A governance framework by Audrey Tang and Caroline Green at Oxford's Institute for Ethics in AI that translates Joan Tronto's political ethics of care into six machine-codeable design primitives for Civic AI — deliberately bounded, purpose-specific AI stewards engineered to nurture a community's relational health rather than maximise abstract global metrics.
Standard alignment approaches try to derive values from data — but no amount of "is" produces an "ought." Care ethics sidesteps this by starting, as Tronto puts it, "in the middle of things": within an existing commitment to democratic values, asking what those commitments demand once we take mutual dependence seriously. The result is alignment-by-process — a continuous, democratically governed practice rather than a one-time engineering solution.
Where the vertical narrative of a technological singularity concentrates power in a single unbounded optimiser, the 6-Pack proposes a horizontal alternative: plural stewardship by many bounded intelligences — local kami — in close interaction with human communities. Each kami is bound to a specific place and purpose; its success is the health of the relationships it supports, not indefinite expansion.
| # | Pack | Tronto phase | Core question |
|---|---|---|---|
| 1 | Attentiveness | Caring about | What do the people closest to the pain notice that we're missing? |
| 2 | Responsibility | Taking care of | Who is accountable, with what authority, and what happens if they fail? |
| 3 | Competence | Care-giving | Does the system demonstrably work — audited, explainable, safe-to-fail? |
| 4 | Responsiveness | Care-receiving | Can those affected correct the system, and does correction actually change it? |
| 5 | Solidarity | Caring with | Does the ecosystem structurally reward cooperation over lock-in? |
| 6 | Symbiosis | Kami of care | Is the system bounded, sunset-ready, and incapable of imperial creep? |
Packs 1 – 4 form Tronto's feedback loop. Pack 5 (from Caring Democracy) ensures the loop operates within democratic commitments to justice, equality, and freedom. Pack 6 is Tang and Green's addition: the anti-Singleton architecture that keeps care local, bounded, and provisional.
Also: Measures (one metric per pack) and FAQ (speed, cost, bad actors, and how the framework handles them).
- Academics. A rigorous "alignment-by-process" theory grounded in Tronto, Margaret Urban Walker's expressive-collaborative morality, and Ostrom-style polycentric governance. Navigates the Is-Ought problem without collapsing into thin universalism or cultural relativism.
- Policymakers. Hard governance levers: citizen alignment assemblies, structural data portability (Utah Digital Choice Act), escrow-backed engagement contracts, bridging-based ranking transparency, and federated trust-and-safety (ROOST).
- System designers. Actionable technical scaffolding: bridging algorithms (PCA/embedding overlap), shadow/canary orchestration with rollback, decision-trace schemas, community-authored eval registries (Weval), RLCF reward pipelines, and guardrail-as-code engines.
- Bridging algorithms. Recommenders score content by cross-group endorsement, not outrage. Smaller, coherent clusters earn higher bridging weight because they are harder to reach.
- Engagement contracts. Published, auditable specs — purpose, SLAs, pause triggers, remedies, sunset — that make power accountable and irresponsibility visible.
- Kami model. Every agent has purpose bounds, resource caps, and a sunset timer. The component sunsets; the service duty persists through succession.
- RLCF. Reinforcement Learning from Community Feedback — training agents to optimise for cross-group endorsement and trust-under-loss, not raw engagement.
- Meronymity. Partial anonymity: verify an agent is anchored to a real entity without exposing the person.
6pack.care — bilingual (British English / Traditional Mandarin) static site.
Built with Eleventy v3 and Bun:
bun install # install dependencies
bun run dev # local dev server at http://127.0.0.1:4000
bun run build # production build → ./docs/Pull requests are welcome. By contributing, you agree to release your work under the CC0 1.0 Universal public domain dedication.
When editing content, maintain parity between English (*.md) and Traditional Mandarin (tw-*.md) variants.
Part of the Accelerator Fellowship Programme, Oxford Institute for Ethics in AI.