The Gendered Geographies of AI Labor AI 劳动的性别地理

"Silicon Valley wife" usually arrives as a joke: yoga, school districts, stock options, Whole Foods, a husband who is always almost changing the world. The joke works by making her seem adjacent to technology rather than inside it. But if we take the phrase seriously, and rescue it from contempt, it points to one of AI's least acknowledged infrastructures: the household that makes the high-intensity AI worker reproducible.

That household labor is not sentimental background. It organizes meals, sleep, child care, housing, visas, school choices, moves between cities, elder care, social calendars, emotional repair, and the management of future risk. It makes possible a form of labor that can present itself as pure technical intensity. The engineer, founder, researcher, or investor may appear as an individual genius, but the working day has already been prepared elsewhere.

This is where a feminist geography of AI begins: not only with the campus, model, chip, cloud, or data center, but with social reproduction. The question is not whether the Silicon Valley wife is privileged. Often she is. The sharper point is that even privilege can expose an infrastructure. The supposedly private household helps produce the public capacity of AI work.

The second figure is the "woman coder", or in Chinese internet language, the chengxuyuan, the "program girl". She is inside the formal labor system: writing code, building models, debugging products, designing interfaces, translating user needs into technical decisions. Yet the same workplace often assigns her the work that holds the system together without naming it as invention: documentation, communication, onboarding, mentoring, ethics, safety, user empathy, conflict resolution, and project memory.

This is the gender of maintenance inside a culture of innovation. AI work celebrates breakthroughs, benchmarks, launches, and founders. It is less fluent at recognizing the work that makes collaboration reliable. Women in AI are not only underrepresented at the top; they are also frequently overrepresented in the relational work that allows technical teams to keep moving.

The third figure is the "optimized" digital woman worker. AI does not meet a neutral labor market. It arrives in offices where administrative support, clerical processing, customer service, HR, translation, scheduling, reporting, and routine writing have long been feminized. The first effect is not always disappearance. More often, the job is optimized: one task is automated, one is outsourced, one becomes monitoring, one becomes exception handling, and one is folded into someone else's workload.

The numbers make the pattern difficult to ignore. Brookings estimates that 36% of female workers, compared with 25% of male workers, are in occupations where GPT-4 could save at least half of task time. A 2026 Brookings analysis identifies 6.1 million U.S. workers who face both high AI exposure and low adaptive capacity; about 86% of them are women, many in clerical and administrative roles. The point is not to say that exposure equals job loss. It is to say that AI exposure lands on an already gendered map of savings, age, local opportunity, training access, and bargaining power. "Optimization" names this quietly violent transition: work is made measurable before it is made replaceable.

The fourth figure is the prompt girlfriend. The phrase is unstable on purpose: part tutorial persona, part AI companion script, part platform fantasy of endless emotional availability. She may be teaching a user how to speak to a model, performing intimacy around the model, or becoming the feminized surface through which the model feels warmer, softer, and more obedient. In this role, prompting is not only technical skill. It is affective interface labor.

The fifth figure is the influencer. She may be teaching prompts, selling AI templates, narrating career survival, turning optimization anxiety into content, or performing the optimism of the side hustle. In this scene, AI anxiety becomes platform labor. The woman who might be made insecure by automation is also invited to display her adaptability in public: learn the tool, brand the transition, monetize the lesson, reassure the audience that no one has to be left behind.

This is not fake empowerment. It is labor under platform visibility. Making a self into proof of adaptation takes time, affect, editing, search optimization, audience management, and risk. The influencer converts uncertainty into a feed. Her work makes AI feel learnable, survivable, and intimate, even as it often privatizes the responsibility for surviving structural change.

The sixth figure is the clickworker. She sits at the end of an outsourcing chain where judgment becomes a sequence of small paid acts: label this image, rank this answer, test this refusal, flag this harm, compare these outputs, click the box that makes the model safer. The word matters because it strips away the romance of "training AI." A model is trained through fragmented attention, repeated decisions, and piecework that is easy to count and hard to honor.

These figures should not be collapsed into one universal woman. They occupy different class positions, regions, contracts, languages, and degrees of visibility. The Silicon Valley wife may live near wealth. The woman engineer may hold a scarce credential. The optimized digital woman worker may have fewer savings and fewer local alternatives. The prompt girlfriend may turn intimacy into interface work. The influencer may turn visibility into income or exhaustion. The clickworker may work through an outsourcing chain in a city far from the platform whose system she improves.

Gender does not erase these differences. It organizes how they connect. It helps decide which labor is called genius, which is called support, which is called flexibility, which is called low skill, and which is not named at all. A gendered geography of AI labor therefore asks where AI workers are reproduced, where AI risks are deposited, where AI skills are made visible, and where AI value returns after passing through hidden work.

The familiar map of AI begins with Silicon Valley, Seattle, Shenzhen, Beijing, data centers, GPUs, venture capital, and cloud contracts. That map is not wrong. It is incomplete. The gendered map adds kitchens, bedrooms, Slack channels, HR queues, customer-service dashboards, TikTok studios, companion scripts, vocational classrooms, clickwork queues, and moderation screens. It shows that AI is not only built. It is made livable, teachable, intimate, marketable, safe, and profitable through forms of work that have long been coded feminine.

“硅谷娇妻”通常像一个玩笑出现：瑜伽、学区房、期权、Whole Foods、以及一个永远在“改变世界”的丈夫。 这个玩笑之所以成立，是因为它把她放在技术之外，好像她只是 AI 产业旁边的附属人物。 但如果我们认真对待这个词，并把它从轻蔑里救出来，它其实指向 AI 最少被承认的一层基础设施： 那个让高强度 AI 工作者得以持续存在的家庭。

这种家庭劳动并不是温情背景。它组织饮食、睡眠、育儿、住房、签证、学区、城市迁移、照顾老人、 社交安排、情绪修复和未来风险管理。它让某种劳动可以把自己呈现为纯粹的技术强度。 工程师、创业者、研究员或投资人看起来像独立的天才，但他的工作日早已在别处被准备好了。

女性主义的 AI 地理，可以从这里开始：不只是从园区、模型、芯片、云或数据中心开始， 也从社会再生产开始。问题不是“硅谷娇妻”是否拥有阶层特权，很多时候她当然拥有。 更锋利的问题是：即使是特权，也会暴露一种基础设施。看似私人的家庭，参与生产了公共的 AI 劳动能力。

第二个形象是“程序媛”。她在正式劳动系统内部：写代码、做模型、调产品、设计界面， 把用户需求翻译成技术决策。但同一个工作场所，也常常把维持系统运转的劳动交给她： 文档、沟通、入职培训、带新人、伦理、安全、用户理解、冲突协调和项目记忆。 这些工作让团队能够继续前进，却很少被命名为“创新”。

这就是创新文化内部的维护性别。AI 工作喜欢庆祝突破、榜单、发布和创始人。 它不太擅长承认让协作变得可靠的劳动。AI 中的女性并不只是“高层人数不足”； 她们也经常被放到关系性、维护性、解释性工作的位置上，让技术团队看起来可以顺畅运作。

第三个形象是“被优化的数字女工”。AI 并不是进入一个中性的劳动市场。它进入的是早已性别化的办公室： 行政支持、文书处理、客服、人事、翻译、排期、报告和常规写作，本来就常被视作女性化劳动。 AI 的第一重影响也不一定是岗位直接消失。更多时候，工作被“优化”：一部分自动化，一部分外包， 一部分变成监控，一部分变成异常处理，还有一部分被塞进另一个人的工作量里。

数据让这个模式很难被忽视。Brookings 估计，36% 的女性劳动者所在职业中，GPT-4 可能节省至少一半任务时间； 男性劳动者对应比例为 25%。Brookings 2026 年的另一项分析指出，美国有 610 万劳动者同时处在高 AI 暴露和低适应能力位置， 其中约 86% 是女性，许多人集中在文书和行政岗位。这里的重点不是把“暴露”直接等同于“失业”， 而是说明 AI 暴露落在一张已经性别化的地图上：储蓄、年龄、地方机会、培训渠道和议价能力，都分布不均。 “优化”正是这个安静而暴力的过渡：劳动先被度量，随后才变得可替换。

第四个形象是提示词女友。这个词本身就不稳定：它一部分是教程人设，一部分是 AI 伴侣脚本， 一部分是平台对无限情绪可用性的幻想。她可能在教用户如何对模型说话，也可能围绕模型表演亲密， 或者成为那个让模型显得更温柔、更顺从、更像“懂你”的女性化界面。在这个位置上，提示词不只是技术能力， 也是一种情感界面劳动。

第五个形象是网红博主。她可能在教提示词，卖 AI 模板，讲职业转型，把优化焦虑变成内容， 或者表演一种副业式的乐观。在这个场景中，AI 焦虑变成了平台劳动。 一个可能因自动化而不安的女性，同时被邀请在公共空间里展示自己的适应力： 学会工具，包装转型，出售课程，安慰观众，证明没有人一定会被时代抛下。

这不是假的赋权，而是平台可见性之下的劳动。把自我变成“适应 AI 的证据”，需要时间、情绪、 剪辑、搜索优化、受众管理和风险承担。博主把不确定性转化为信息流。她让 AI 看起来可学习、 可存活、可亲近，但也常常把结构性变化的生存责任重新推回个人身上。

第六个形象是点击女工（clickworker）。她在外包链条末端，把判断变成一次次可计价的小动作： 标注这张图、排序这个回答、测试这次拒绝、标记这个伤害、比较这组输出、点击那个让模型显得更安全的方框。 这个词重要，因为它把“训练 AI”的浪漫感拆掉。模型是在碎片化注意力、重复判断和计件劳动中被训练出来的； 这些劳动容易被计数，却很难被尊重。

这些形象不能被压成一个统一的“女性”。她们处在不同的阶层、地区、合同、语言和可见性之中。 硅谷娇妻可能生活在财富附近。女性工程师可能拥有稀缺文凭。被优化的数字女工可能缺少储蓄和地方替代机会。 提示词女友可能把亲密转化为界面劳动。网红博主可能把可见性变成收入，也可能变成耗竭。 点击女工可能通过外包链条工作，身处一个离平台总部很远、却持续改善平台系统的地方。

性别并不会抹平这些差异。它组织这些差异之间的连接。它帮助决定什么劳动被称为天才， 什么劳动被称为支持，什么劳动被称为灵活，什么劳动被称为低技能，什么劳动干脆没有名字。 因此，AI 劳动的性别地理要问的是：AI 工作者在哪里被再生产？AI 风险在哪里沉积？ AI 技能在哪里变得可见？AI 价值又在穿过哪些隐形劳动之后回流？

我们熟悉的 AI 地图，从硅谷、西雅图、深圳、北京、数据中心、GPU、风险资本和云合同开始。 这张图并没有错。它只是不完整。性别化的地图还要加入厨房、卧室、Slack 频道、人事队列、 客服看板、短视频工作室、伴侣脚本、职业教室、点击队列和审核屏幕。它说明，AI 不只是被建造出来； 它还通过那些长期被编码为女性化的劳动，被变得可生活、可教学、可亲密、可营销、可安全、可盈利。