The Institutional Capacity Gradient and the Bidirectionality Criterion
The Institutional Capacity Gradient (ICG) framework holds that the effectiveness of policy interventions varies systematically with the institutional capacity of the implementing context. Applied to cultural governance, it yields a moral criterion — bidirectionality — that distinguishes governance from predation: institutional spaces sustain cultural governance only when they deliver to both the state and the community through the same channels.
Policy mechanisms must match the capacity of the context, not the standards of the deploying institution. Low capacity ≠ deficit; it determines which mechanisms are available.
Ambiguity sustains governance only when institutional spaces extract and provide through the same channels. Unidirectional extraction = predation. Applied here to AI cultural projects.
When all four state frames — leader, regime, administration, commonwealth — collapse simultaneously, even community-level cultural governance infrastructure is destroyed, not merely absent.
Calibrating policy mechanism and AI governance to institutional capacity level
Legislative protection + state-led regulatory oversight. Four-tier national registry, designated inheritor (传承人) system, Cultural Ecological Protection Zones.
State-directedAI deployed by professional institutions (national libraries, cultural ministries) with full infrastructure support. Community access rights formally guaranteed.
Gallica model referenceStandards-setting and best-practice dissemination. Establish China as non-Western reference model, weakening implicit equation of high-capacity governance with Western liberal design.
Normative coordinationCommunity–state hybrid co-governance. Traditional authorities co-validate AI training data and outputs. Cultural AI infrastructure embedded within state administrative capacity-building programs.
Capacity-generatingShared validation architecture. Kinyarwanda NLP program (early 2020s) as model: cultural infrastructure built for administrative efficiency generates community digital resources as byproduct.
Bidirectional exchangeActive technology transfer and South–South knowledge exchange. Rwanda's experience directly applicable to other mid-capacity African states; more relevant than any European model.
Technical assistanceCommunity-Controlled Cultural AI. Bypass absent state intermediary. Community as primary governance actor. Goal: rebuild Xeer-equivalent community governance structures through which cultural knowledge can be transmitted.
Community-sovereignOffline-first architecture (TensorFlow Lite, Masakhane model). Community holds data selection, validation, and access rights. No cloud dependency. Masakhane: 30+ language benchmarks, community-as-researcher model.
Offline-first · No cloudDirect community empowerment: fund technical capacity training for community members, support offline-first AI deployment, establish legal frameworks for community data sovereignty independent of absent states.
Community empowermentThe moral boundary separating cultural governance from cultural extraction
Extracts: compliance, labor, performed allegiance, cultural data, linguistic corpora
Receives: justice, infrastructure, civic standing, administrative capacity, data sovereignty
Extracts: oral knowledge, training data, model weights, archival content, linguistic corpora
Receives: nothing — no governance rights, no data sovereignty, no access, no benefit-sharing
Tiered Digital Cultural Governance Protocol — four mandatory requirements
Before any AI cultural project, a standardized ICA must assign the context to one of three tiers. Determines which governance model is mandatory. Prevents default deployment of Tier 1 mechanisms in Tier 3 contexts.
Data sovereignty agreements signed with communities as primary parties, not national governments as proxies. Specifies ownership, access conditions, revenue sharing, repatriation obligations.
All Tier 2/3 AI systems must function without stable internet or continuous cloud access. Forces project architects to build for infrastructure realities of target communities, not deploying institutions.
Community verification of AI outputs is a mandatory design component, not optional QA. Validation protocols designed for community contexts. Masakhane model: community-as-researcher, not passive informant.
Implementation: TDCGP compliance module added to UNESCO Periodic Reporting template. States parties document tier assignment, ICA process, data sovereignty terms, offline-first compliance, and community validation outcomes.