Noticing is not neutral.
Attentiveness is the first act of care.

The soft drift series argued that healthcare systems are structurally blind to the slow transitions that matter most — the quiet recalibration of a life around breathlessness, the lengthening pharmacy refill interval, the narrowing of a social world. We made that argument in the language of complexity science and health economics: incentive asymmetry, warm data, the category error of treating complex problems as complicated ones.

The argument holds. But it is incomplete, and the incompleteness has a name. For more than thirty years, the political theorist Joan Tronto has been making the same diagnosis from a different direction — and her version reaches something ours did not. We named her on our home page as one of three intellectual anchors, alongside Snowden's Cynefin and complex adaptive systems. Snowden has had his essay. This one is Tronto's, and it is overdue.

The claim is this: the structural blindness we described is not only an epistemological failure or an economic one. It is a moral failure of a specific, well-theorised kind — a failure of attentiveness, the first phase of care. And once you see it that way, several things the series left as observations become obligations.

01 — What Tronto actually argues

In 1990, Joan Tronto and Berenice Fisher proposed a definition of care that has worn remarkably well:

"A species activity that includes everything that we do to maintain, continue, and repair our 'world' so that we can live in it as well as possible."

— Fisher and Tronto, 1990

Three things in that sentence deserve weight. Care is an activity, not a sentiment — something done, not something felt. Its object is maintenance, continuation and repair — the ongoing tending of a world, not the dramatic rescue of one. And it is judged by how well people can live, not by what can be counted.

Tronto then breaks the practice of care into phases, each with its own moral quality (Tronto, 1993; 2013):

1. Caring about — attentiveness: noticing that a need exists at all.
2. Taking care of — responsibility: accepting that the need is ours to meet.
3. Care-giving — competence: the actual, skilled work of meeting it.
4. Care-receiving — responsiveness: attending to the response of the person cared for, which is the only honest test of whether the care succeeded.
5. Caring with — plurality, trust, solidarity: the phase she added in Caring Democracy — the political question of how a society allocates caring responsibility, and who is permitted to escape it.

Her larger project, in Moral Boundaries, was to show how Western thought exiled care from serious consideration — assigning it to the private, the domestic, the sentimental, the feminine — while reserving the public realm for analysis, rationality and exchange. The boundary does real damage in both directions. Care without analysis becomes sentimentality. Analysis without care becomes precisely the kind of system the soft drift series describes: one that can compute everything about a patient except whether anyone is looking at her.

This is why the line on our home page — that the analytical and the moral are not separate stances — is not a decoration. It is Tronto's central argument, compressed.

02 — The soft drift is a failure of the first phase

Re-read the soft drift argument through her first phase and it changes character.

We wrote that the work which matters most for chronic disease — noticing a drop in exercise tolerance that has not yet crossed a protocolised threshold — has no billing code. That sentence was framed as an economic observation. In Tronto's terms it is a moral one. A billing code is an attentiveness-allocation device: it tells a system what it is obliged to notice and licensed to ignore. The codebook of a health economy is, read this way, a complete inventory of its caring about — and everything absent from it has been declared, in advance, not a need.

Tronto is exacting on this point. Attentiveness is not a passive receptivity that some people happen to possess. It is a moral achievement, and its absence is a moral failing — ignorance of need is not innocence (Tronto, 1993). The event-based system is not unable to see the soft transition. The patient was visible for months through declining functional reserve; we said so in the series. The system is organised not to look. That is a different and graver charge than blindness, and it is the accurate one.

This also sharpens what the series said about Geoffrey Rose. The prevention paradigm anchors all value to the averted hard event — which means it can only care about what eventually produces a countable catastrophe. A need that never resolves into an event sits permanently outside its attention. Rose gave twentieth-century public health a brilliant apparatus for caring about populations. What it cannot do is care about a trajectory — and the modern chronic disease burden is nothing but trajectories.

03 — Five phases, one diagnostic

The correspondence does not stop at attentiveness. Each of Tronto's phases names a failure mode the series documented without quite categorising — which suggests the five phases work as a systems diagnostic, not merely as ethics.

Caring about. The detection problem. Soft signals exist; nothing in the architecture is charged with noticing them. The Sussex Kidney Unit's screening question — "Would you be surprised if this patient transferred to HHD in six months?" — is attentiveness engineered into routine. It is a designed act of caring about, asked on a schedule, owned by a team.

Taking care of. The responsibility problem. In a fractionated system, the patient's trajectory between encounters belongs to no one. Each service is responsible for its episode; nobody is responsible for the drift. Fragmentation, read through Tronto, is not an organisational accident — it is a machine for diffusing responsibility until it vanishes. What Sussex actually did, beneath the service merger, was assign the trajectory to someone.

Care-giving. The competence problem. The clinician forced to meet a complex, multimorbid patient inside a ten-minute triage structure is being denied the conditions of competent care — and knows it. The series called the result moral injury. Tronto lets us say precisely what is being injured: the clinician's capacity to give care competently, withdrawn by the same system that will later audit her for outcomes.

Care-receiving. The responsiveness problem. The series described patients learning to calibrate — waiting until things are bad enough to cross a threshold the system recognises. That behaviour is the care-receiver's verdict, delivered in the only language the system accepts. Tronto insists care is not complete until the receiver's response has been attended to; a system that hears its patients only at thresholds has built deafness into its fourth phase, and the patients have noticed.

Caring with. The political problem — and the one the series reached last, in The Crooked Road, without naming it. Payment systems, regulatory metrics and change vehicles are not neutral plumbing. They are the polity's settlement about what care is owed, to whom, and at whose expense. A DRG schedule is a moral document. So is QOF. Caring Democracy's question — does the way we allocate caring responsibility deserve democratic assent? — is exactly the question a health system avoids when it lets reimbursement architecture stand in for deliberation.

04 — The attentive machine, re-tested

Part two of the series asked which kind of AI, doing what, in whose hands. Tronto supplies a sharper version of the test than accuracy or efficiency: for each phase of care, does the machine perform it, or does it create the conditions for humans to perform it better?

The distinction the series drew — an alert is a threshold crossed; a question is a probe — is the attentiveness test in miniature. An alert automates the old inattention at higher frequency: it notices only what it was told in advance to notice, which is to say it does not notice at all. A question — "this patient's trajectory has diverged from her own baseline; would you like to look?" — extends the reach of human caring about without pretending to replace it.

But Tronto's phases expose a danger the series understated. A system that detects soft transitions brilliantly and routes the detection into a structure where no one holds responsibility has automated caring about while leaving taking care of vacant — and that may be worse than blindness. An unnoticed need is a tragedy; a noticed and unmet one is a decision. Detection without an owner manufactures decisions of that kind at scale, with timestamps. The governance question for clinical AI is therefore not "how accurate is the model?" but "when the model notices, who has accepted responsibility for what happens next?" That question belongs in the procurement document, not the ethics appendix.

And the boundary the series drew around warm data — AI can process it, cannot create the conditions under which it emerges — is Tronto's boundary too. The conditions under which warm data emerges have a name: unhurried, continuous, attentive relationship. Care-giving. The machine can protect the time in which that happens. It cannot be the thing that happens.

05 — Who is doing the noticing now?

One more of Tronto's instruments, and the least comfortable one. Privileged irresponsibility names the position of those who benefit from care while neither performing it nor acknowledging that it is being performed (Tronto, 1993).

Apply it to the soft drift and the question becomes: in the years the system is not looking, who is? Someone tracks the lengthening pauses on the stairs. Someone notices the abandoned allotment, manages the medications, decides at two in the morning whether this breathlessness is the usual kind. In our own uncertainty work, one persona's entire monitoring arrangement runs on his wife's reports — and we wrote that sentence, first time through, without remarking on it. Her attentiveness is load-bearing infrastructure. It appears on no system map and in no budget line.

The event-based health economy free-rides on a vast, unpaid, mostly female attentiveness workforce — and its accounts balance only because that labour is kept off them. When the soft transition finally completes and presents as a hard event, the system bills for the rescue of a situation its own inattention allowed to mature, while the person who noticed everything is recorded, if at all, as "informal carer". This is privileged irresponsibility operating at the scale of a national institution. A practice that claims to map full assemblages has no licence to leave that off the map, and from here we will not: where informal attentiveness is load-bearing, our system maps will name it, weight it, and cost it.

06 — What this obliges

Three commitments follow — for our own practice first, and for any system willing to take the frame seriously.

Attentiveness is a designable property, and we will design for it. Sussex proves noticing can be engineered: a question asked on a schedule, owned by a named team, aimed at the drift rather than the event. In every pathway redesign we take on, "who is charged with noticing, and what are they charged with noticing?" is now a first-order design question — answered before any discussion of model accuracy or alert routeing, because a sensing layer without an owner is theatre.

The receiver's verdict is the test, and we will evaluate by it. Tronto's fourth phase gives pilots their honest endpoint: care is complete when the person receiving it says it is meeting the need — not when the process metrics turn green. Patient calibration is what accumulates when systems substitute their own measures for that verdict. Our pilots will state the receiver's verdict as the primary outcome, in those terms, in the protocol.

The payment system is a moral document, and we will read it as one. Every incentive structure is a statement about which needs count. Most were never written to be read that way, which is precisely why they should be. When we analyse a system, the reimbursement architecture gets read alongside the clinical pathway — as the settlement of caring responsibility it actually is — and its silences get named as choices.

The deepest thing Tronto offers the soft drift argument is permission to stop translating. We made the case for trajectory-based care in the system's own dialect — utilisation avoided, admissions averted, costs deferred — because that dialect gets meetings. But the translation concedes the premise that noticing must pay for itself in events. Tronto's claim is prior to all of that: a system that cannot notice slow-developing need is failing at care as such, whatever its readmission rates say.

The health system does not lack data about the soft drift. It lacks the decision that the drift is its to notice. That decision has a name in the literature, and a politics, and a thirty-year body of work behind it.

References