体力劳动产业 AI 全景图 · 一次身体的代劳

ASPINE · 主脊The spine

一次身体代劳的一生 · 一单劳务生命周期八节点The life of one body’s errand · the eight-node job lifecycle

体力服务劳动的最小单元不是「一个工种」,而是一次「用自己的身体替别人的身体做事」——打扫、照护、看守、跑腿、整理。每一单都要走一遍生命周期:需求→定义→派单→到岗→身体交付→监督→验收→复购。AI 沿这条链先接管「管活」(派单/监督/验收),再逐格逼近「做活」(身体交付)。每节点标注替代节奏:已被管住 / 挤压中 / 硬骨头 / 分化;渗透 % 为编者评估值。The smallest unit of physical-service labour is not «an occupation» but one instance of «using your body in place of someone else’s» — cleaning, caring, guarding, running errands, tidying. Every job walks the lifecycle: demand → definition → dispatch → arrival → body delivery → monitoring → inspection → re-booking. Along it, AI takes over «running the work» (dispatch/monitoring/inspection) first, then closes in node by node on «doing the work» (body delivery). Each node is tagged by tempo: MANAGED / SQUEEZED / THE BONE / SPLIT; penetration % are editorial estimates.

A+SUPPLY SIDE · 供给侧The supply side

从业者的一生:进城 → 培训 → 接单 → 损耗 → 退出A worker’s lifecycle: arrive → train → take jobs → wear out → exit

一单劳务有生命周期,干活的人也有——诚实层就藏在这条线上:谁把人送进系统、系统怎么消耗人、人最后带着什么离开。The job has a lifecycle; so does the worker — the honesty layer lives on this line: who feeds people into the system, how the system consumes them, and what they leave with.

01 · 进城ARRIVE

谁来干Who comes

骑手 81.6% 来自县域乡村(2022⚑);上海调研近八成学历高中及以下,60.19% 因薪资入行——「高收入预期」是入口的引力。81.6% of couriers came from counties and villages (2022 ⚑); a Shanghai survey found nearly 80% with high-school education or below, 60.19% drawn by pay — the income promise is the doorway’s gravity.

02 · 培训TRAIN

被打标签Getting tagged

「阿姨大学」、VR 课、80 课时岗前——证书化的本质,是把非标劳动转成算法可读取的「数字标签」:进入派单系统的通行证。«Auntie universities», VR courses, 80-hour onboarding — certification’s essence is converting non-standard labour into algorithm-readable «digital tags»: the pass into the dispatch system.

03 · 接单WORK

大进大出High churn

蓄水池的真相:高频骑手(年接单≥260 天)仅约 11%(约 82 万),约 48% 年接单不足 30 天;站点年流转率达七成——极好的过渡带,极差的长期职业。The reservoir’s truth: high-frequency couriers (≥260 days/yr) are only ~11% (~820K); ~48% take orders fewer than 30 days a year; station turnover hits 70% — an excellent transition zone, a terrible career.

04 · 损耗WEAR

身体先坏The body breaks first

2017 上半年上海平均每 2.5 天一名骑手伤亡;护理员腰背伤是离职首因;养老专业毕业生到岗流失率第一年 30%、第三年超 70%(单源)。In H1 2017 Shanghai averaged one courier casualty every 2.5 days; back injuries are carers’ top exit reason; care-major graduates leave at 30% in year one, 70%+ by year three (single source).

05 · 退出EXIT

带不走什么Leaving with little

多数人无养老/工伤积累而被动退出;宏观底色:全国养老金结余曾被预警 2035 年耗尽(2019 报告;2024 更新称延迟退休可推后 8–9 年)。Most exit passively with no pension or injury accrual; the macro backdrop: national pension reserves were once projected to run dry by 2035 (2019 report; a 2024 update says delayed retirement buys 8–9 more years).

中介盘剥换皮:线下介绍费/押金/培训费 → 平台佣金、加盟站管理费、设备租赁费、保险费(专送约 6 元/天、众包约 3 元/天,从配送费中扣)、流量费、排名费;传统月嫂中介费=一个月工资的 20–100%(一次性),平台把它改写成了 30.5% 的持续抽成。灵活就业超 2 亿,参加职工养老仅 7057 万——「自由」的底色,是三分之二的人没有职工养老。The middleman re-skinned: offline referral fees/deposits/training fees → platform commissions, franchise-station management fees, equipment rentals, insurance (≈¥6/day dedicated, ≈¥3/day crowdsourced, deducted from delivery pay), traffic fees, ranking fees. The traditional maternity-nanny agency fee was 20–100% of one month’s wage, one-off; the platform rewrote it as a 30.5% continuous take. 200M+ in flexible work, only 70.57M in employee pension — the ground colour of «freedom» is two-thirds without one.

A2BANDS · 五职业带Five occupational bands

同一生命周期,落到五种「身体动作物理场景」上One lifecycle, landing on five «physical scenes of the body»

工种的差别,只在「身体动作」落到哪种物理场景:能被结构化、可远程监控的地方 AI 跑得最快;必须落到非标、贴身、担责的地方 AI 卡在接口。同一台机械臂,在工厂是产线自动化(→factory),在这里是「喂饭机械臂仍是 demo」。Occupations differ only in which physical scene the «body’s action» lands on: where it can be structured and remotely monitored, AI runs fastest; where it must land on the non-standard, the intimate, the liable, AI jams at the interface. The same robot arm is line automation in a factory (→factory) and «a feeding arm that is still a demo» here.

BTHE MACHINES · 技术代理层The technical proxies

机器先吃哪一头:规模化早期,不是替代完成时Which end the machine eats first: early scale, not a finished replacement

替代身体动作的技术代理,几乎全部落在「户外结构化」那头——商用清洁、视觉安防、封闭配送已规模化早期,护理科技是「辅助不替代」,人形机器人入户仍在概念→实验室→早期试点。数字多为厂商/机构口径(⚑/D),配戒律看:演示口径 ≠ 你的开放非标场景。The technical proxies that replace the body’s actions land almost entirely on the «outdoor-structured» end — commercial cleaning, visual security and enclosed delivery are at early scale, care tech is «assist not replace», humanoids in homes are still concept → lab → early pilot. Most figures are vendor/institutional (⚑/D): a demo figure ≠ your open, non-standard scene.

TWO PARADOXES · 两个悖论Two paradoxes

避风港在漏气,照护堰塞湖在涨水The harbour leaks; the care reservoir rises

这是全图最沉重的两格:一个关于「往下退还有没有退路」,一个关于「最需要人的活为什么最不值钱」。The two heaviest cells: one about whether «stepping down» still has a floor, the other about why «the work that needs people most» is worth the least.

⚠ PARADOX Ⅰ — 避风港悖论the harbour paradox

「大不了去送外卖」,这块缓冲垫正从边缘漏气«Worst case, I’ll deliver food» — the cushion is leaking from the edges

当白领面临 AI 危机、试图向下降维寻求体力劳动的庇护,却发现底层同样在被自动化吞噬——只是先从户外结构化那头漏。骑手恰恰是最早被算法深度管理、也最早面对无人配送倒计时的身体劳动者:3 公里送餐时限从 2016 年的 1 小时,被算法压到 2018 年的 38 分钟;专送单价从约 8 元降到约 7 元;准时率低一点就扣款。无人车已在园区/校园/酒店先切走标准化好单(全国 6000+台、上百城市路权),留给人的是复杂路况的长尾。漏气是真的,清零是假象——被「最后一百米」的非标难题与成本曲线死死拖慢。而蓄水池还有一个被忽略的真相:它本来就是大进大出的(高频骑手仅约 11%,近半数年接单不足 30 天)——极好的过渡带,极差的长期职业;漏气最先漏掉的不是总量,是单价与自主权。When white-collar workers face an AI crisis and step down seeking the shelter of physical labour, they find the bottom is being devoured too — only leaking from the outdoor-structured end first. The courier is precisely the body-worker managed most deeply by algorithms and facing the driverless countdown first: the 3km delivery window was compressed by algorithms from 1 hour in 2016 to 38 minutes in 2018; per-order pay fell from ~¥8 to ~¥7; a dip in on-time rate means a deduction. Autonomous vehicles already skim the standardised good orders in campuses/hotels (6,000+ nationwide, road rights in 100+ cities), leaving people the long tail of messy conditions. The leak is real, the zeroing-out is an illusion — held back hard by the non-standard «last hundred metres» and the cost curve. And the reservoir has an overlooked truth: it was always high-churn (high-frequency couriers only ~11%, nearly half take orders fewer than 30 days a year) — an excellent transition zone, a terrible career; what leaks first is not volume but price and autonomy.

3 公里送餐时限3km delivery window 2016 · 60min→ 2018 · 38min→ 单价 ¥8 → ¥7 · 准时率扣款pay ¥8 → ¥7 · on-time deductions

⚠ PARADOX Ⅱ — 照护悖论the care paradox

最需要人的工作,社会支付意愿最低The work that needs people most, society is least willing to pay for

老龄化制造了未来二十年最确定的劳动缺口(日本 2040 缺约 57 万,中国失能老人约 3500 万、缺口数百万至千万级),而「把老人从床上安全扶起」恰是具身智能的珠峰。偏偏中国养老护理员报酬压在服务业最底部——月薪约 2272 元接近城市最低工资、劳动合同签订率仅略高于 1%、超 42% 无任何社保。于是年轻人不愿入行、护理员老龄化、养老专业毕业生到岗流失率第一年 30%/第三年超 70%。价格与价值严重倒挂,这里不会漏气,反而越裂越大。行业内部的定价锚更刺眼:养老护理员月薪只有月嫂的约 37%、时薪只有保洁的约 42%;而缺口的 85–95% 在居家社区,不在机构。政策在补课:「十五五」末养老护理员持证率目标 80%、2026 年设「养老服务师」——证书化是把照护从「家务的延伸」改写成「可定价职业」的第一步。Ageing creates the most certain labour gap of the next two decades (Japan short ~570K by 2040; China ~35M disabled elderly, a gap of millions to tens of millions), and «lifting an elder safely off a bed» is the Everest of embodied AI. Yet Chinese elder-care wages sit at the very bottom of the service economy — ~¥2,272/month near the urban minimum, labour-contract signing just over 1%, over 42% with no social insurance at all. So the young avoid the field, carers age, and care-major graduates leave at 30% in year one, over 70% by year three. Price and value are deeply inverted; this end does not leak — it only widens. The internal anchors are harsher still: an elder-care worker earns ~37% of a maternity nanny’s monthly wage and ~42% of a cleaner’s hourly rate; and 85–95% of the gap sits in homes and communities, not institutions. Policy is catching up: an 80% certification target for carers by the plan’s end, an «elder-service worker» qualification in 2026 — certification is the first step in rewriting care from «an extension of housework» into «a priceable profession».

陪聊音箱·付费意愿低chat speaker · low willingness

监测辅具·减负monitoring aids · relief

移位机·辅助transfer lifts · assist

▲ 安全扶起失能老人·珠峰▲ lifting the disabled elder · the summit

B+GLOBAL · 四种答案Four answers

同一道缺口,四种补法——每一种都有自己的账单One gap, four fixes — each with its own bill

老龄化与体力劳动缺口是全球共题:WHO 已把 2030 年全球卫生与照护人力缺口上调至 1110 万。四大经济体交出了四份互不相同的答卷——把它们并排放,才能看清「中国方案」的真实坐标。Ageing and the physical-labour gap are a global question: the WHO has raised its 2030 health-and-care workforce shortfall to 11.1 million. Four economies handed in four different answers — only side by side do you see where China’s approach actually sits.

中国 · 算法 + 无人配送China · algorithms + driverlessA/B

用算法解吞吐、用无人配送压成本;监管开始把算法拉回「可申诉、可追溯」(公示栏/取消超时罚款/12h 强制下线),社保修补进行时(职业伤害保障 2026 全国推开)。答卷的强项是速度,账单是保障真空。Solve throughput with algorithms, compress cost with driverless delivery; regulators are pulling the algorithm back to «appealable and traceable» (disclosure boards, scrapped overtime fines, 12h forced-offline), welfare patching underway (nationwide injury protection in 2026). The strength is speed; the bill is the protection void.

日本 · 补贴 + 机器人Japan · subsidies + robotsB

政策最系统:经产省九大护理机器人重点领域、移位机纳入介护保险租赁,官方估产业 2035 年约 4000 亿日元、产量超其后四国总和——但 2019 年 9000 多家养老机构仅约 10% 引入过任何护理机器人。三十年实验的结论:机器人给腰减负,补不了 2040 年缺的 57 万人。The most systematic policy: nine priority care-robot fields, transfer lifts inside long-term-care insurance rental, an official ~¥400B industry estimate by 2035, output exceeding the next four countries combined — yet by 2019 only ~10% of 9,000+ elder-care facilities had ever adopted any care robot. Thirty years of experiments conclude: robots relieve backs; they cannot supply the 570K missing people of 2040.

欧美 · 移民 + 平台The West · migration + platformsB/C

靠移民补照护供给(德国 2035 年前需新增 13 万+专业护理、欧洲 2050 无移民或缺 4400 万劳动力);平台侧高抽成(外卖对商家佣金 15–30%、网约司机扣费后平均时薪约 11.77 美元)。美国直接照护工 2024 中位时薪 17.36 美元、36% 处于或接近贫困——而居家健康助理偏偏是未来十年新增岗位最多的职业(2024–34 预计 +17%)。最缺人的职业,时薪在 50 个州都垫底。Care supply patched by migration (Germany needs 130K+ more professionals by 2035; Europe may lack 44M workers by 2050 without it); platform-side high takes (15–30% merchant commissions, ride-hail drivers averaging ~$11.77/hr after fees). US direct-care workers earned a median $17.36/hr in 2024, 36% in or near poverty — while home-health aides are the fastest-adding occupation of the decade (+17% projected 2024–34). The most needed job pays bottom-tier in all 50 states.

东南亚 · 输出 + 债务Southeast Asia · export + debtC

全球照护/家政劳动力的净输出池:中介与培训费通过 7–9 个月工资扣除偿还、部分以房产车辆抵押——「出国前先负债」;菲律宾法律禁收中介费但仍收「相关费用」,印尼靠 2017 年移民工人保护法监管(部分数据系较早年份,见口径)。别人的照护缺口,是她们的债务起点。The net export pool of global care and domestic labour: agency and training fees repaid via 7–9 months of wages, some collateralised on family homes — «in debt before departure»; the Philippines bans agency fees yet «related charges» persist, Indonesia regulates via its 2017 migrant-worker law (some figures are from earlier years, see notes). Someone else’s care gap is where their debt begins.

CHARD BONES · 硬骨头Hard bones

AI/机器人攻不进的七块:结构性地难,非算法可解Seven things AI and robots cannot breach: structurally hard, not algorithmically solvable

利润与地位正向这七堵墙内迁移:能做非标入户、能承担情感与责任、有专项证书与直客关系的位置,最后失守。每一块都不是「机器还不够强」,而是「问题的结构不给机器留位置」。Profit and status are migrating behind these seven walls: the positions that handle non-standard in-home work, carry emotion and liability, and hold specialist certificates and direct clients fall last. Each is hard not because «machines aren’t strong enough yet» but because the problem’s structure leaves machines no seat.

DSTANCE MAP · 立场地图Stance map

你站在哪:按角色 × 三档替代节奏过滤Where you stand: filter by role × three tempos of replacement

选「我是谁」高亮与你相关的卡;切三档信号灯看「两头挤压」——已被管住(算法) / 挤压中(机器) / 硬骨头(人的堡垒)。开关本身就是「两头挤压、中间收窄」论点的交互证明。Pick «who I am» to highlight your cards; toggle the three signal lights to see the «squeeze from both ends» — MANAGED (algorithm) / SQUEEZED (machine) / THE BONE (human fortress). The toggles are the interactive proof of «squeezed from both ends, narrowing in the middle».

我是谁WHO I AM 打工者 · 骑手/保洁/护工/保安Worker · courier/cleaner/carer/guard 平台 · 中介Platform · agency 雇主 · 机构Employer · institution 机器人 · 物理 AIRobot maker · physical AI 家庭消费者Household 监管 · 社保Policy · welfare

哪一档WHICH TEMPO 已被管住 · 算法Managed · algorithm 挤压中 · 机器Squeezed · machine 硬骨头 · 人的堡垒The bone · human fortress

DEBATES · 争议对垒Live debates

四场没打完的仗Four unfinished fights

四份底稿对下面四题立场不一——这本身就是重要信息。本页各方并置,再给编者合议,不裁掉任何一方。The four sources disagree on the questions below — that disagreement is itself information. Each side is shown, then an editorial synthesis; no side is cut.

TIMELINE · 时间线Timeline

2015–2026:从算法派单到会跑的机器2015–2026: from algorithmic dispatch to machines that move

底稿对 2015–2019 极薄(骑手/家政/养老/保安行业基线为主,标 ※ 为编者补注),2020 疫情催化后渐厚,2023–2026 最厚(无人配送规模化+算法治理+护理科技+人形机器人+社保政策)——薄区如实留稀。Sources are thin on 2015–2019 (industry baselines only; ※ marks editorial notes), thicken after the 2020 pandemic catalyst, and are thickest in 2023–2026 (delivery at scale, algorithm governance, care tech, humanoids, welfare policy) — sparse zones left honestly sparse.

FIELD GUIDE · 产品指南Field guide

四种人的「现在能用 / 别碰 / 戒律」«Use now / avoid / discipline» for four roles

厂商效率数字一律 D 级(演示口径);所有清单是行业结构信息,不构成就业或投资建议。Vendor efficiency figures are D-grade (demo scope); these lists are industry-structure information, not employment or investment advice.

FIVE LAWS · 五条规律Five laws

把整张图压缩成五句判断The whole map, compressed into five judgments

管活 > 做活:算法先于机器人替代了自主权Running > doing: algorithms replaced autonomy before robots did

最先被替代的不是身体动作,是围绕这单劳务的「找人、派单、监控、验收、结算、复购」。机器还在平整地面学步,算法已经把骑手变成「系统的执行末端」。看一个岗位危不危险,先看它的「管活」层被数据化到什么程度。What gets replaced first is not the body’s action but the «finding, dispatching, monitoring, inspecting, settling, re-booking» around the job. The machine still learns to walk on flat floors while the algorithm has made the courier «the system’s executing endpoint». To gauge a job’s risk, first see how datafied its «running the work» layer is.

两轴分化:结构化 + 低情感 = 先被吃Two axes: structured + low-emotion = eaten first

替代沿「环境结构化程度」和「情感劳动含量」两轴走:平整、可远程监控、低情感的活(商用清洁/门岗盯屏/封闭配送)先被机器吃;非标、贴身、担责、高情感的活(入户照护/月嫂/临终陪护)最后失守。手,比脑更难自动化。Replacement runs along two axes — environmental structuredness and emotional-labour content: flat, remotely-monitorable, low-emotion work (commercial cleaning, screen-watching, enclosed delivery) is eaten by machines first; non-standard, intimate, liable, high-emotion work (in-home care, maternity nannies, end-of-life companionship) falls last. The hand is harder to automate than the mind.

缓冲垫从边缘漏,不是中心塌The cushion leaks from the edges, not the centre

「大不了送外卖」的失业避风港正在漏气,但先从户外结构化那头漏(配送/清洁/安防),被「最后一百米」的非标与成本曲线拖慢——漏气是真的,清零是假象。10 年内替代比例仍有限而非清零,别把「早期」误读成「过去时」。蓄水池的底色本就是大进大出(高频骑手仅约 11%)——它是好的过渡带,坏的职业。The «worst case, deliver food» harbour is leaking, but from the outdoor-structured end first (delivery/cleaning/security), slowed by the non-standard «last hundred metres» and the cost curve — the leak is real, the zeroing-out is an illusion. Replacement stays limited, not zero, within a decade; do not misread «early» as «past tense». And the reservoir was always high-churn (high-frequency couriers ~11%) — a good transition zone, a bad career.

照护是堰塞湖:缺口越裂越大,报酬却在底部Care is a rising reservoir: the gap widens while pay sits at the floor

照护是未来二十年最确定的劳动缺口,也是定价与价值最倒挂的一格——最需要人的活报酬最低。机器人补得了体力缺口与监测盲区,补不了商业逻辑的断裂:能解失能刚需的高端机器人太贵、付费意愿低的陪伴音箱又没用。照护是制度题、文化题,不是纯 AI 题。Care is the most certain labour gap of the next two decades and the cell where price and value are most inverted — the work that needs people most pays least. Robots can fill the physical gap and monitoring blind spots but not the broken business logic: the high-end robot that meets real disability needs is too expensive, the affordable companion speaker useless. Care is a question of institutions and culture, not pure AI.

重新分层:证书 + 直客 + 情感 + 担责 > 能吃苦Re-stratification: certificate + direct clients + emotion + liability > endurance

AI 的净效应是把劳动者推向两端、中间收窄。未来的赢家不是「最能吃苦的人」,而是「有专项技能、证书、口碑、直客、会用工具的人」。月嫂持证平均月薪 16246 元 vs 通用保洁被商用机器人首先替代——把『劳务』做成『可定价、可分级的产品』,向机器管不住、也换不掉的那端走。AI’s net effect pushes workers to two ends and narrows the middle. The winners are not «those who can endure most» but «those with specialist skills, certificates, reputation, direct clients and tool fluency». A certified maternity nanny averages ¥16,246/month while generic cleaning is replaced first by commercial robots — turn «labour» into «a priceable, tiered product», and move toward the end machines can neither run nor replace.