拧开水龙头，你判断不了这水能不能喝。
它无色无味，你尝不出里面有没有铅，也做不了检测。你能做的只有查一份水质公告，而公告出自一个你没见过的机构，用的是一套你没法复核的方法。
于是有件事被悄悄换掉了：你想知道的是「这水能不能喝」，你实际能做的判断是「这个出公告的机构值不值得信」。这道替换一旦发生就退不回去，而它发生在你每天碰到的几乎所有危险上：食品、药、房子的结构、孩子学校的空气。
1986 年春天，德国社会学家乌尔里希·贝克（Ulrich Beck）快写完一本书。四月底切尔诺贝利炸了，他在德文版序里补了一段，然后书出版，成了那十年德语社会学最热的一本，叫《风险社会》。
他的起手式是个对照。工业社会争的是好东西怎么分：工资、房子、上学的机会。他说有一类新的争夺在长出来，争的是坏东西怎么分：谁住在下风口、谁家的井在化工厂下游、谁吃到了出问题的那一批。
这些坏东西跟从前的灾祸不是一回事，有三个特点。一是制造出来的——不是地震台风，是某个技术决策的副产品，原则上有人做过那个决定。二是感官测不到——你看不见辐射也闻不出添加物，「我有没有被暴露」本身要靠仪器和专家才能成立。三是不受时空约束——过境、跨代，你没法搬家躲开。
三条合起来推出一句要紧的话：在这种风险面前，「我受了害吗」不是你自己能知道的事实，是一个要用知识和文件去确立的资格。
这话有个很具体的样子。人类学家阿德里亚娜·佩特里纳（Adriana Petryna）在 1990 年代的乌克兰做了几年田野，写成《暴露的生命》（2002）。切尔诺贝利之后，新独立的乌克兰把大量财政转成了辐射相关的补助与残疾认定。她看到普通人必须把自己变成半个专家——学会剂量单位、学会哪些诊断挂得上「与辐射相关」——才能把身上的损害兑成一个有人承认的身份，她叫它生物学公民（biological citizenship）。这是民族志个案，强在细，弱在推不到别处。
贝克还有句被引到滥的话：「贫穷是等级的，烟雾是民主的。」他指新风险有回旋镖效应，造它的人自己也躲不开。这句争议很大：环境不平等这一支几十年来相当一致地发现，暴露其实高度分层，填埋场和化工园区更常落在穷人和少数族群聚居的地方。一种读法说他夸大了均等化；另一种说他讲的不是暴露均等，而是「没有一处绝对安全」这件事是新的。还得说清楚：《风险社会》是理论框架，不是实证发现。
EN In the spring of 1986 the German sociologist Ulrich Beck was close to finishing a book. Chernobyl exploded at the end of April, he added a passage about it to the preface of the German edition, and the book — Risk Society — became the most talked-about work of German-language sociology of the decade. His opening move is a contrast. Industrial society fights over how good things are shared out: wages, housing, places at school. A new kind of contest is growing up alongside it, he says, over how bad things are shared out: who lives downwind, whose well sits downstream of the chemical plant, who got the batch that was contaminated. These bads are not the old calamities, and they have three features. They are manufactured — not earthquakes and typhoons but side effects of some technical decision, so in principle someone made that decision. They are below the threshold of the senses — you cannot see radiation or smell an additive, so even the fact of your own exposure has to be established with instruments and experts. And they are unbounded in space and time — they cross borders and generations, and you cannot move away from them. Together those three yield one consequential line: in the face of risk like this, "have I been harmed" is not a fact you can know by yourself. It is an entitlement that has to be established with knowledge and paperwork. That line has a very concrete shape. The anthropologist Adriana Petryna spent several years doing fieldwork in 1990s Ukraine and wrote Life Exposed (2002). After Chernobyl, newly independent Ukraine turned a large share of its budget into radiation-related benefits and disability designations. What she saw was ordinary people having to make themselves into half-experts — learning dose units, learning which diagnoses can be attached to "radiation-related" — in order to convert the damage in their bodies into a status somebody recognizes; she calls it biological citizenship. This is a single ethnographic case, strong in detail and weak as a basis for generalization. Beck also has a line that has been quoted to death: "poverty is hierarchic, smog is democratic." He means that the new risks have a boomerang effect, so that those who produce them cannot finally stay clear of them. The line is heavily contested: decades of work on environmental inequality find fairly consistently that exposure is in fact steeply stratified, with landfills and chemical parks landing more often where poor people and minority communities live. One reading says he overstated the equalization, and that bads are still distributed along the old hierarchy; another says his claim was never about equal exposure but about the novelty of there being no absolutely safe position. And it should be said plainly: Risk Society is a theoretical framework, not an empirical finding.
1986 年 4 月底，切尔诺贝利的烟云飘到英国，在英格兰西北的坎布里亚山区下了雨，铯落在草上。6 月 20 日，农业部对坎布里亚和威尔士的羊下了移动与屠宰禁令，涉及约五千个农场、两百多万只羊，通知里写的是三周。三周之后禁令延长了，然后一延再延。最后一批限制取消是 2012 年，二十六年。
为什么能错这么远？科学家的预测基于一个模型：铯落进土里会被黏土矿物牢牢吸住，进不了草，几周就锁死。低地那种偏碱性的黏土土壤确实如此。坎布里亚的山地不是——酸性泥炭土，铯在里面保持着化学上的可移动，被草根吸回去，再进羊。模型用错了土。
社会学家布赖恩·温（Brian Wynne）1980 年代末在那里做了长期访谈，1992 年写成《被误解的误解》。他的发现不是「农民不懂科学」：农民对铯的危害量级几乎没有异议，也照着建议做。他们不接受的是另外两样——科学家为了做实验要求把羊圈在小块试验地里，农民说山地羊群根本不这么放，圈起来的羊和山上的羊不是一回事，这话被当成外行话；以及科学家一次次给出确定的时间表，又一次次改口，却始终不承认模型可能是错的。
温的结论少见地硬：普通人对科学的信任，判断的对象不是知识内容，是提供知识那套机构的举止——它承不承认自己可能错、认不认你的处境里有它不知道的事、你跟它是什么依赖关系。坎布里亚还多一层：山那边就是当地大雇主塞拉菲尔德核设施，不少农户家里有人在那上班。这是定性案例研究，影响极大，也被批评为把公众的理性说得太乐观。
这件事的骨架，安东尼·吉登斯（Anthony Giddens）在《现代性的后果》（1990）里给了名字。他说现代社会靠两类机制把关系从「当地」拔出来，叫脱域（disembedding）：一类是货币这样的符号标志，你收下一张纸不必认识付钱的人；另一类是专家系统（expert systems），把技术知识组织起来，替你担保你没法自己检查的事。你坐飞机不懂空气动力学，这不是你的知识缺口，是脱域正常工作的样子。
但脱域不免费。你信的不是人，是一套看不见的系统——吉登斯叫它「无面目的承诺」。系统只在少数地方露脸，他叫接入点（access points）：柜台、诊室、公告会。你对整套系统的估计几乎全部来自这几个点上的举止，权重大得不成比例。坎布里亚那些会议室就是接入点，而在那里发生的是：确定的语气加上反复改口。
EN At the end of April 1986 the Chernobyl plume reached Britain and rained out over the Cumbrian fells in north-west England; caesium settled on the grass. On 20 June the ministry of agriculture banned the movement and slaughter of sheep in Cumbria and Wales, covering some five thousand farms and more than two million animals. The notice said three weeks. After three weeks the ban was extended, and then extended again and again. The last restrictions came off in 2012 — twenty-six years. How could the estimate be that far out? The scientists' forecast rested on a model in which caesium falling on soil is bound tightly by clay minerals, never reaches the grass, and is locked away within weeks. In the alkaline clay soils of the lowlands that is what happens. The Cumbrian uplands are not that: they are acid peat, where the caesium stayed chemically mobile, was taken up again through the grass roots, and went back into the sheep. The model had the wrong soil. The sociologist Brian Wynne ran long interviews there in the late 1980s and in 1992 published "Misunderstood misunderstanding". What he found was not that the farmers failed to understand science: they barely disagreed about the magnitude of the hazard from caesium, and they followed the advice. What they refused was two other things. To run an experiment the scientists wanted sheep confined to small fixed plots; the farmers said that is not remotely how a hill flock is grazed, and that penned sheep and hill sheep are not the same animal in the relevant sense — and that was treated as amateur talk. And the scientists issued confident timetables, revised them, issued them again, and never conceded that their model might be wrong. Wynne's conclusion is unusually hard for this field: what lay people judge when they decide whether to trust science is not the content of the knowledge but the conduct of the institution supplying it — whether it concedes that it might be wrong, whether it grants that your situation holds things it does not know, and what your relation of dependence on it is. Cumbria had a further layer: over the hill sat Sellafield, the big local employer, with someone from many of those households on its payroll. This is a qualitative case study, hugely influential, and also criticized for painting lay rationality too optimistically. The skeleton of the episode was named by Anthony Giddens in The Consequences of Modernity (1990). Modern societies, he says, lift relations out of their local settings by two kinds of mechanism, which he calls disembedding: one is symbolic tokens like money, where you accept a piece of paper without needing to know who paid; the other is expert systems, technical knowledge organized so that it vouches, on your behalf, for what you cannot check yourself. You fly without knowing aerodynamics. That is not a gap in your education; that is disembedding working as designed. But disembedding is not free. What you trust is not a person but a system you cannot see — Giddens calls it a faceless commitment. The system shows a face only in a few places, which he calls access points: the counter, the consulting room, the briefing. Your estimate of the whole system comes almost entirely from conduct at those few points, which gives them wildly disproportionate weight. Those meeting rooms in Cumbria were access points, and what happened in them was a confident tone plus repeated revisions.
1986 年 1 月，挑战者号升空七十几秒后解体。流行的说法是有人明知危险还批了发射。社会学家黛安·沃恩（Diane Vaughan）花九年读完调查留下的几万页材料、补做访谈，1996 年出了《挑战者号发射决策》，答案更难受。
出事的部件是固体助推器分段之间的 O 形密封圈，它在之前的飞行里就出现过侵蚀。每一次，工程师都按程序写报告、开评审会、把这回的侵蚀量跟以往比一比，判定「在已有经验范围内」，放行。下一次侵蚀多一点，可参照的基线已被上一次挪过去，于是又在范围内。沃恩把这叫偏离的正常化（normalization of deviance）：一个本来算异常的信号，通过一次次合规、认真、留了文件的判断，被逐步收进「正常」里。
要紧的是她的否定：这不是一群莽人，也没有人隐瞒，每一步都能在会议记录里找到理由。危险不是被谁决定的，是被累加出来的。贝克给这种结构起过名字：组织化的不负责（organized irresponsibility）——制度一边生产危险，一边用归责与举证的规则把危险变成没人可归责的事。提醒一句：这是单案例的档案研究，而「偏离的正常化」后来被用得非常松，套得越广解释力越弱。
那信任还能靠什么维系？有条经得起检验的线索指向一个不太像答案的地方：不在内容上，在举止上。常见的担心是「说不准就会掉信任，所以别给区间」。2020 年范德布勒斯、斯皮格哈尔特等人做了五项实验、合计约五千八百人，其中一项直接嵌在 BBC 新闻网站上：给出数值区间确实让人觉得这个数字更不确定（本来就该如此），但对来源的信任只掉一点点；掉得相对多的反而是用含糊的词说不确定那种。效应量都不大，方向稳定。
放回坎布里亚：农民抱怨的既不是禁令太严，也不是铯的毒性数据，是被当成不懂事的人，和一个不肯说「我们可能弄错了」的语气。信任是对机构举止的一个估计，而这个估计有可观察的后果：下一次公告出来，有多少人照着做。
EN In January 1986 the space shuttle Challenger broke apart some seventy seconds after launch. The popular account is that someone approved a launch they knew to be dangerous. The sociologist Diane Vaughan spent nine years working through the tens of thousands of pages the investigation left behind, added interviews of her own, and in 1996 published The Challenger Launch Decision with a more uncomfortable answer. The part that failed was the O-ring seal between segments of the solid rocket booster, and erosion of it had shown up on earlier flights. Each time, engineers wrote the report, held the review, compared this instance of erosion against previous ones, judged it "within the existing experience base", and cleared the flight. Next time the erosion was a little worse — but the baseline it was compared against had already been moved by the time before, so it was within the base again. Vaughan calls this the normalization of deviance: a signal that counted as an anomaly is gradually absorbed into the normal through a sequence of compliant, conscientious, fully documented judgements. What matters is her negative claim. These were not reckless people, nobody concealed anything, and every step has a reason in the minutes. The danger was not decided by anyone; it accumulated. Beck had a name for a structure like that: organized irresponsibility — institutions producing hazards while the rules of attribution and proof turn those hazards into something nobody can be held to account for. One caution: this is a single-case historical and archival study, and "normalization of deviance" has since been used very loosely, losing explanatory force as it widens. So what can trust rest on? One line of evidence that holds up points somewhere that does not look much like an answer: not the content, the conduct. The standard worry runs, hedge and you lose trust, so do not give ranges. In 2020 van der Bles, Spiegelhalter and colleagues ran five experiments with roughly 5,800 participants in total, one of them embedded directly in the BBC News website. Giving a numerical range did make people see the number as more uncertain — as it should — but trust in the source fell only slightly; what cost relatively more was expressing the uncertainty in vague words. The effects are all small and the direction is stable. Back to Cumbria. What the farmers complained about was neither the severity of the ban nor the toxicology of caesium. It was being treated as people who knew nothing, and a tone that would not say "we may have got this wrong". Trust is an estimate of an institution's conduct, and that estimate has an observable consequence: how many people follow the next announcement.