Key Terms
Alignment Assembly. A structured deliberative process in which a randomly selected, statistically representative group of citizens deliberates on a specific policy question with AI-assisted sensemaking tools, producing recommendations that the convening authority has pre-committed to present to the legislature. The 2024 Taiwan deepfake assembly is the primary example in this book.
Boundedness. The design principle that an AI system should serve a specific community, in a specific place, for a specific purpose, with no aspiration to expand beyond what the community needs. Derived from the kami metaphor. The opposite of unbounded optimisation.
Bridge index. The headline public measure for solidarity (Chapter 7). Tracks whether shared decisions show genuine cross-group participation and co-endorsement rather than separate silos.
Bridging map. A visual representation of where participants agree, where they disagree, and where apparent disagreements dissolve under closer examination. Produced by tools like Polis. A tool of attentiveness.
Care cycle. The feedback loop formed by the first four packs: Attentiveness → Responsibility → Competence → Responsiveness → back to Attentiveness. Each phase generates information and obligations that feed the next.
Care-washing. The use of care language — "we care about our users," "community-first" — to disguise extractive practices. The civic equivalent of greenwashing.
Caring deficit. Joan Tronto's term for the structural condition in which there are too many demands on people's time to care adequately for children, older people, and one another. Not a natural condition but the product of political choices about how to organise work, markets, and public life.
Civic AI. AI designed and governed as civic infrastructure — attentive to particular communities, responsible to specific relationships, bounded by the needs of the people it serves. The alternative to extractive AI built for infinite scale.
Civic Care Licence. A public rulebook encoding scope, consent rules, portability requirements, and shutdown duties for a deployed kami. The machine-enforceable expression of the engagement contract.
Data as oil. The extractive paradigm in which collective culture, languages, and tacit knowledge are treated as resources to be drilled, refined, and locked behind proprietary APIs. Depleting and non-regenerative.
Data as soil. The restorative alternative in which data is a community asset to be tended and cultivated. AI models are moved into local contexts and governed by the communities that produce the data. Regenerative.
Engagement contract. A publicly posted agreement specifying who is responsible for what, with what authority, for how long, and what happens if they fail. The core artefact of responsibility (Chapter 4).
Geothermal engine. A metaphor for treating polarisation not as a fire to extinguish but as an energy source — harnessing the heat of disagreement to generate co-creation by amplifying overlap rather than outrage.
Kami. A concept from Japanese Shinto tradition describing a helpful spirit associated with a particular place — a river, a tree, a mountain. Used in this book as a design metaphor for bounded, local, purpose-specific AI systems that serve their community and recede when their care is no longer needed.
Kami of Care. An AI system designed according to the 6-Pack principles: bounded, local, attentive to a specific community, accountable through engagement contracts, and sunset-ready. The antidote to the Singularity vision.
Mohui (默會知識). Tacit knowledge — knowledge held in the hands and habits of human practitioners that cannot easily be written down as a manual. The kind of knowledge that the Public Master model is designed to capture and transmit.
Perspective receipt. A record that allows each contributor to find and verify how their input was represented in a summary or decision. A tool of attentiveness that builds trust by making the process transparent.
Plurality. The vision of AI as augmenting cooperation across human differences rather than flattening those differences. A horizontal alternative to the vertical singularity, in which many bounded local intelligences tend their own communities rather than a single centralised intelligence governing from above.
Privileged irresponsibility. Joan Tronto's term for structural arrangements that allow those with power to avoid the work of care. Three common passes: the protection pass (security concerns exempt from care duties), the production pass (earning exempts from household/community care), and the bootstrap pass (everyone should arrange their own care through the market).
Public Master (公共師傅). A paradigm in which AI serves as a digital apprentice alongside a human master, learning by observing and recording decision traces. Governed as a public asset by intermediate local institutions rather than owned by a vertical monopoly.
Relational health. The quality of human connection that civic AI aims to support and that extractive AI tends to degrade. Encompasses three relationship types: people-to-people (how AI mediates human relationships), people-to-AI (direct interaction), and AI-to-AI (interoperability and federated safety).
Representation gap. The headline public measure for attentiveness (Chapter 3). Tracks which materially affected groups are still missing or badly under-represented in the record.
Satisficer. A system that seeks to meet needs adequately rather than to optimise toward a single maximum. Contrasted with an optimiser. The ethics of care favours satisficing design.
Singularity. The vision of a single, centralised, unbounded superintelligence that optimises humanity from above. The book's primary alternative.
6-Pack of Care. Six design principles that translate care ethics into institutional practices for AI governance: Attentiveness (Chapter 3), Responsibility (Chapter 4), Competence (Chapter 5), Responsiveness (Chapter 6), Solidarity (Chapter 7), and Symbiosis (Chapter 8). Not a consumer product but muscles to be trained.
Thick alignment. Alondra Nelson's concept (drawing on Clifford Geertz) for AI alignment that engages with the full social context of human values — cultural distance, power relationships, community-specific needs — rather than merely satisfying stated preferences. Contrasted with thin alignment.
Thin alignment. A narrow, technical approach to AI alignment that focuses on getting the objective function right without engaging with the deeper layers of meaning, context, and power that shape human values.
Trust-under-loss. The headline public measure for responsiveness (Chapter 6). After a bad outcome and attempted repair, do affected people report that the system became more trustworthy rather than less?
Uncommon ground. Ideas that people with different starting views can still find reasonable. Not the centrist average or the consensus position, but the proposals that earn cross-group endorsement. The output of a well-facilitated bridging process.