看得见的那一个：为什么责任不随人数增长
A named person outweighs a hundred statistical ones; the problem is neither indifference nor an absence of responsibility, but that obligation is allocated by identifiability, distance and the moment accounts close
One Person Against a Hundred
A child needs money for an operation. There is a name, a photograph, a diagnosis and a deadline, and the sum is raised within days. In the same period, an infrastructure change that would reduce the same class of risk for far more people stalls, because the case is long, the benefit cannot be attached to anyone in particular, and no one can say when the job will be finished. The difference between the two is not the size of the consequence. It is that the first is a story with an ending.
The question worth asking is why the strength of an obligation fails to grow with the number of people affected, and instead decays with how identifiable they are. Two ready explanations should be set aside. It is not a shortage of sympathy: responses to a named individual are often immediate and generous. Nor is it an absence of responsibility, in the sense of a task spread across so many hands that nobody has to answer for it. Here the decision maker knows the extent of their authority and knows the consequences, and still directs resources toward the few who can be seen. The trouble lies in weighting. Obligation is not computed from the scale of outcomes but from identifiability, temporal distance and the length of the causal chain.
Psychology calls the first of these the identifiable victim effect: willingness to give is higher for a single concretely described person than for a group presented as a statistic. Work associated with Paul Slovic describes a stranger companion phenomenon under the name psychic numbing, in which felt concern and willingness to act fall as the number in need rises; early findings have since been revisited in replication and extension studies. These are paraphrases, not quotations. Together they point to a plainer claim: obligation needs an object, and an object must be identifiable before it enters anyone's calculation. Numbers supply scale; they do not supply an object.
A Story Can End; a Statistic Cannot
A named individual carries three things a statistic lacks. It has a name, and so can be referred to. It has a situation that has already happened, and so requires no imaginative completion. And it has a definite ending: the money arrives, the operation takes place, the matter closes. The psychological precondition for acting is precisely that the matter can be brought to a close. A completable action has a clear beginning, a bounded cost and a result that can be retold.
A statistic offers probabilities, distributions and intervals instead. It describes events that have not yet happened, and it can never be finished: after one cohort's risk is reduced there is another, and saving a number of people shows up only as a small movement in a rate. Help therefore cannot be experienced as a completion, only as a marginal improvement. A marginal improvement produces no ending and no story worth repeating; without a story there is no picture capable of mobilising anyone.
The easily missed point is that identifiability is not a quantity of information but a shape of information. Listing the ages, occupations and addresses of a hundred people does not multiply concern by a hundred. As the file thickens, processing load rises and felt concern tends to thin. More information producing less weight means the weighting function simply does not read number as an input. The literature on noise concerns a different failing, the random scatter of the same judgment across occasions; that is a problem of reliability, whereas this is a problem of allocation. Both degrade judgment, and they need different repairs.
The first mechanism, then, is that visibility is not about how much is known but about whether a matter can be closed. Obligation is triggered by a story that is unfinished yet finishable, and a statistic satisfies neither condition.
Distance Discounts Obligation
The second mechanism is distance, in three forms. Temporal distance is the hardest: people in the future have not been born or have not entered any register, hold no position, cannot complain, and appear in no feedback loop for the current period. Spatial distance comes next, since those far away enter neither the daily field of view nor the local chain of accountability. The length of the causal chain is the most obscure. When a decision reaches a person only after a dozen steps, each step can describe itself as mere execution, and responsibility thins along the chain until it cannot be attributed to anyone.
All three distances converge on an asymmetry of traces. Acting leaves budgets, contracts, signatures and minutes, all auditable, questionable and attributable. Not acting leaves almost nothing: no document, no signature, no nameable moment of failure. For anyone occupying an auditable position, a bias toward inaction is therefore not a moral defect but a reasonable response to the recording system. This differs from the problem of defaults, where an omitted choice is read by the system as consent. Here the point is that inaction generates no record to be reviewed at all: in one case a meaning is written onto silence, in the other no trace is produced.
The outcome is more serious than partiality. Distant people are not rejected; they never enter the candidate set. A rejection at least has an opposing party, a reason and an argument that can be answered. A loss that never became a candidate has no name, and so no one can appeal it. Obligation decays with distance, and at the limit it does not become small. It becomes zero.
Ledgers Close in the Present
Moving from persons to institutions makes the structure clearer. Budget years, terms of office, project cycles, incident reports and performance indicators all demand results visible within the period. None was designed with malice; each was built so that an organisation could be counted and called to account. Their shared effect, however, is that the ledger has a closing date, and people and losses beyond that date have no line item. With no line item there is no pending task, and no absence that needs explaining.
A boundary has to be drawn here with the argument about removing prices. That discussion concerned what happens when a price tag is taken away: queues, lotteries, eligibility and connections take its place, and the cost of rationing remains, merely booked to people least able to object. The cost is transferred. What is at issue here is a cost that was never booked at all: not displacement but absence. Both involve people who cannot complain, yet one is about a substitute carrier for allocation and the other about the temporal boundary of an accounting scheme.
Institutions have in fact built tools to resist that absence. Cost-benefit analysis of public projects discounts future benefits and prices risk reduction through the value of a statistical life. Thomas Schelling introduced that concept in his 1968 essay "The Life You Save May Be Your Own" to address a famous asymmetry: a named six-year-old girl needing surgery money would fill a post office with donations, whereas repealing a sales tax that lets medical equipment decay and causes avoidable deaths draws almost no letters. What the two tools share is the conversion of unidentifiable people into units that can be entered in a ledger, forcing the distant to appear. The delays and boundaries described in systems thinking are not abstractions at this point; they are an actual blank column in the accounts.
The Countercase: Commensuration Is How the Distant Become Visible
The argument so far slides easily toward a cheap conclusion, that concern ought to be spread more evenly. Three strong objections come first.
First, the preference for identifiable people has cognitive justification. Nearby obligations are executable, verifiable and completable; concern for abstract large numbers tends to remain at the level of declaration and can even produce moral licensing, where expressing concern lowers the obligation actually discharged. Requiring people to allocate care strictly by headcount is both unrealistic and likely to weaken the obligations that can genuinely be honoured. Work on the moral circle suggests that concern expands outward from concrete relationships rather than contracting inward from an abstract principle; condemning that structure outright asks people to operate with a cognitive apparatus they do not have.
Second, discount rates and the value of a statistical life were built precisely to counteract that bias. Abolishing them and retreating to intuitive care would not produce greater fairness; it would hand still more to those nearby, named and vocal. Beneath the cold surface of abstract conversion lies an effort to bring strangers inside the calculation. The claim that large-scale cooperation among strangers depends on shared fictions can be read quite literally here: a common unit of account is one such fiction, unwarm but sufficient to let people who will never meet be counted together.
Third, perfectly impartial concern is psychologically unsustainable and dilutes concrete commitments. A system equally responsible to everyone often bears no enforceable responsibility to anyone in particular. At an emergency scene, attending first to the person in front is not a moral error but a necessary division of labour.
The objection carries its own limit, and the limit should not be skipped. Placing life, health, time and liberty on a single ruler means the dimension that is dropped appears neither in the result nor in a footnote. This is distinct from the problem of measures becoming targets, where an indicator is gamed once it is rewarded and is optimised into distortion. Here a dimension is erased at the moment of conversion, whether or not anyone cheats. The real question is therefore not whether to convert, but who sets the conversion coefficients, whether they can be contested in public, and whether what was discarded has been written down.
Make Inaction Auditable
The workable conclusion is neither greater kindness nor the abandonment of concrete intuition. It is to write invisible people into the accounting scheme, from which three tests follow.
First, decision papers should present the named case and the statistical scale together, and state explicitly what weight each receives and on what grounds. The aim is not to neutralise the force of an individual case but to move the weighting from implicit to explicit, so that why this one case outranks that cohort becomes a question with an answer.
Second, distant and future losses need a line item and a named owner. Without a line item there is no eligibility for consideration, and specifying who must maintain that column does more than repeated appeals to take the future seriously.
Third, discount rates, values of a statistical life and assessment conventions must be public, reproducible and open to challenge, and must state which dimension they discard. Publishing a coefficient is not a bid for greater precision; it is a way of returning the erased dimension to discussion.
The three tests have limits of their own, worth recording alongside them. Converting lives into money and then naming what was dropped does not compensate for what was dropped. Record-keeping adds paperwork and can invite performative concern, in which having registered a matter substitutes for having dealt with it. Frequent revisions to conventions break long-run comparison and leave room for manipulation. These are not a scorecard but three reminders.
That returns to the opening question. Obligation fails to scale with numbers not because numbers are too abstract, but because the ledger has no column for them. Research on poverty keeps showing that a vague category becomes discussable only when it is broken into concrete, measurable entries; the comparative history of welfare states keeps showing that who is counted inside the community is itself an institutional choice. Add the column and numbers begin to enter the calculation. The test of a system's reliability is not whom it has saved but whether it can state what it did not do for whom, and why. Making inaction leave an auditable trace is the only improvement that does not depend on goodwill.
