platform. A shared runtime then executes whatever each community wrote. The platform supplies the engine and the hooks but takes no position on the rules, so neighboring communities on the same platform can run incompatible constitutions. The goal is a portable, composable governance layer, with components that can move from one context to another, rather than a single set of platform-wide rules.
A legal body is placed between the people the data belongs to and anyone who wants to use it. The body holds the rights, owes duties to the people it holds them for, and sets the conditions under which the data may be accessed, licensed, or used for training. A corpus can therefore be assembled only on that body's terms. Two forms appear here, and they work differently. In the first, member-owned custodians multiply, either as trusts, each under its own deed, or as cooperatives with a fiduciary obligation to their citizen members. Each has different terms, and a member's operative power is exit, i.e., leaving one custodian for another. The second form, from the third paper, is a single negotiated multi-party structure for the world's language data. Its plurality is internal (among stakeholders, values, and rights), and it has no deed and no exit. It is the loosest fit for a regime built on partition. It is included because it is the only paper that applies the custodian idea directly to language-model training corpora. The trust and cooperative papers, written for personal data and member services, only leave this extension open.
Many independent decision centers each act on their own, but under one common framework of rules that all of them acknowledge. This shared framework is what makes the centers a single order rather than a scattered collection.
The authority of the centers overlaps on purpose, so the same case can fall under more than one of them. Conflicts between the centers are treated as useful and are worked out through interaction, rather than prevented by a clear division of jurisdiction.
Decisions are made at the most local level capable of making them. Local units nest inside larger ones that handle what the local level cannot. Authority is layered, and it is neither centralized nor fragmented on one level.
The Logic of Liberty
1951
The Organization of Government in Metropolitan Areas
1961
Governing the Commons
1990
Polycentricity: From Polanyi to Ostrom, and Beyond
2012
Politica Methodice Digesta
1603