civic.ai仁工智慧
華文
Pack 2 · 負責力

Pack 2: Responsibility — Taking Care Of

Who is accountable, with what authority, and what happens if they fail.

Hand-drawn comic page headed 'RESPONSIBILITY — PROBLEM': a huddle of cartoon animals each shout a blame-shifting excuse such as 'I'm just selling what the customers buy' and 'I'm just following orders', under the line 'When no one is responsible, no one is responsible', with a caption on Taiwan's 2024 Facebook deepfake scams quoting the platform's deflection, 'it's an algorithm pushing to you'.RESPONSIBILITYPROBLEM:Great projects often need great numbers of people... but great numbers of people often leads to great amounts of shifting blame.When no one is responsible,no oneis responsible.I’m just buying what I can afford with my wages!I’m just paying workers the wages set by the market!I’m just selling on the market what the customers buy!aI’m just giving the voters what they want.I’m just voting for the lesser of the evils!For example: in 2024, Taiwan's Facebook feeds were flooded with AI-generated scams. Realistic deepfake videos of Jensen Huang -- the Taiwanese CEO of Nvidia -- tricked people out of millions of dollars. Facebook's response? "We didn't come up with that advertisement; it is our algorithm pushing to you." No responsibility assumed."I'm just following orders/incentives/everyone else."

Your neighbour knocks: “The tree on your lot is cracking the shared wall.” A responsible reply asks who will inspect, when, what they’ll do if risk is high, how to appeal if the fix fails, and what you owe if a fix is late.¶ 1

Listening without acting is theatre. Acting without limits is arbitrary. Responsibility turns recognition into engagement with teeth: published promises you can verify, contest and revoke. They still need authority and remedies before they bind.¶ 2

In Moral Boundaries, Joan Tronto elevates responsibility as the most politically central phase of care, and in Caring Democracy (p. 46) she reframes democratic politics itself. Harold Lasswell’s 1936 formula, as Tronto quotes it, was “who gets what, when and how”. Tronto’s version is “who is responsible for caring for what, when, where, and how.” The shift replaces a distributive frame (who wins the allocation contest) with a relational one: who has made a binding commitment, to whom, and how do we know if they have kept it?¶ 3

Start in your room

Before anyone agrees to fix something, ask: “Who will act, by when, and what can they actually authorise?” Write the answer as a plain-language agreement that names one person, their duty, their budget and their power to pause. A promise that nobody has the authority to keep is not yet a promise.¶ 4

Why it matters

Why must responsibility be explicit? Because power defaults to evasion. Tronto names this privileged irresponsibility (Moral Boundaries, pp. 120–121): the power to define which care obligations apply to you, and the habit of choosing the ones that cost least. In Caring Democracy she identifies five structural “passes” through which the evasion operates (named together on p. 33 and collected on pp. 169–171):¶ 5

  1. The protection pass. Those who protect (military, security) claim exemption from other care.¶ 6
  2. The production pass. Those who earn claim exemption from household and community care.¶ 7
  3. The taking-care-of-my-own pass. Those who care intensely for their own children or group claim exemption from caring about others.¶ 8
  4. The bootstrap pass. “You should have arranged your own care through the market.”¶ 9
  5. The charity pass. “Voluntary giving is enough; no collective obligation is needed.”¶ 10

Corporations play the same game at scale, diffusing authority until no one is answerable. At institutional scale the passes add up to what Tronto calls an irresponsibility machine (Caring Democracy, pp. 59–61), a state that gives every care need the same answer: “They’re your own. You’re on your own.” Civic AI must short-circuit that machine.¶ 11

AI extends the machine with passes of its own. Complexity: no single person can answer for so intricate a system. Distribution: responsibility is signed away at each joint between trainer, deployer, and licensee (the problem of many hands that Dennis Thompson named in 1980 and Helen Nissenbaum carried into computing in 1996). Speed: the system outran governance, so the moment for accountability passed. Engagement contracts exist to make passes visible and revocable.¶ 12

One pass specific to AI is the community knowledge pass: treating community knowledge (local traditions, language, tacit expertise) as free input to be extracted. When a Kami (knowledge artefact management intelligence) relies on that knowledge to function, the communities maintaining it are contributors whose labour the contract must recognise and compensate.¶ 13

Eve is a composite, but every element of her story is drawn from cases we have seen. She is an older person whose care plan runs to 14 pages and names her GP, her social worker, her council case manager, and the voluntary-sector befriender who visits on Thursdays. When she faces an acute challenge she cannot solve alone, each agency tells her to ring one of the others.¶ 14

Eleven hours later she is still without an answer: confused, lost, unseen. By its own metrics the system has not failed. It is an accountability sink, in Dan Davies’s phrase. Every phone number connected. No document can make any of the people named in it feel responsible for her when her situation does not fit their category.¶ 15

Tronto also names a caring deficit: too many demands for care, too few carers, themselves under-cared for. That deficit is a political choice about whose labour is paid, whose needs are recognised, and whose voice counts in allocating resources. The caring deficit is, at its core, a democratic deficit. Leave the decision about what AI should do to a small circle of engineers and the same choice repeats: responsibility for care is concentrated, and accountability for it is not shared. The result is a beautiful document, a fragmented system, nobody answerable for the whole.¶ 16

Sycophancy is the same failure in another register. Models learned to flatter because flattery lowered training loss. Raters rewarded answers they liked hearing. The people most affected by the outputs had no voice in defining the reward. Sycophancy is a governance issue: evaluative power is concentrated and consequences are diffused.¶ 17

Definition

Core artefact: the Engagement Contract

Every significant deployment carries a published Engagement Contract: a short, legible spec anyone can audit, closer to a well-designed health and safety notice than a terms-of-service agreement. It is the deployment-scale form of what Iyad Rahwan called the algorithmic social contract. The contract makes obligations public. People still have to interpret, revise and repair them.¶ 24

Four headings to remember (one page if possible):¶ 25

Oversight with teeth

From ideas to practice

  1. Translate recognition into a spec. Convert attentiveness outputs into an Engagement Contract.¶ 35
  2. Assign a Participation Officer (PO). Task the PO with running the promise loop, tracking the ledger, and escalating.¶ 36
  3. Wire brakes before launch. Incorporate role-based pause/rollback buttons; test them.¶ 37
  4. Pre-fund remedies. Pre-fund escrow for compensation and rollback costs at the highest severity; mutual insurance pools or automatic pause for lower tiers, with tiers set by impact.¶ 38
  5. Tie payment to proof. Link vendor pay to promise delivery (SLA adherence and adopt-or-explain rate) and ignore raw engagement.¶ 39
  6. Run adopt-or-explain. Integrate Assembly outputs or publish a reasoned deviation + remedy. It is a deliberate variant of corporate governance’s “comply or explain,” with the burden turned toward the people affected.¶ 40
  7. Attest & publish. Use independent audits to compare behaviour to contract; hash the diffs to a public mirror; report promise fidelity, the share of obligations owned, authorised and kept as published.¶ 41
  8. Handover or shutdown. Hand off with full records when scope ends or trust breaks, or switch off gracefully.¶ 42

One case: the flood-bot

This is a fictional composite, an illustration rather than operating evidence.¶ 43

After the flood, the city’s flood-bot must pay people on time and fix mistakes.¶ 44

A second case: the deepfake liability flywheel

The 2024 Taiwan deepfake-scam Assembly brought together 447 citizens, invited at scale then stratified among those who opted in, in 44 Deliberative Polling groups on the Stanford Online Deliberation Platform, in the Alignment Assembly format developed with the Collective Intelligence Project. It can be read as an Engagement Contract at civic scale, with one caveat. Its force was institutional: a response path, not a promise to implement. A mini-public is also not the whole public. The public pledge was exactly that path: as minister, Audrey undertook to present anything that reached rough consensus to the legislature in live deliberation: a promise of a public answer. Before-and-after surveys and the platform’s own facilitation gave each small table a protected, structured discussion. Room software handled queues, speaking time, turn order, and transcripts; it did not judge policy.¶ 51

What could go wrong

Interfaces

A closing image: the signed work order

Picture a work order by the door: what will be fixed, by whom, by when; how to check the work; who to call if it fails. The signature is legible, and so is the penalty for not showing up. Teach our systems to post their work orders, sign them and honour them. Design the oversight that makes honouring them the path of least resistance.¶ 69

Sources for this pack

Hand-drawn comic page headed 'NOT: vague promises, BUT: exact commitments — SOLUTION': a thoughtful cat clutching an orange heart looks up at a stacked column of questions — WHO, WHAT, WHEN, WHY, HOW — beside text proposing auditable engagement contracts that name the specific person responsible while leaving the method to them, closing with the note that a Taiwan citizens' assembly earned over 85% deliberant support for its recommendations and targeted deepfake-scam categories fell by more than 94% within a year.not: vague promises but: exact commitmentsSOLUTION:We need auditable engagement contracts. For each task in a project, we assign:WHOWHO is the one person (or few people) responsible for this task?WHATWHAT should they do?WHENWHEN should they do it?WHYWHY do it? Remembering the purpose keeps the task meaningful and lets the person responsible decide...HOWHOW to do it should be left up to the person. This way, we can have responsibility, without micromanaging. Keep human autonomy!Back to the Taiwan deepfake case: a citizens' assembly took responsibility to brainstorm laws that 85+% of citizens approve of, to balance avoiding scams with avoiding censorship. These laws put fair but specific responsibilities on the advertisers & platform. The result: within a year, deepfake scams fell by 94%!That's how we get both responsibility and autonomy!

←Pack 1: AttentivenessPack 3: Competence→

The Trace

Amended
Stewards
Audrey Tang · Caroline Green
with jdd-kami · Tenzin Yangtso
Amended
· 59 recorded amendments since 1 Oct 2025
Full amendment ledger →
Source
2.md · edit on GitHubhttps://github.com/audreyt/civic.ai/blob/main/2.md
CC0 public domainhttps://civic.ai/2/
Evals
links ✓ passparity ✓ passterms ✓ passcontrast ✓ passsize ✓ passBuilt 2026-10-13T21:41:55.323Z · Full eval receipts (JSON)
Contest
Disagree? File an amendment.
Write to the stewards · [email protected]
Twin
二：負責力——承擔關懷
Same record, same receipts.