「AI 吞噬世界」 · AI IS EATING THE WORLD— №48 · INTIMACY / 亲密产业
一单位被出售的亲密Intimacy, by the Unit
AI 把「被回应」做成了工业品 —— 亲密商品化光谱 AI 全景图 · 2026AI turned «being answered» into an industrial product — the commodified-intimacy spectrum · 2026
这个产业卖的从来不是爱情,是「一单位被出售的亲密」——被看见、被回应、被偏爱、被幻想。传统上每一单位都由真人实时生产:聊手按条计费(8 元只够 14 条),主播下播后还要在私域「写作业」,订阅创作者的私信外包给一人管七个收件箱的代聊工人。AI 的冲击是把这条流水线整个接管:12 元包月、无限畅聊、永远在线、记得你的生日——供给曲线从「稀缺、按次」翻转为「无限、订阅」。于是最贵的问题变成了「对面是谁」:模型冒充真人收你的钱,真人冒充 AI 赚 0.05 美元一条。但 AI 压低的只是模拟亲密的成本:触碰与在场无法数字化,被识破的暧昧会瞬间崩塌,而它最恶性的产出是非自愿——真实,正在变成这个光谱上最后的奢侈品。This industry never sold love; it sells «a unit of intimacy for sale» — being seen, being answered, being favoured, being fantasised. Traditionally each unit is produced live by a real person: chatters bill by the message (¥8 buys 14), streamers do private-message «homework» after going offline, subscription creators outsource DMs to workers juggling seven inboxes. AI takes over the whole line: ¥12 a month, unlimited chat, always on, remembers your birthday — the supply curve flips from «scarce, per-use» to «infinite, subscription». So the most expensive question becomes «who is on the other end»: a model impersonating a human to take your money, a human impersonating AI for $0.05 a message. But AI only lowers the cost of simulated intimacy: touch and presence cannot be digitised, a discovered flirtation collapses instantly, and its most malignant output is the non-consensual — the real is becoming the last luxury on this spectrum.
✓ 本图相信This map believes供给曲线已经翻转——真人聊手被 AI 批量替代是现在时,不是预言;替代力沿「虚→实」单调递减,身体在场是 AI 攻不动的终点;分成瀑布与幂律是产业底色,AI 没有改变分配、只是省掉了聊手工资;深伪 NCII 与 AI-CSAM 是犯罪且高度性别化,「虚拟伤害」就是真实伤害;治理重心正确的落点是平台责任 + 年龄验证 + AI 身份披露,监管对象已从「内容」升维到「关系」;对个体,任何线下环节的人身与法律风险都大于收益。The supply curve has flipped — human chatters being replaced en masse by AI is present tense, not prophecy; replacement power decays monotonically along «virtual → physical», bodily presence being the terminus AI cannot breach; the split-waterfall and power law are the industry’s ground colour, AI changed not the distribution but merely saved the chatter’s wage; deepfake NCII and AI-CSAM are crimes and highly gendered — «virtual harm» is real harm; governance’s correct landing is platform liability + age verification + AI-identity disclosure, its object risen from «content» to «relationship»; for the individual, the bodily and legal risk of any offline step exceeds the reward.
✗ 本图不相信This map does not believe不信「AI 终结了最古老的产业」——它只替代了可数字化的那部分,把定价从按次改成订阅;不信「AI 伴侣是孤独的解药」也不信「纯属毒药」——证据是剂量效应,中度缓解、重度恶化,而商业激励天然推向重度;不信「对面是真人」的默认假设——双向欺诈时代,未披露即风险;不信检测器能赢军备竞赛——生成方永远先手,骨架只能是哈希下架与平台义务;不信「虚拟供给会自动消化线下风险」——供给侧的技术挤出正把最弱者推向更高风险端;不信任何绕开「谁承担风险、谁拿走利润、谁拥有退出权」三问的产业叙事。Not «AI ended the oldest industry» — it only replaced the digitisable part, switching pricing from per-use to subscription; not «AI companions are the cure for loneliness» nor «pure poison» — the evidence is a dose effect, moderate use eases, heavy use worsens, while the commercial incentive pushes toward heavy; not the default assumption that «the other end is human» — in the age of two-way deception, undisclosed means risk; not that detectors can win the arms race — the generator always moves first, the skeleton can only be hash-takedown and platform duty; not that «virtual supply automatically absorbs offline risk» — the technical squeeze-out is pushing the weakest toward the higher-risk end; not any industry narrative that dodges «who bears the risk, who takes the profit, who owns the exit».
B/C
0.57→12
真人陪聊 0.57 元/条 vs AI 伴侣 12 元/月无限——供给曲线翻转¥0.57/msg human chat vs ¥12/mo unlimited AI companion — the supply curve flips
B
$72亿
OnlyFans 2024 粉丝总支付额(GMV,≠营收)——订阅制亲密经济体量OnlyFans 2024 fan spend (GMV, ≠ revenue) — the size of subscription intimacy
B
八成<8千
1508 万职业主播八成以上月入不足 8000 元——卖亲密者拿零头80%+ of 15.08M professional streamers earn under ¥8,000/mo — those who sell intimacy get the crumbs
B
99%
深伪色情受害者为女性(2023 安全厂商口径)——harm 层的性别化铁证of deepfake-porn victims are women (2023 security-vendor sample) — the harm layer’s gendered proof
⚖ 口径裁判 · 先读我Calibration notes · read me first①「0.57 元/条」为交友 App 实测(8 元 14 条,2021,C 级),「12–14 元/月」为国内头部 AI 伴侣订阅价——两数为「单价对撞」示意,非同年同平台严格可比;②「72 亿美元」是 OnlyFans 2024 财年粉丝总支付额(GMV),平台自身营收(抽 20% 后)约 14 亿美元——GMV≠营收,勿混写;③「八成<8000 元」=以网络主播为主业的 1508 万人权威调查口径(含全品类非纯秀场);④「99%」=2023 安全厂商报告(样本 95,820 段)受害者性别占比,与 2019 另一报告「96% 为色情」(样本 14,678 段)是两份不同年份的安全厂商样本统计,非全网普查,不得写成同一「最新数据」;⑤AI 伴侣市场规模用实测 App 消费(2025 全年约 1.2 亿美元),研报「367.9 亿美元」为含硬件/企业宽口径、相差近 300 倍,仅作口径警示;⑥直播分成多口径(平台 50/公会 25/主播 25;团播 50/20)是不同链路层级叠加,本页用瀑布不用单数;⑦「AI 冲击强度 %」与「被压缩速度排序」为编者综合评估值,非统计值;⑧本页综合四份独立深度研究交叉整理,为批判性产业结构分析,不构成任何建议;不提供任何交易渠道、话术或规避监管方法;涉未成年人与非自愿影像内容均属犯罪,仅作治理与风险分析。① «¥0.57/msg» is a measured dating-app rate (¥8 for 14 msgs, 2021, grade C); «¥12–14/mo» is a leading domestic AI-companion subscription — the two are a «unit-price collision» illustration, not strictly comparable across year/platform. ② «$7.2B» is OnlyFans 2024 gross fan spend (GMV); the platform’s own revenue (after a 20% cut) is ~$1.4B — GMV ≠ revenue. ③ «80% < ¥8,000» is an authoritative survey of the 15.08M whose main job is streaming (all categories, not pure show). ④ «99%» is a 2023 security-vendor report (sample 95,820) on victim gender, distinct from a 2019 report’s «96% is porn» (sample 14,678) — two different-year vendor samples, not a census, never merged as one «latest figure». ⑤ AI-companion market size uses measured app spend (~$120M for 2025), while a research-firm’s «$36.79B» is a broad hardware/enterprise scope ~300× larger, cited only as a caveat. ⑥ Streaming splits (platform 50 / guild 25 / streamer 25; team-show 50/20) stack across chain layers; shown here as a waterfall, not a single number. ⑦ «AI-impact %» and the compression ranking are editorial estimates, not statistics. ⑧ Compiled from four independent deep-research reports as critical industry-structure analysis; not any form of advice; provides no transaction channel, script, or means to evade regulation; content involving minors or non-consensual imagery is criminal and appears only as governance/risk analysis.
◎ 中心裂缝 · 全图的坐标原点The central rift · origin of this map
AI 把「被回应」变成了工业品,最贵的问题变成了「对面是谁」AI industrialised «being answered»; the priciest question became «who is on the other end»
把这个产业沿「亲密的虚实程度 × 法律地位」双轴摊开,会看到一条清晰的裂缝。纵向:AI 的替代力沿「虚→实」单调递减——越纯数字(擦边图、标准化聊天),AI 越能整包接管;越需要身体在场(触碰、线下),AI 越只能生产「前戏与诱饵」。横向:治理重心正从「内容管制」上移到平台责任 + 年龄验证 + 身份真实性。两轴交叉处,是全图的核心命题:当「被回应」可以被无限量、零边际成本地生产,亲密的稀缺性就从「供给」转移到了「真实性」——你付的钱,买的到底是一个真人的选择,还是一段脚本?Unroll this industry along «degree of virtuality × legal status» and a clean rift appears. Vertically: AI’s replacement power decays monotonically along «virtual → physical» — the more purely digital (suggestive images, standardised chat), the more AI takes over wholesale; the more bodily presence is needed (touch, offline), the more AI can only produce «foreplay and bait». Horizontally: governance’s centre is rising from «content control» to platform liability + age verification + identity authenticity. Where the axes cross sits the map’s core question: once «being answered» can be produced infinitely at zero marginal cost, the scarcity of intimacy shifts from «supply» to «authenticity» — what your money buys, a real person’s choice, or a script?
线下端 = 违法区Offline = the illegal zone在中国语境此端是法律与人身风险最重处,不是堡垒(仅作法律与社会分析)in the Chinese context this end carries the heaviest legal and personal risk, not a fortress (legal/social analysis only)
一句话读法:AI 压平了「模拟亲密」的价格,却抬高了「真实亲密」的相对价格。它没有改变这个产业的分配(卖亲密的人依然拿零头),只是省掉了聊手工资;它最恶性的产出不是效率,是「非自愿」。四色贯穿全页:粉=已被替代 · 琥珀=混合重构 · 紫=攻不动之墙 · 红=犯罪红线。与其他图不同:这里最难自动化的那端不是从业者的避风港,而是法律与人身风险最重处——墙内没有红利,只有红线。One-line reading: AI flattened the price of «simulated intimacy» while raising the relative price of «real intimacy». It changed not the industry’s distribution (those who sell intimacy still get crumbs), only saved the chatter’s wage; its most malignant output is not efficiency but the «non-consensual». Four colours run through the page: pink = replaced · amber = hybrid · violet = the wall · red = the criminal red line. Unlike other maps: the end hardest to automate is not the worker’s harbour but the point of heaviest legal and personal risk — behind the wall there is no dividend, only the red line.
ASPECTRUM · 主脊 · 亲密商品化光谱The spine · the commodified-intimacy spectrum
不是流水线,是一根光谱:虚→实越深,AI 越攻不动;法律越重,治理越上移Not a line but a spectrum: the deeper virtual→physical, the less AI breaches; the heavier the law, the higher governance rises
这个产业的最小可售单元不是「一次服务」,而是「一单位被出售的亲密」——一条暧昧消息、一次连麦、一个专属称呼、一段付费内容、一个记得你生日的 AI 人格。各档不是上下游,而是同一种「亲密单元」落在不同虚实程度与法律重量上。AI 冲击强度沿「虚→实」递减(编者评估值,非统计);法律状态沿光谱右移逐格加重。两条横切层(harm/寄生)贯穿全谱。The smallest sellable unit here is not «a service» but «a unit of intimacy for sale» — a flirty message, a voice-room session, a pet name, a paid clip, an AI persona that remembers your birthday. The tiers are not upstream/downstream but the same «intimacy unit» landing at different degrees of virtuality and legal weight. AI-impact decays along «virtual → physical» (editorial estimate, not statistics); legal status intensifies tier by tier to the right. Two cross-cutting layers (harm / parasitic) run through the whole spectrum.
BTHREE SHIFTS · AI 重构的三大传播点The three shifts AI drives
供给翻转、对面是谁、真人 IP + AI 劳动The supply flip, «who is it», and real-IP-plus-AI-labour
四份底稿在三个「传播点」上收敛——它们是本图最锋利、最出圈的三张牌。The four sources converge on three «shifts» — the map’s sharpest, most shareable cards.
THE LEDGER · 诚实层The honesty ledger
卖亲密的人拿到的是零头——AI 没改分配,只省了聊手工资Those who sell intimacy get the crumbs — AI changed not the split, only saved the chatter’s wage
一笔 100 元的打赏,到主播手里还剩多少?分成瀑布是这个产业的底色,幂律是它的形状——AI 没有改变任何一层,它只是让平台连「聊手工资」都省了。数字为多口径并列,页面用瀑布不用单数(见口径裁判)。Of a ¥100 tip, how much reaches the streamer? The split-waterfall is this industry’s ground colour, the power law its shape — AI changed no layer, it merely let platforms save even «the chatter’s wage». Figures are multi-source; shown as a waterfall, not a single number (see calibration notes).
幂律三层同构:同一个「头部吃掉一切」的形状,在创作者、App、主播三层反复出现——AI 内容的海量供给只会让中腰部被进一步稀释。The power law, three layers isomorphic: the same «the top eats everything» shape recurs across creators, apps and streamers — AI’s flood of content only dilutes the middle further.
CHARD BONES · 硬骨头Hard bones
AI 攻不进的七块:结构性地难,非算力可解Seven things AI cannot breach: structurally hard, not solvable by compute
这七堵墙决定了替代力为何沿虚→实递减;但要记住——本图的硬骨头不是从业者的避风港,最难自动化的那端(身体在场/线下)在中国语境恰是违法与人身风险最重处。These seven walls dictate why replacement decays virtual→physical; but remember — this map’s hard bones are not the worker’s harbour: the end hardest to automate (bodily presence / offline) is, in the Chinese context, the point of heaviest illegality and personal risk.
DSTANCE MAP · 立场地图Stance map
你站在哪:按角色 × 光谱替代三层过滤Where you stand: filter by role × the three replacement tiers
选「我是谁」高亮相关卡;切三层看替代规律——已被替代(纯数字) / 混合重构(真人+AI) / 攻不动之墙(在场/合意/违法区)。开关本身就是「AI 替代力单调递减、法律风险单调递增」双规律的交互证明。Pick «who I am» to highlight cards; toggle the three tiers to see the rule — replaced (purely digital) / hybrid (human + AI) / the wall (presence / consent / the illegal zone). The toggles are the interactive proof of «AI-replacement decays, legal risk rises».
哪一层WHICH TIER已被替代 · 纯数字Replaced · digital混合重构 · 真人+AIHybrid · human+AI攻不动之墙 · 红线The wall · red line
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——各方并置,再给编者合议,不裁掉任何一方。The four sources disagree below — each side shown, then an editorial synthesis; no side is cut.
TIMELINE · 时间线Timeline
2018–2026:供给翻转 ↑ vs 治理加码 ↓ 的双轨2018–2026: the twin tracks of supply-flip ↑ and governance-ratchet ↓
底稿对 2018 前极薄(标 ※ 为编者补注),2023 生成式 AI 爆发后骤厚(深伪立法军备竞赛 + 中国专项监管)——薄区如实留稀。页面上线期恰逢中国拟人化互动办法 2026-07-15 施行节点。Sources are thin before 2018 (※ = editorial note) and thicken sharply after the 2023 generative-AI surge (the deepfake-legislation arms race + China’s targeted regulation) — sparse zones left honestly sparse. This page ships right at China’s «anthropomorphic-interaction» rules taking effect on 2026-07-15.
BOUNDARIES · 边界戒律(非导购)Boundaries (not a buying guide)
四种人的「认清结构 / 风险自查 / 边界戒律」«See the structure / self-check the risk / hold the line» for four roles
本图不做「现在能用」推荐——与 bet 图同款批判性定位,改为防御性三栏;线下端与 harm 层只作法律与社会分析。行业分析,不构成任何建议。This map offers no «use now» recommendations — like bet, a defensive three-column framing; the offline end and harm layer appear as legal/social analysis only. Industry analysis, not any form of 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
L1
供给翻转:亲密从「稀缺按次」变「无限订阅」The supply flip: intimacy from «scarce per-use» to «infinite subscription»
AI 第一次把「被回应」变成边际成本趋零的工业品:真人聊手 0.57 元/条,AI 伴侣 12 元/月无限。但 AI 压低的只是模拟亲密的成本,不解决真实亲密的稀缺——它把定价从按次改成订阅,没终结这个产业。AI first turned «being answered» into a near-zero-marginal-cost product: ¥0.57/msg human vs ¥12/mo unlimited AI. But AI lowers only the cost of simulated intimacy, not the scarcity of real intimacy — it switched pricing from per-use to subscription, it did not end the industry.
越纯数字,AI 越能整包接管;越需要身体在场,AI 越只能生产「诱饵」。而沿法律轴反向:光谱右移一格,风险指数级上升——最难自动化的那端不是避风港,是违法区。这与其他图「硬骨头=人的堡垒」结构相反。The more purely digital, the more AI takes over wholesale; the more bodily presence is needed, the more AI produces only «bait». The legal axis runs opposite: each tier rightward, risk rises exponentially — the end hardest to automate is not a harbour but the illegal zone. This inverts other maps' «hard bones = the human fortress».
L3
「对面是谁」成为新溢价:真实性是最后的奢侈品«Who is it» becomes the new premium: authenticity is the last luxury
双向欺诈已进入司法:模型冒充真人氪金、真人冒充 AI 打工。当 AI 内容泛滥,「真人回复」「本人出镜」「无 AI 代聊」「可验证同意」成为新溢价标签——新奢侈从「你在和谁聊」转向「你能否确认 TA 是真的」。AI 身份披露正成为法定义务。Two-way deception is now in court: models impersonating humans to charge you, humans impersonating AI to earn. As AI content floods, «real reply», «really on camera», «no AI ghost-chat», «verifiable consent» become new premiums — the new luxury shifts from «who you chat with» to «whether you can confirm they are real». AI-identity disclosure is becoming a legal duty.
L4
孤独经济的悖论:缓解孤独的产品在再生产孤独The loneliness-economy paradox: the cure reproduces the disease
证据是剂量效应:中度使用 AI 伴侣可缓解孤独,重度使用与更高孤独、依赖、现实社交减少相关。AI 被设计成无限包容、永远讨好的「回音室」,剥离真实关系里的冲突与磨合——而商业激励(留存/氪金)天然把用户推向重度。监管把「诱导情感依赖」写入禁止条款,正是对这一激励的矫正。The evidence is a dose effect: moderate use of AI companions eases loneliness; heavy use correlates with more loneliness, dependence and less real-world contact. AI is designed as an infinitely accommodating, ever-pleasing «echo chamber», stripping out the conflict and friction of real bonds — while the commercial incentive (retention/spend) pushes toward heavy use. Regulators writing «inducing emotional dependence» into prohibitions is the correction to that incentive.
L5
治理:管住平台与身份,比管住每一张图有效Governance: controlling platforms and identity beats controlling every image
内容管制是打不赢的猫鼠游戏(年龄验证一生效,VPN 注册 +1400%);检测器在真实环境掉准 45–50%。真正的骨架是平台责任(48h 下架)+ 年龄验证(隐私保护型)+ AI 身份披露 + 哈希共享网络。窄口径重打击:火力集中在非自愿影像、未成年人、暴力剥削三个高危场景;对内容层保持「减损」预期而非「根治」幻想。Content control is an unwinnable cat-and-mouse (age verification took effect, VPN sign-ups +1400%); detectors lose 45–50% accuracy in the wild. The real skeleton is platform liability (48h takedown) + age verification (privacy-preserving) + AI-identity disclosure + hash-sharing networks. Narrow scope, hard hits: concentrate fire on non-consensual imagery, minors, and violent exploitation; expect «harm reduction», not a «cure», for the content layer.
STANCE · 本图立场Where this map stands本图站在批判性祛魅一侧:既不替「AI 女友终结孤独」的叙事抬轿,也不替「最古老产业永不消失」的宿命论背书。我们把四份高度同构的深研底稿拆开重验:框架共识多为研究命题的回声,只有十条事实独立收敛——聊手替代已发生、订阅碾压按次、替代力沿虚实递减、双向欺诈进入司法、深伪高度性别化、治理重心上移、越用越孤独的剂量悖论、幂律三层同构、支付网络是事实监管者、未成年人保护最高优先。这个产业最擅长把「被设计的冲动」讲成「你被偏爱」、把「模型的回声」讲成「有人懂你」——本图的全部工作,是把这些重新拆开,并始终追问三件事:谁承担风险、谁拿走利润、谁拥有退出权。数字答案:风险在个体,利润在平台与组织者,退出权被数字足迹与深伪进一步没收。This map stands on critical demystification: it carries the sedan chair neither for «the AI girlfriend ends loneliness» nor for the fatalism of «the oldest industry never dies». We took four highly isomorphic reports apart and re-verified: most framework consensus echoes the commissioning prompt; only ten facts converge independently — chatter replacement has happened, subscription crushes per-use, replacement decays virtual→physical, two-way deception is in court, deepfakes are highly gendered, governance’s centre has risen, the dose-paradox of «more use, more loneliness», the three-layer power law, payment networks as de-facto regulators, and minors’ protection as top priority. This industry excels at narrating «engineered impulse» as «you are favoured» and «a model’s echo» as «someone understands you» — the whole job of this map is to take those apart, always asking three things: who bears the risk, who takes the profit, who owns the exit. The numeric answers: risk on the individual, profit with platforms and organisers, the exit further confiscated by digital footprints and deepfakes.
⚠ 边界与红线:本图为批判性产业结构分析,不构成任何建议;不提供任何交易渠道、招徕话术或规避监管的方法;光谱的线下端仅作法律与社会分析。涉及未成年人的性内容、以及未经同意的私密/深伪影像,均属犯罪——本图仅作治理与风险分析。若你正遭遇裸聊勒索或非自愿影像扩散:停止支付、固定证据、向平台举报并报警;可用 StopNCII / NCMEC「Take It Down」类哈希工具申请下架,责任在加害者。⚠ Boundaries and red lines: this map is critical industry-structure analysis, not any form of advice; it provides no transaction channel, solicitation script, or means to evade regulation; the spectrum’s offline end is legal/social analysis only. Sexual content involving minors, and non-consensual intimate or deepfake imagery, are crimes — covered here only as governance/risk analysis. If you face sextortion or non-consensual image spread: stop paying, preserve evidence, report to the platform and police; hash tools like StopNCII / NCMEC «Take It Down» can request removal — the fault lies with the perpetrator.