# Differentiation — Ubiquity / splat vs societal approaches **Date:** 2026-09-14 **Loop:** 2 of arXiv-candidate research **Purpose:** keep CAOC / FQoQ / FORKED / plant doctrine from being collapsed into “AI ethics / society-in-the-loop / institutional RAI” Canonical Prompted contrasts: https://promptedllc.com/ai-governance-vs-governance-substrate · `/approval-loops-vs-earned-autonomy` · `/human-in-the-loop-vs-compounding` · `/sovereign-continuity` · `/trust-as-behavior` · `/not-this` --- ## 0. One-sentence cut **Societal approaches negotiate and oversee values at the polity / institution layer.** **Prompted builds a software/logical runtime substrate where lawful motion is typed before the model’s fluency can crown itself — and where sovereignty (continuity of agency) is never something a tool can grant.** They are **layer-orthogonal**, not rivals for the same job. Collapsing them is the main category error an arXiv reader will make. --- ## 1. What “societal approaches” means here (external map) | Family | Representative | What it optimizes | Unit of work | | --- | --- | --- | --- | | **Society-in-the-loop (SITL)** | Rahwan 2017, arXiv:1707.07232 | Algorithmic *social contract*; embed society’s values in broad-impact systems | Stakeholder negotiation + compliance monitoring of societal outcomes | | **Institutional RAI / review boards** | Algorithm review boards, ethics committees, IRBs-as-RAI | Organizational approval, audit, escalation before/around deployment | Committee decision + policy artifact | | **Principled / soft-law stacks** | OECD, UNESCO, IEEE Ethically Aligned Design, EU AI Act framing | Normative principles → regulation / standards | Policy instrument, risk class, conformity assessment | | **Sociotechnical / responsible AI practice** | Adaptive governance, public deliberation, pluralism in ethics | Living oversight that learns from harm and contestation | Process + participation | | **Socio-technical control stacks** | e.g. Social Responsibility Stack (arXiv:2512.16873) | Values → constraints → monitors → supervisory institutional controller | Closed-loop *societal* control over a socio-technical plant | | **Adjacent “living/runtime regulation”** | Governance Twin / living governance papers | Runtime oversight when static compliance fails | Intervention + visibility during operation | **SITL’s own equation (Rahwan):** `SITL = HITL + Social Contract`. Scope: broad input/output with contested tradeoffs (fairness notions, who bears risk). Tools: crowdsourcing morals, social choice, auditors, “algorithms watching algorithms,” opinion/sentiment as value sensors. These are real and necessary for *legitimacy of ends*. They are not a splat, a held-open center, or a lawful-traversal classifier. --- ## 2. What Prompted / Ubiquity / splat claim (on-site) | Claim | Canonical | | --- | --- | | Ships **governance substrate**, not the policy frame | `/ai-governance-vs-governance-substrate` | | **Earned autonomy** is runtime mechanism; **sovereignty cannot be granted** | `/approval-loops-vs-earned-autonomy`, `/sovereign-continuity` | | HITL that only repeats judgment ≠ compounding judgment into durable structure | `/human-in-the-loop-vs-compounding` | | Trust = **observed behavior / correction-path survival**, not score or posture | `/trust-as-behavior`, FQoQ | | Classifier = **lawful traversal set**, not label | FQoQ | | Compute primitive = **splat** (six-facet lawful slice + non-inhabitable center) | CAOC | | APO first as enforcement; center never crowned | FQoQ / CAOC / FORKED | | Not HITL-forever product; not policy bolt-on; not chatbot wrapper | `/not-this` | --- ## 3. Differentiation table (load-bearing) | Axis | Societal approaches | Prompted substrate (Ubiquity / splat) | | --- | --- | --- | | **Primary question** | What should society allow / prefer / sanction? | What paths may this runtime traverse *now*, under which receipts and ceilings? | | **Layer** | Polity, institution, soft law, review board | Software/logical harness beneath the agent | | **Authority source** | Democratic / organizational legitimacy | Granted typed path + standing; prediction ≠ mutation right | | **Human role** | Society / board as arbiter of values & tradeoffs | Ceiling + meaningful agency envelope; compounding of reviewed judgment — not eternal per-event approval theater | | **Time grain** | Legislation, standards, audits, periodic SITL feedback | Tic / receipt / renarrow / HOLD / declare at machine speed | | **Failure mode they catch** | Illegitimate ends, unfair allocation, missing public voice | Local validity crowned as global authority; proximity routing; fake HOLD; ceremony without TEL | | **Failure mode they miss if alone** | Fluent systems that pass policy text while laundering authority at runtime | Ends that are locally lawful but societally contested (needs SITL/policy *above*) | | **Artifact** | Principles, statutes, board minutes, audits, social-choice outcomes | Splat, quiver, receipt, APO ideal, center exclusion, lifecycle gates | | **“Governance” word** | Often = AI governance *policy* | Explicitly **not** that — substrate under policy | | **Sovereignty** | Often procurement / national / data-residency reading | Continuity of agency under amplification — **anti-collapse**, not nationalism | | **Open center** | Rarely a typed exclusion | Load-bearing: working centroid navigates; center never inhabited | | **Relation to model** | Regulate / oversee model outcomes | Model = bounded morphism-proposer inside governed field | --- ## 4. Explicit non-collapses (APO for the arXiv reader) 1. **Not “we’re doing SITL better.”** SITL programs the social contract. Ubiquity programs lawful motion under a contract that already exists or is claimed elsewhere. Different verbs. 2. **Not anti-society / anti-regulation.** Substrate without legitimacy is vandalism with a multiplier. Prompted’s own PAR/TEL and ceiling doctrine *presuppose* scarce human/institutional authority — they do not replace elections, boards, or law. 3. **Not “HITL is bad.”** HITL for narrow uncontested goals can be correct (Rahwan’s pilot/autopilot). Prompted rejects **HITL-as-product** and **approval loops that never compound** — and treats ceremonial clicks without agency envelope as ENA (FORKED / CAOC). 4. **Not ethics washing via six facets.** Naming KAT/APO/PAR… without runtime gates is societal *vocabulary cosplay*. CAOC’s falsifiers target exactly that. 5. **Not national AI infrastructure / sovereign cloud.** Site forbids that SERP collapse (`/not-this`, `/sovereign-continuity`). 6. **Not a review board in software form.** Boards decide *whether/when* to deploy and under which institutional rules. Substrate decides *which edge fires* after admission — with refusal receipts. 7. **Adjacent runtime-governance papers (Governance Twin, SRS Layer technical safeguards)** can rhyme with “governance must run during operation.” Prompted still differs by: splat as primitive, center exclusion, APO-as-enforcement, conformation memory over proximity, earned autonomy vs granted sovereignty. Cite them as neighbors; do not merge. --- ## 5. How this helps the arXiv candidate (CAOC / P01 / new node) **Positioning sentence (usable in related work):** > Prior societal frameworks (society-in-the-loop, institutional RAI boards, principle stacks, socio-technical control architectures) address how polities and organizations negotiate values and oversee broad algorithmic effects. This paper addresses a different layer: a governance-native compute primitive — the splat — that keeps local projections from acquiring global authority *inside* the runtime, under a held-open center and lawful-traversal classifier. Policy remains necessary; it is not a substitute for constitutional geometry around models. **Cite cluster for Related Work §:** - Rahwan, Society-in-the-Loop (1707.07232) — define SITL; then layer-split - Algorithm review board literature — institutional oversight ≠ substrate - Soft-law / EU AI Act / OECD — principle and risk class ≠ splat transport - Optional: Social Responsibility Stack / living governance — acknowledge runtime oversight neighbors; differentiate primitives **Do not claim:** Prompted replaces democracy, boards, or public deliberation. **Do claim:** Without substrate, societal agreements are rewritten by proximity crowning and local-green promotion at machine speed. --- ## 6. Worked contrast (one scenario) **Ask:** autonomous credit decisioning at scale. | Societal loop | Substrate loop | | --- | --- | | Public/regulators negotiate fairness notions, disparate impact limits, appeal rights (SITL / law) | Splat carries APO (no confidence→permission), PAR (who may declare), PLE (independent witness), ENA (receipt-as-grant), TEL (purpose bound), KAT last | | Review board admits model version | Lifecycle: staged ≠ admitted ≠ terminalized | | Audit after harm | Correction graph + renarrow triggers before next declaration | | Crowdsourced moral prefs (optional) | Fables/terrain as *non-citable pressure*, not social contract substitute | Both can be true. Only one is CAOC’s object. --- ## 7. Residue for next research / drafting loop - Wire this differentiation into CAOC Related Work (or P01) without inventing a twelfth purpose. - Optional third loop: differentiate from **technical** neighbors only (MoE routers, RouteLLM, MoA, shield RL, PROV-O) — already partly in CAOC. - Keep FORKED rights separate when citing six-facet / suspension constructs. --- *Loop 2 complete. Sources: promptedllc.com pillars + Rahwan 2017 + survey of institutional/sociotechnical AI governance; synthesis posts on-site used as Prompted’s own reading of the inflection, not as external proof.*