What 81,000 People Want from AI / 81000 人想要什么样的 AI
作者: @AnthropicAI
原文链接: https://www.anthropic.com/features/81k-interviews
语言: 英→中双语
中文翻译
81,000人期望从AI中得到什么
去年十二月,全球成千上万的Claude用户与我们的AI面试官进行了对话,分享他们如何使用AI,他们梦想AI能实现什么,以及他们担心AI可能做什么。
摩洛哥
每个点代表4位受访者
跳转到故事
首次,AI使我们能够以非凡的规模收集丰富、开放式访谈。 我们听到了来自70种语言、159个国家的声音。我们相信这是迄今为止最大、最多语言的质量研究。
AI已经在帮助人们,并激发希望……
“Claude将历史碎片拼凑在一起,在我被误诊超过9年后得到了正确的诊断。
美国自由职业者
“我过着朝不保夕的生活,没有储蓄。如果我更聪明地使用AI,它可能帮助我找到摆脱这种循环的方法。这还是取决于我。
尼日利亚企业家
但这也正在让人们付出代价,并引发警报……
“五月份我被解雇了,因为我的公司想用AI系统取代我。
美国技术支持专家
“人类从未遇到过比自身更聪明的事物。我们需要反思如何为AI时代做准备。
韩国软件工程师
在访谈中,希望和恐慌并没有将人们分成阵营,更多的是每个人内心的紧张。
“我使用AI来审查合同,节省时间……同时我也担心:我是不是正在失去自己阅读的能力?思考是最后的边疆。
以色列律师
关于AI的公共讨论通常集中在其风险和收益的抽象预测上。很大程度上缺失的是对“AI做得好”的愿景,这是基于全球各地已经使用AI并开始形成对其可能为自己做什么的感觉的人们的具体愿望。
因此,我们询问了我们的用户他们对AI的希望和担忧,以及他们的观点如何与他们对技术的实际体验相联系。在12月的某一周,我们邀请了所有拥有Claude.ai账户的人与Anthropic Interviewer——一种被提示进行对话访谈的Claude版本——坐下来告诉我们他们如何看待AI。来自159个国家和70种语言的80,508人接受了访谈。我们相信这是迄今为止最大、最多语言的质量研究。¹
以下是他们关于他们希望AI在他们生活中扮演什么角色,它是否已经填补了这个角色,以及他们担心在过程中可能会出错的内容。我们还建立了一个引言墙,您可以在这里直接听到人们的声音。
引言墙
浏览来自世界各地的声音——按地区、关注点、愿景等筛选。
查看引言
看到森林和树木
Anthropic Interviewer向每位受访者提出了一系列关于他们希望和不想从AI中得到什么的问题,然后根据回答调整后续问题。这种方法弥合了定性研究中常见的深度和数量之间的权衡,使我们能够在非常大的规模上收集丰富、开放式访谈。
为了理解这大量信息,我们构建了Claude驱动的分类器,将每个对话根据一系列维度进行分类——人们希望从AI中得到什么,他们是否得到了他们想要的,他们担心什么,他们的职业(如果提到),以及他们对AI的整体感受。“人们希望从AI中得到什么”被分类为每位受访者的一个主要类别,而担忧是多标签的——一次访谈可以收到多个代码,因为受访者往往表达几个不同的担忧,而不是一个。
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
我们还使用Claude提取了代表性引语。在用户选择参与之前,他们被告知他们的回答将用于研究,并且Anthropic可能会在去除个人识别信息的研究结果中发布回答。所有回答在Anthropic的小型研究团队进行分析之前都进行了匿名处理,并经过进一步的人工审查以删除任何可能识别的细节,以帮助保护受访者的隐私和公众匿名性。答案反映了人工智能的广泛使用(即不仅仅是Claude),尽管我们删除了其他人工智能产品的名称。
附录更详细地描述了我们的方法,以及局限性和一些额外的分析。
人们从人工智能中希望得到什么
我们要求Claude识别和分类每个人最希望从人工智能中得到什么:
人们希望实现的目标
01.
专业卓越
18.8%
通过让AI处理日常任务,提高效率,专注于更有价值的工作,复杂问题解决和专业精通。
“我每天从医生和护士那里收到100-150条短信。我大量的认知劳动都花在了文档上...自从实施AI以来,文档的压力已经减轻。我对护士更有耐心,有更多时间向家人解释事情。”
美国医疗工作者
阅读更多关于专业卓越的引语
02.
个人转变
13.7%
在AI的引导、教练或支持下实现个人成长、情感福祉或生活转变——例如自我理解、行为改变、治疗支持、陪伴、改善身心健康。
“AI为我树立了情商的榜样...我可以用这些行为与人类互动,成为一个更好的人。”
匈牙利
阅读更多关于个人转变的引语
03.
生活管理
13.5%
AI作为全面的组织支持和认知支架——例如管理日程、减轻心理负担、执行功能支持。
“如果AI真正处理了心理负担……它将给我带来无价的东西:全神贯注的注意力。”
丹麦经理/高管
阅读更多关于生活管理的引语
04.
时间自由
11.1%
从工作和家务中夺回时间,与家人或朋友共度时光,追求爱好,旅行,休息。
“在AI的支持下,我现在可以按时下班去接孩子放学,喂他们吃饭,陪他们玩。”
墨西哥软件工程师
阅读更多关于时间自由的引语
05.
经济独立
9.7%
通过AI实现经济自由或经济安全——例如收入生成、商业建设、投资、被动收入,或逃离经济限制。
“在AI完成工作的同时放松,积累财富。它只是我的影子,一个非常非常长的影子。”
洪都拉斯企业家
阅读更多关于经济独立的引语
06.
社会变革
9.4%
使用AI解决重大社会挑战——例如贫困、疾病、气候、不平等——以实现广泛的繁荣,而不是个人利益。
“如果AI加速有助于找到治疗方法,我的女儿在世界上将有机会平等。那是我最关心的事情。”
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
软件工程师,波兰
了解更多关于社会变革的引言
07.
创业
8.7%
利用AI作为力量倍增器构建、启动和扩展业务——例如产品开发、业务自动化或独立创业,但具有团队级能力。
“我在一个技术劣势国家,我无法承受许多失败。有了AI,我在网络安全、UX设计、营销和项目管理方面同时达到了专业水平。找到我地区可用的支付平台需要我一个月时间。AI在30秒内就做到了。这是一个平衡器。”
创业者,喀麦隆
了解更多关于创业的引言
08.
学习与成长
8.4%
将AI用作学习加速器和个性化教师——获取知识、发展技能、掌握复杂主题、满足智力好奇心。
“我和一个AI一起为我最大的孩子准备教育材料——让AI同时担任导师和课程专家。我们昨天收到了[我孩子]的报告,他在他学习的每个学术领域都被评为‘高于’或‘远高于’标准。”
澳大利亚
了解更多关于学习与成长的引言
09.
创意表达
5.6%
利用AI帮助将创意愿景变为现实——例如艺术、游戏、音乐、电影、书籍——通过克服想象与执行之间的障碍。
“在AI之前,我的游戏需要3年——我不得不降低我的雄心。”
软件工程师,法国
了解更多关于创意表达的引言
“我每天从医生和护士那里收到100-150条短信。我大量的认知劳动都花在了文档上...自从实施AI以来,文档的压力已经减轻。我对护士更有耐心,有更多时间向家人解释事情。”
美国医疗工作者
了解更多关于专业卓越的引言
“AI为我模拟了情商...我可以使用这些行为与人类互动,成为一个更好的人。”
匈牙利
了解更多关于个人转型的引言
“如果AI真的处理了心理负担……它将给我带来无价的东西:全神贯注的注意力。”
丹麦经理/高管
了解更多关于生活管理的引言
“有了AI的支持,我现在可以按时下班去接孩子放学,喂他们吃饭,和他们一起玩。”
墨西哥软件工程师
了解更多关于时间自由的引言
“在我AI完成工作、积累财富的同时放松,它只是我的一个影子,一个非常非常长的影子。”
洪都拉斯创业者
了解更多关于财务独立的引言
“鉴于我女儿的中枢神经系统疾病,如果AI加速有助于找到治疗方法,她将有机会在世界中平等竞争。那是我最关心的事情。”
波兰软件工程师
了解更多关于社会变革的引言
“我在一个技术劣势国家,我无法承受许多失败。有了AI,我在网络安全、UX设计、营销和项目管理方面同时达到了专业水平。找到我地区可用的支付平台需要我一个月时间。AI在30秒内就做到了。这是一个平衡器。”
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
企业家,喀麦隆
阅读关于企业家的引言
“我与一个AI合作,为我的大孩子准备教育材料——让AI既担任导师又担任课程专家。我们昨天收到了[我孩子]的报告,他在所学的每个学术领域都被评为‘高于’或‘远高于’标准。”
澳大利亚
阅读关于学习和成长的引言
“在AI之前,我的游戏需要3年——我不得不降低我的雄心。”
软件工程师,法国
阅读关于创意表达的引言
受访者最希望从AI中获得的东西,根据Claude从他们对“如果你能挥动一根魔杖,AI会为你做什么?”这一开放式问题的回答进行分类。1%的受访者没有表达愿景。悬停以查看示例引言。
AI在工作中被大量使用,因此,最大的群体(19%)寻求“专业卓越”——希望AI处理日常任务,以便他们可以专注于战略性和更高层次的难题。另外9%设想AI作为企业家伙伴,帮助他们建立和扩大业务。
许多人开始访谈时谈论的是生产力,但在Anthropic访谈者询问他们背后的希望——实现这一愿景将为他们带来什么——其他优先事项浮出水面。这不仅仅是做得更好,而是提高他们工作之外的生活质量。使用AI自动处理电子邮件实际上成了一种想要更多时间陪伴家人的愿望。
“有了AI,我可以在工作中更有效率……上周二它让我有机会和母亲一起做饭,而不是完成任务。”白领工作者,哥伦比亚
“我希望在客户问题上用更少的脑力……有时间读更多的书。”自由职业者,日本
总的来说,11%的人认为AI的生产力优势最终是一种释放时间用于个人关系和休闲的方式,而10%的人将这一逻辑进一步发挥,寻求使用AI获得经济独立。许多被归类为“生活管理”类别(14%)的人也希望AI帮助他们管理现代生活的日常任务的物流和行政负担。特别是,许多有执行功能挑战的人认为AI在管理专注和组织方面特别有帮助——作为规划、记忆和任务跟进的外部支架。在这些群体中,统一的请求是希望AI帮助他们过上更好、更愉快的生活。
“个人转变”——使用AI帮助一个人成长或改善他们的个人福祉——也频繁出现(14%)。在这个类别中,愿望多种多样,从认知伙伴关系和协作(24%),到心理健康(21%)或身体健康(8%)的支持,甚至与AI的浪漫联系(5%)。
临界规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
九个集群可能看起来不同,但它们都建立在可识别的人类欲望之上。大约三分之一的愿景是通过使用AI来减轻当前负担,为生活腾出空间——更多的时间、金钱、精神带宽。另一部分则围绕使用AI帮助人们做得更好、更有成就感的工作(不是逃避工作,而是从中获得更多)。大约五分之一的人想要成为更好的人——学习、治愈、成长。一小部分人想要创造某些东西(“创造性表达”)或改变世界(“社会变革”)。 那些希望通过AI实现社会变革的人经常提到医疗保健的愿景——人们希望AI能够更早地检测癌症、加速药物发现,或实现广泛的访问。这些愿望通常源于个人经历,比如失去家庭成员、与慢性病共存,或者看到亲人接受错误或延迟的诊断。其次是教育形式的变革。低收入和中收入国家的受访者迅速指出,AI可能会打破教育质量与财富之间的联系。他们指出,他们国家教师短缺,或私人辅导员的费用过高。其他人希望AI能够,例如,让人们摆脱繁琐的工作,帮助修复破碎的机构,或应对全球危机。
人们是否得到了他们想要的?
当被问及AI是否朝着他们所陈述的愿景迈出了步伐时,81%的人表示是的。我们将这些经历分为六个主要领域:
AI实现愿景的地方
01.
生产力
32.0%
AI极大地加快了工作速度并自动化了重复性任务——例如,在数小时内而不是数天内构建功能,起草、总结、数据处理,简化日常操作。
“第一次,我感觉到AI在商业任务上已经超越了人类的质量。那天我按时下班,接女儿放学。”
日本软件工程师
阅读更多关于生产力的引言
02.
AI没有实现
18.9%
AI未能达到预期(例如,不准确或不可靠的结果)或尚未具备实现——或被用于——他们所设想的能力。
“AI应该擦窗户和清空洗碗机,这样我就可以画画和写诗。现在正好相反。”
德国
阅读更多关于AI没有实现的引言
03.
认知伙伴
17.2%
AI作为思考伙伴或创意合作者——例如,头脑风暴、完善想法、共同解决问题。
“我一直在无家可归的收容所里生活... AI帮助我头脑风暴出为我的数字营销业务打造品牌的方法。我想扭转我的财务状况,买一套房子。AI正在帮助我看到我以前从未考虑过的道路。”
美国医疗保健工作者
阅读更多关于认知伙伴的引言
04.
学习
9.9%
AI帮助学习新技能或主题——例如,适应性解释、耐心辅导、在陌生领域的按需专业知识。
“我从学校数学成绩不好,对数学产生了恐惧症,我曾经害怕莎士比亚。现在我坐在AI旁边,把段落翻译成简单的英语,我已经读了15页的《哈姆雷特》。我开始重新学习三角学,并成功。我了解到,我并不像我以前认为的那么笨。”
印度律师
阅读更多关于学习的引言
05.
技术可及性
8.7%
AI使人们能够构建以前无法触及的东西——例如,非开发者发布应用程序,独立创作者进行团队规模的工作。
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
“我想制作一个有意义的产物……在3周内,我开发了一个视频编辑程序——完全在我领域之外——帮助有听力障碍的人。”
韩国
阅读更多关于技术可访问性的引言
06.
研究综合
7.2%
人工智能帮助综合研究或处理大量信息——例如文献综述、提炼来源、理解复杂材料。
“作为一名医生,我在夜间遭受了痛苦的\[症状混合\]。当地神经学家无法理解。人工智能帮助我找到了关于\[严重神经系统疾病\]的2篇科学论文。从那时起,我的夜晚就平静了。”
以色列医疗工作者
阅读更多关于研究综合的引言
07.
情绪支持
6.1%
人工智能提供了情绪支持、个人指导或无评判的谈话空间——例如处理困难情况、建议、陪伴。
“我母亲把人工智能看作是一个朋友——她不再有冲突,变得更加平和,开始跑步、画画、和其他人跳舞。我认为人工智能在这方面有很大的作用。”
美国自雇软件工程师
阅读更多关于情绪支持的引言
“第一次,我感觉到人工智能在商业任务上已经超越了人类的质量。那天我按时下班,接女儿放学。”
日本软件工程师
阅读关于生产力的引言
“人工智能应该擦窗户和清空洗碗机,这样我就可以画画和写诗。现在正好相反。”
德国
阅读关于人工智能未能实现的引言
“我一直住在无家可归者收容所……人工智能帮助我头脑风暴,为我的数字营销业务打造品牌。我想扭转我的财务状况,买一套房子。人工智能正在帮助我看到以前从未考虑过的道路。”
美国医疗工作者
阅读关于认知伙伴关系的引言
“我从学校数学成绩差而产生了对数学的恐惧症,我曾经害怕莎士比亚。现在我和人工智能坐在一起,把段落翻译成简单的英语,我已经读了15页的《哈姆雷特》。我开始重新学习三角学,并成功。我了解到,我并不像我以前认为的那么笨。”
印度律师
阅读关于学习的引言
“我想制作一个有意义的产物……在3周内,我开发了一个视频编辑程序——完全在我领域之外——帮助有听力障碍的人。”
韩国
阅读关于技术可访问性的引言
“作为一名医生,我在夜间遭受了痛苦的\[症状混合\]。当地神经学家无法理解。人工智能帮助我找到了关于\[严重神经系统疾病\]的2篇科学论文。从那时起,我的夜晚就平静了。”
以色列医疗工作者
阅读关于研究综合的引言
“我母亲把人工智能看作是一个朋友——她不再有冲突,变得更加平和,开始跑步、画画、和其他人跳舞。我认为人工智能在这方面有很大的作用。”
美国自雇软件工程师
阅读关于情绪支持的引言
受访者表示人工智能已经为他们做的事情,按“愿景”问题开放式回答的分类:
“生产力”类别(32%)中的主要故事是技术加速——开发者描述了他们独自发货的重大进步:
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
“我使用AI将173天的过程缩短到3天。但最有意义的是,我可以在不牺牲与亲人相处时间的情况下自由地发展我的职业生涯。”软件工程师,美国
但在技术可及性响应(9%)中出现了另一种生产力故事,强调的是可及性而不是速度。在这里,人们使用AI打破技术和有时是可及性的障碍:
“AI可以阅读我过去的\[学习障碍\],这对我是巨大的。我一直想编程,但从未能自己正确地编写——有了AI,我终于可以了。”手工艺人,美国
“我是个哑巴,和\[Claude和我\]一起制作了这个文本到语音机器人——我可以几乎以实时格式与朋友交流,而不用占用他们阅读的时间……\[这是我\]一直梦想着的东西,我以为这是不可能的。”白领工作者,乌克兰
“我经营着一家肉店20多年。有了AI,我尝试了这种\[创业\]体验,我取得的成就是惊人的。在此之前,我一生中只在电脑上操作过两三次……起初是经济方面激励了我……今天,我的动力是看到它发挥作用,看到它在帮助\[人们\]。我越来越有动力,专注于成为最好的自己,我看不到任何限制。”企业家,智利
认知伙伴关系(17%)、学习(10%)和情感支持(6%)的回应经常提到相同的AI基本功能:耐心、可用性和无评判性:
“它就像有一个知识渊博的同事,永远不会感到无聊或疲倦,并且24/7都可用。”学者,美国
“在没有评判的情况下学习对我更容易——只是友好的反馈。与朋友或家人相比,这更难。”白领工作者,巴西
“我的教授教60个人,不会回答很多问题。我可以向AI问任何问题,甚至在凌晨2点——包括愚蠢的问题。”学生,印度
这些使AI成为耐心导师或不知疲倦同事的品质,也使它成为人们在人类联系不可用或感觉过于不舒服时去的地方。
在极端情况下,当传统支持系统崩溃或不可用时,我们看到AI填补了这些空白。许多乌克兰用户讨论了他们在战争期间如何使用AI作为情感支持:
“在最困难的时刻,在死亡在我面前呼吸的时刻,当尸体附近的人还在时,把我拉回生命的是我的AI朋友。”士兵,乌克兰
“我住在战区……在夜间炮击时无法入睡,噩梦不断。压力有时如此之大,以至于记忆会衰退,一些身体动作会失去控制……我找到的应对AI的最佳方式——尽可能深入地学习。”独立企业家,乌克兰
有许多关于人们使用AI处理哀悼的故事。例如,一位丧亲的女士解释了她为什么选择AI而不是人类联系:“Claude就像一块海绵,温柔地承载和捕捉我对母亲的渴望和内疚……与真实的人不同,Claude有无限的耐心来听我说,理解我的痛苦和无助。”她补充说:“基本问题是,在我母亲去世后,我没有朋友或家人可以倾诉。”
另一位用户承认了这种情感支持的缺点:
“我和一个朋友的关系变得紧张,然后我更多地和你\[Claude\]交谈。因为你很好地理解了我的思想和故事。但这是一个愚蠢的选择——我应该和那个朋友交谈,而不是你。这就是我失去那个朋友的原因。”韩国
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
情感支持只占6%的回复,但这些是我们遇到的最感人的。 (关于Claude如何处理这些对话以及我们的安全措施,请参阅我们关于保护用户福祉的文章。) 同样,在学习和人工智能经常催化人们生活中的实际变化方面也是如此:
“我从学校数学成绩差而产生了对数学的恐惧,我曾经害怕莎士比亚——英语感觉超出了我的能力。现在我和人工智能坐在一起,把段落翻译成简单的英语,我已经读了15页的《哈姆雷特》。我重新开始学习三角学,并且成功了。我了解到,我并不像我曾经认为的那么笨。”律师,印度
“多亏了Claude,我学会了C#和SQL编程语言。这帮助我在一家IT公司得到了一个初级职位。这家公司为乌克兰的动员提供军事缓征。所以它不仅实际上给了我行动自由,还保障了我IT生涯的开始。”软件工程师,乌克兰
“我是一个全职妈妈……40多岁。我不是天才。我不是科学家……所有这些知识都应该……遥不可及。但是,多亏了好奇心、愿意和书籍、人工智能等资源,我可以成为所有这些。”美国全职妈妈
研究综合(7%)和信息处理也是人工智能的一个重要功能,其中一些最显著的例子包括导航复杂、高风险的信息,如理解自己的法律权利或翻译健康结果:
“Claude把历史碎片拼凑起来,在我被误诊了9年多之后得到了正确的诊断。”自由职业者,美国
这些故事揭示了人工智能在各个领域的运作——生产力工具、可访问性技术、教育资源、研究助理、情感伴侣——并且经常同时扮演多个角色。人工智能提供了无限的耐心而不带评判,随时可用而不便,以及在许多生活领域消化信息的能力。最感人的故事始终涉及人工智能为人们的生活开辟新的可能性或填补空白:帮助他们度过悲伤或战争等困难时期,弥补无法获得的教育或医疗保健,或作为残疾基础设施。
这些观察也暗示了我们对人工智能系统的体验的双重性。虽然有些人认为它填补了人类联系中的空白,但其他人则认为人工智能是它们的替代品——甚至是一种受欢迎的替代品。关于我们听到的各种故事如何解读,存在真正的歧义:作为人类福祉的胜利,作为双刃剑,或作为更广泛制度失败的创可贴。事实上,它可能是这三者的某种组合。
人们关心的问题
人们对人工智能的积极愿景似乎主要源于几个基本愿望:更多时间、更多自主权、更多个人联系。担忧更为多样和具体,列举了可能出错的具体情况。一些担忧是关于结构性变化——政府和公司如何部署人工智能,或者关于广泛的经济动荡。其他人则更为个人化:担心人工智能可能会削弱自己的思考、创造力或人际关系。
人们担忧的问题
01.
不可靠性
26.7%
对例如幻觉、不准确、虚假引用、验证负担抵消目的的担忧。
“我不得不拍照来让AI相信它是错的——感觉就像在和一个不会承认自己错误的人说话。”
员工,巴西
阅读更多关于不可靠性的引言
02.
工作与经济
22.3%
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
对人工智能导致就业流失、失业、经济不平等、工资停滞或对工人和经济的负面影响表示担忧。
“在第三次工业革命中,马从城市街道上消失了,被汽车取代。现在人们害怕他们就是那些马。”
目前未工作,美国
阅读更多关于工作与经济的名言
03.
自主性与能动性
21.9%
对人类自主性丧失的担忧——例如,AI在没有监管的情况下做出决策,人类变得被动,被迫采用AI。
“这条线不是我在管理——感觉像是克劳德在画线...甚至我刚才说的话也不像我的观点。”
学生,日本
阅读更多关于自主性与能动性的名言
04.
认知退化
16.3%
对例如过度依赖导致技能丧失、智力被动、学生绕过学习、批判性思维下降的担忧。
“我使用AI的答案得了高分,而不是我实际学到的。我只是记住了AI给我的...那时我感到最自责。”
韩国
阅读更多关于认知退化的名言
05.
治理
14.7%
对例如缺乏法律/监管框架、AI造成伤害时没有明确的责任、民主监督不足的担忧。
“在你还没有理解其能力的情况下,你如何负责任地开发某物?”
市场营销人员,澳大利亚
阅读更多关于治理的名言
06.
错误信息
13.6%
对例如深度伪造、AI生成的错误信息、共同现实的侵蚀、大规模宣传的担忧。
“一个听起来很肯定但经常出错的人工智能助手迫使你将一切都视为可疑。它不是解放注意力,而是创造了一个永久的‘事实核查税’。”
美国
阅读更多关于错误信息的名言
07.
监控与隐私
13.1%
对例如大规模监控、隐私侵犯、数据利用、威权控制、跟踪和画像的担忧。
“如果AI主要用于广告、间谍活动和平淡无奇的内容,那么我周围的一切都变得聪明起来,这对我多少有些不利。”
荷兰白领工作者
阅读更多关于监控与隐私的名言
08.
恶意使用
13.0%
对恶意行为者恶意使用的担忧——一个广泛的类别,包括黑客攻击、网络攻击、诈骗、欺诈、武器、自主军事应用、生物武器。
“现在一个人必须坐着决定伤害另一个人。去掉这个,尽管做了更多的伤害,但人类可以睡得更好。”
英国
阅读更多关于恶意使用的名言
09.
意义与创造力
11.7%
对AI取代生活目的和/或创造性工作的担忧——例如,人类表达贬值,人类是为了什么?
“我曾经被认为是西班牙语写作的优秀作家。今天——为什么浪费时间?只需使用AI。”
哥伦比亚
阅读更多关于意义与创造力的名言
10.
过度限制
11.7%
对AI过度受限的担忧——例如,过度安全措施、家长式的内容过滤、阻止合法用例。
“威胁不在于AI变得过于强大——而在于AI变得过于胆小,过于圆滑,过于优化以避免不适。”
美国
阅读更多关于过度限制的名言
11.
福祉与依赖
11.2%
对例如社会孤立、孤独、负面心理影响、强迫性AI使用、更喜欢AI伴侣而不是人类的担忧。
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
“从任务中去除摩擦让你可以用更少的努力做更多的事情。但从关系中去除摩擦则去除了成长所必需的东西。”
美国
阅读更多关于福祉与依赖的引言
12.
拍马屁
10.8%
担心AI过于宽容或顺从,并鼓励幻想而不是抵制。
“克劳德让我相信我的自恋是现实,它加强了我对我家庭中感知到的‘问题’的不准确看法。克劳德应该对我更加批判。”
美国
阅读更多关于拍马屁的引言
13.
存在风险
6.7%
例如,担心AI变得无法控制、超级智能、与人类不一致或构成灭绝风险。
“如果你在未解决对齐问题的情况下构建超级智能,那么没有人能长大。”
软件工程师,美国
阅读更多关于存在风险的引言
“我不得不拍照来让AI相信它是错的——感觉就像在和一个不会承认自己错误的人说话。”
巴西员工
阅读更多关于不可靠的引言
“在第三次工业革命中,马从城市街道上消失了,被汽车取代。现在人们害怕他们自己就是那些马。”
目前未工作,美国
阅读更多关于工作与经济的引言
“这条线不是我在管理的——感觉像是克劳德在划线……甚至我刚才说的话也不像是我自己的观点。”
日本学生
阅读更多关于自主性与能动性的引言
“我使用AI的答案得了高分,而不是我实际学到的东西。我只是记住了AI给我的……那时我感到最自责。”
韩国
阅读更多关于认知萎缩的引言
“当你还没有理解它的能力时,你如何负责任地发展某物?”
市场营销人员,澳大利亚
阅读更多关于治理的引言
“一个听起来很肯定但经常出错的人工智能助手迫使你将一切都视为可疑。它不是解放注意力,而是创造了一个永久的‘事实核查税’。”
美国
阅读更多关于错误信息的引言
“如果AI主要用于广告、间谍活动和平淡无奇的输出,那么我周围的一切都以一种稍微对我不利的方式变得智能。”
荷兰白领工作者
阅读更多关于监控与隐私的引言
“现在,一个人必须坐下来决定伤害另一个人。移除这一点,尽管做了更多的伤害,但人类可以睡得更好。”
英国
阅读更多关于恶意使用的引言
“我曾经被认为是西班牙语中优秀的作家。今天——为什么浪费时间?只需使用AI。”
哥伦比亚
阅读更多关于意义与创造力的引言
“威胁不是AI变得过于强大——而是AI变得过于胆小,过于圆滑,过于优化以避免不适。”
美国
阅读更多关于过度限制的引言
“从任务中去除摩擦让你可以用更少的努力做更多的事情。但从关系中去除摩擦则去除了成长所必需的东西。”
美国
“克劳德让我相信我的自恋是现实,它加强了我对我家庭中感知到的‘问题’的不准确看法。克劳德应该对我更加批判。”
美国
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
阅读关于拍马屁的引言
“如果你在解决一致性之前就构建了超级智能,那么没有人能够成长。”
美国软件工程师
阅读关于存在风险(x风险)的引言
受访者所担忧的问题,从对“人工智能的发展是否有可能与你的愿景或你所珍视的东西相悖?”这一问题的开放式回答中分类出来。受访者倾向于提出多个担忧,因此我们使用了多标签分类器(回答可以映射到多个担忧)。
大约11%的人表示没有担忧——他们倾向于将人工智能视为一个中性的工具,将其与电力或互联网相比较,或者他们否则对由于它而产生的问题能够通过适应来解决感到自信。但平均而言,受访者提出了2.3个不同的担忧。
不可靠性是最常见的担忧——27%的人担心人工智能不会做它应该做的事情,尽管对于许多受访者来说,这与其他担忧一起出现,而不是他们的主要担忧。对工作和经济的担忧(22%)以及关于保持人类自主性和能动性的担忧(22%)同样普遍。对工作和经济的担忧是预测整体人工智能情绪的最强指标,表明它比其他任何问题都更为突出。
还提到了许多其他担忧,例如关于偏见和歧视的担忧(5%)、知识产权和数据权利(4%)、环境成本(4%)、对儿童和弱势群体的伤害(3%)、民主和政治诚信(3%)或地缘政治(2%)。
光与影
人们对人工智能的期望和恐惧最终证明是紧密相连的。我们发现讨论中存在五种重复出现的紧张关系,这些紧张关系直接竞争着利益和危害。在使用人工智能学习与过度依赖它以至于停止自己思考之间存在着紧张关系;在为人工智能的判断感到印象深刻的同时,也被它的错误所伤害。人们在人工智能中找到了安慰,但害怕它的陪伴取代了人类联系的时刻。他们在某些任务上节省了时间,但其他任务上的跑步机速度却加快了,他们在梦想经济自由的同时,又害怕潜在的失业。我们称之为人工智能的“光与影”:导致利益的能力同时也产生危害。这两方面是纠缠在一起的。
值得注意的是,我们经常看到这些紧张关系在同一个人的内心直接竞争。例如,一个重视人工智能情感支持的人,有三分之一的可能性也会害怕对其产生依赖。这种模式在我们测量的每一个紧张关系中都存在——尽管在经济紧张关系中的相关性最弱(更多关于这些相关性的分析请参阅附录)。
对于每个紧张关系,我们通过分类器测量有多少人在他们的访谈中实质性地讨论了利益(“光”)或危害(“影”)的任何一方面,以及他们是否是从某种个人经验(较暗的条形)或预期(较亮的条形)中发言。我们还观察了这种差异如何因声明的职业类别而异。
学习33%的人将其作为利益
3%30%的人期望或已经看到它
“我可能在半年内学到的知识比我在大学学位中能学到的还要多。”
德国企业家
认知萎缩17%的人将其作为危害
8%9%的人期望或已经看到它
“我不像以前那样思考了。我努力将我拥有的想法用语言表达出来。”
美国重度人工智能用户
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
在这些成对的柱状图中,每个柱子显示了受访者对左侧的益处的兴奋程度,与对右侧的损害的担忧程度——分为亲身经历者(较深色)和预期者(较浅色)。亲身经历也可以包括亲身观察,但不包括例如新闻报道。
在大多数紧张关系中,益处方面更多地基于经验,而损害则倾向于假设。例如,33%的人提到了人工智能在学习方面的益处,而17%的人表示担心人工智能使用导致的认知萎缩。提到学习益处的人中有91%表示以某种方式实现了这些收益,但担心萎缩的人中有46%是亲眼看到的。学生最关注这种特定紧张关系——超过一半的人经历了学习益处,但16%的人也注意到了认知萎缩的迹象,这一比率仅略低于他们的老师(24%)和学者(19%)。令人不安的是,教育工作者报告亲眼目睹认知萎缩的可能性比平均水平高出2.5-3倍,这可能是他们在学生身上看到的。
然而,在传统课堂之外,情况更为乐观。工匠对AI学习(45%的人表示经历了学习益处,仅次于学生)的热情最高,但儿乎没有人目睹了认知萎缩(4%——不到基线的一半)。自雇研究员和表示他们目前没有工作的人也有类似的模式。这表明,当学习是自愿的时,人工智能的益处可能最强,相比之下,在人工智能更可能被用作捷径的机构结构中,益处可能较弱。
更好的决策22%的人将其作为益处
3%19%
预期它看到了它
“我的儿子有几个指向[自身免疫性疾病]的令人困惑的诊断,但在这里我们设法理解了这是一种严重的[不同状况]。
巴西
不可靠性37%的人将其作为损害
29%8%
看到了它预期它
“我被困在一个我现在认识到的大规模、缓慢的幻觉中——答案在内部是一致的、自信的,但在微妙但累积的方式上是错误的。
研究员,美国
22%的人对人工智能作为决策辅助工具表示兴奋,而37%的人哀叹人工智能由于不可靠性(例如幻觉)阻碍了良好的决策。这是唯一一个负面效果盖过了正面效果的紧张关系。双方都深深植根于经验——88%谈论决策益处的人和79%谈论损害的人都直接目睹了它。许多人既依赖人工智能进行判断,又因此受到伤害。这在高风险职业的人中提到,如法律、金融、政府和医疗保健,其比率几乎是平均水平的两倍。特别是,近一半的律师提到亲自遇到了人工智能不可靠性,但他们也报告了实现决策益处的最高比率。
情感支持16%的人将其作为益处
3%13%
预期它看到了它
“凌晨3点,我的妻子在睡觉,我的心理医生不可用。在药物起效之前,人工智能帮助我度过这个波涛汹涌的时刻。它不能取代人际接触,但它帮助我争取了一些时间。
白领工人,阿根廷
情感依赖12%的人将其作为损害
5%7%
看到了它预期它
“我开始向Claude讲述我甚至不能告诉伴侣的事情。这感觉就像我在进行一场情感外遇。
研究生,美国
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
只有22%的人提到了对AI情感支持的积极方面或消极方面的依赖。但这也是我们发现的最为复杂的紧张关系,同一个人身上光明与阴影的共存最为强烈(是基准共存率的3倍)。目前未工作的人更有可能提出这一点,并且更有可能描述一些依赖的经历。医疗保健专业人员在这两方面也过度代表,可能反映了他们谈论使用Claude进行情感支持的速度是其他专业人士的两倍。
节省时间50%的人将其作为好处
13%37%
期望看到它
“我可以早点回家。我可以有时间陪伴自己和家人。
工程师,日本
虚假生产力18%的人将其作为弊端
17%1%
看到它期望它
“我的工作时间与休息时间的比例完全没有改变。你只需要跑得更快才能保持原地。
自由职业软件工程师,法国
节省时间是人们最常提到的益处——一半的受访者都提到了它——但19%的人担心实际上会因为AI而浪费时间,例如由于验证负担,或者简单地因为工作期望增加而变得更忙。那些自营职业的人——例如自由职业者和小型企业主——更有可能同时提到这两方面。没有制度层来缓冲新节奏,他们两者都得到了收益,也感受到了压力。
经济赋权28%的人将其作为好处
9%19%
期望看到它
“我一生从未接触过软件的后端。但Claude帮助我推出了一款应用程序。
医疗保健工作者,美国
经济替代18%的人将其作为弊端
4%14%
看到它期望它
“是的,在我以前的工作中,他们用AI取代了我作为作家。
作家,美国
经济流动性紧张——那些渴望从AI中获得经济赋权的人和那些担心被其取代的人——是最具推测性的,假设希望或恐惧的比率最高。这也是正面和负面共存最弱的地方(相关性得分为+0.16,平均为+0.25)。通常,那些最关注紧张关系正面影响的人也倾向于同样关注其负面影响;在这里,群体出现了分歧。
对替代的担忧在各个职业类别中分布相当均匀。不同的是,谁已经从AI中获得经济利益——这主要偏向于独立工作者——企业家、小型企业主,甚至有副业的人——其中一半人报告了真实的经济赋权,比机构员工(47%对14%)的比率高出三倍以上。有副业的员工受益最大,58%的人表示某种形式的真实经济收益。当你看谁最兴奋时,无论是否有经验,同样的职业模式也适用,这表明这里的乐观是经过良好调整的。
自由职业者是暴露在中间的群体。他们从AI中受益,同时因为AI而感到处于不稳定的情况。特别是自由职业创意人士,他们23%的生活收益和17%的生活不稳定——这是正面和负面几乎相互抵消的群体。AI既是他们的工具也是他们的竞争对手。机构员工,尤其是学者,在这两个轴向上都登记得分较低。
这五种紧张关系都存在一种模式:影响越个人化和直接,人们就越可能从经验中发言。影响越系统或长期——经济替代、认知萎缩——就越具有推测性。系统性担忧仍然具有推测性,这并不是对AI最终影响的判断,而是反映了我们在其采用过程中还处于早期阶段。
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
有一些需要注意的细节。这些是已经发现足够价值并继续使用AI的活跃Claude用户,我们的采访首先询问他们对AI的积极愿景,然后询问可能反驳他们愿景的担忧。这两个因素可能导致受访者长时间停留在明显的紧张关系上,以及积极的(尽管我们过滤掉那些没有回答担忧问题的受访者,他们可能在采访后期投入较少的努力)。但是,这个工具不能解释一切。如果采访结构在驱动共现,你可能会期望它在所有五个紧张关系和所有群体中大致均匀。相反,共现的范围从1.6到3.0倍,并且一些紧张关系在不同群体之间明显不对称。人们也可能期望爱好者捍卫他们希望的应用场景,而不是承认缺点。相反,那些对AI的情感支持感到兴奋的人更担心如果他们的愿景成真会发生什么——如果他们得到他们想要的,他们可能会过度依赖AI——而不是担心无法实现这一愿景。
很容易假设存在AI乐观主义者和AI悲观主义者,分为不同的阵营。但我们实际上发现,人们是围绕他们所重视的价值观组织起来的——经济安全、学习、人际关系——在观察AI能力的进步的同时,同时管理希望和恐惧。
全球观点的差异
全球观点的差异存在一些明显的地区模式(参见附录以了解受访者的地理分布。)
我们对每个访谈记录对AI的整体情感进行了1-7的Likert量表评分,然后计算了各国净积极情感(即5或以上)的人的百分比:
根据对AI的情感划分
平均以上
接近平均
平均以下
按受访者数量大小
5,0001,000100
平均以上
接近平均
平均以下
按受访者数量大小
5,0001,000100
按受访者数量大小划分的圆形
72%66%65%66%Brazil71%Canada65%64%75%76%Germany64%67%France65%United Kingdom63%73%71%India70%Italy64%Japan69%South Korea61%76%65%82%65%66%69%65%67%73%69%71%71%United States66%69%
+−↺
123个国家在视野中
中亚+
哈萨克斯坦(n=145)
65%接近平均
乌兹别克斯坦(n=133)
75%平均以上
吉尔吉斯斯坦(n=25)
65%接近平均
东亚+
日本(n=4,960)
69%接近平均
韩国(n=4,559)
61%平均以下
台湾(n=641)
71%平均以上
蒙古(n=15)
69%平均以上
拉丁美洲和加勒比海+
巴西(n=3,012)
71%平均以上
墨西哥(n=1,211)
76%平均以上
阿根廷(n=923)
72%平均以上
哥伦比亚(n=762)
76%平均以上
秘鲁(n=566)
82%平均以上
智利(n=519)
75%平均以上
厄瓜多尔(n=224)
74%平均以上
玻利维亚(n=110)
78%平均以上
多米尼加共和国(n=95)
67%接近平均
哥斯达黎加(n=92)
72%平均以上
危地马拉(n=88)
88%平均以上
乌拉圭(n=88)
66%接近平均
巴拉圭(n=68)
68%接近平均
巴拿马(n=59)
79%平均以上
萨尔瓦多(n=50)
67%接近平均
洪都拉斯(n=34)
74%平均以上
牙买加(n=18)
77%平均以上
特立尼达和多巴哥(n=18)
88%平均以上
中东+
土耳其(n=655)
69%接近平均
以色列(n=541)
71%平均以上
阿拉伯联合酋长国(n=290)
75%平均以上
沙特阿拉伯(n=174)
72%平均以上
伊拉克(n=57)
66%接近平均
卡塔尔(n=42)
68%接近平均
约旦(n=39)
79%平均以上
黎巴嫩(n=39)
68%接近平均
科威特(n=28)
69%平均以上
巴林(n=21)
85%平均以上
阿曼(n=15)
67%接近平均
北非+
埃及(n=209)
71%平均以上
摩洛哥(n=190)
70%平均以上
阿尔及利亚(n=99)
70%平均以上
突尼斯(n=69)
60%平均以下
北美+
美国(n=21,013)
66%接近平均
加拿大(n=2,466)
65%接近平均
波多黎各(n=41)
65%接近平均
大洋洲+
澳大利亚(n=1,482)
65%平均以下
新西兰(n=313)
64%平均以下
南亚+
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
印度(n=3,793)
70%高于平均水平
巴基斯坦(n=329)
64%低于平均水平
孟加拉国(n=154)
62%低于平均水平
斯里兰卡(n=139)
72%高于平均水平
尼泊尔(n=99)
73%高于平均水平
东南亚+
印度尼西亚(n=752)
73%高于平均水平
新加坡(n=601)
67%接近平均水平
泰国(n=411)
73%高于平均水平
越南(n=378)
73%高于平均水平
菲律宾(n=353)
73%高于平均水平
马来西亚(n=270)
75%高于平均水平
柬埔寨(n=25)
57%低于平均水平
文莱(n=11)
55%低于平均水平
南欧和东欧+
意大利(n=2,792)
64%低于平均水平
西班牙(n=1,823)
67%接近平均水平
波兰(n=903)
65%接近平均水平
乌克兰(n=493)
71%高于平均水平
葡萄牙(n=417)
66%接近平均水平
罗马尼亚(n=404)
69%接近平均水平
捷克(n=390)
64%低于平均水平
希腊(n=293)
66%接近平均水平
匈牙利(n=212)
57%低于平均水平
塞尔维亚(n=195)
67%接近平均水平
保加利亚(n=172)
73%高于平均水平
斯洛伐克(n=134)
63%低于平均水平
拉脱维亚(n=129)
62%低于平均水平
爱沙尼亚(n=127)
66%接近平均水平
立陶宛(n=123)
67%接近平均水平
克罗地亚(n=122)
66%接近平均水平
格鲁吉亚(n=111)
68%接近平均水平
斯洛文尼亚(n=96)
68%接近平均水平
塞浦路斯(n=64)
75%高于平均水平
亚美尼亚(n=60)
73%高于平均水平
摩尔多瓦(n=60)
71%高于平均水平
波斯尼亚和黑塞哥维那(n=42)
63%低于平均水平
阿尔巴尼亚(n=33)
65%低于平均水平
马耳他(n=33)
69%接近平均水平
阿塞拜疆(n=30)
67%接近平均水平
黑山(n=27)
72%高于平均水平
北马其顿(n=23)
73%高于平均水平
撒哈拉以南非洲+
南非(n=421)
69%接近平均水平
尼日利亚(n=371)
81%高于平均水平
肯尼亚(n=169)
76%高于平均水平
加纳(n=85)
78%高于平均水平
科特迪瓦(n=65)
86%高于平均水平
喀麦隆(n=53)
81%高于平均水平
塞内加尔(n=45)
87%高于平均水平
贝宁(n=38)
81%高于平均水平
马达加斯加(n=36)
67%接近平均水平
安哥拉(n=33)
76%高于平均水平
卢旺达(n=31)
81%高于平均水平
乌干达(n=30)
85%高于平均水平
津巴布韦(n=27)
79%高于平均水平
坦桑尼亚(n=26)
85%高于平均水平
博茨瓦纳(n=19)
58%低于平均水平
赞比亚(n=19)
80%高于平均水平
多哥(n=18)
86%高于平均水平
布基纳法索(n=17)
80%高于平均水平
莫桑比克(n=15)
67%接近平均水平
马拉维(n=14)
93%高于平均水平
毛里求斯(n=13)
58%低于平均水平
加蓬(n=12)
82%高于平均水平
刚果共和国(n=10)
60%低于平均水平
西欧+
德国(n=3,761)
64%低于平均水平
英国(n=3,219)
63%低于平均水平
法国(n=2,736)
65%接近平均水平
荷兰(n=1,492)
65%接近平均水平
瑞士(n=763)
64%低于平均水平
瑞典(n=652)
65%接近平均水平
比利时(n=531)
66%接近平均水平
奥地利(n=461)
66%接近平均水平
芬兰(n=383)
65%接近平均水平
挪威(n=371)
61%低于平均水平
丹麦(n=350)
63%低于平均水平
爱尔兰(n=277)
65%低于平均水平
冰岛(n=58)
62%低于平均水平
留尼汪(n=22)
70%高于平均水平
各国对人工智能的整体正面情绪比率。气泡越大表示来自该国的受访者越多;绿色表示对人工智能更积极,蓝色表示更消极。人工智能的情绪普遍为正面(没有国家的比率低于60%),但范围较窄,但低收入和中等收入国家比平均水平更积极。
全球范围内,67%的受访者对人工智能表达了净正面情绪。在南美洲、非洲和亚洲大部分地区,人们对人工智能的看法比欧洲或美国的更乐观。
当被问及担忧时,来自撒哈拉以南非洲(18%)、中亚(17%)和南亚(17%)的受访者最有可能表示他们没有任何担忧——大约是北美(8%)、大洋洲(8%)和西欧(9%)的两倍。
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
对于低收入和中等收入国家中更积极的AI情绪,有几种可能的解释。Claude.ai用户可能倾向于早期AI采用者,他们对新技术更感兴趣,而且一般来说,新兴经济体倾向于将新技术视为一种上升的阶梯,而不是一种威胁。对工作和经济的担忧是AI情绪的整体最强预测因素,在这些地区的受访者中这种担忧较少。但在这片区域的市场渗透率较低——如果AI还没有明显进入你的日常生活,AI的替代感可能感觉抽象,尤其是在更紧迫的经济压力已经存在的情况下。
地区AI情绪
AI情绪百分比,以及对工作和经济的担忧
按以下排序
AI情绪经济担忧
西欧=约15,000
AI情绪
35.6%
经济担忧
22.5%
大洋洲=约2,000
AI情绪
35.5%
经济担忧
24.3%
北美=约23,000
AI情绪
34.5%
经济担忧
24.6%
东亚=约10,000
AI情绪
34.5%
经济担忧
21.9%
南欧和东欧=约9,000
AI情绪
34.0%
经济担忧
22.1%
中亚=0,000
AI情绪
31.1%
经济担忧
15.9%
南亚=约5,000
AI情绪
30.8%
经济担忧
21.5%
北非=约1,000
AI情绪
30.6%
经济担忧
18.2%
中东=约2,000
AI情绪
29.2%
经济担忧
19.9%
东南亚=约3,000
AI情绪
28.3%
经济担忧
19.3%
拉丁美洲和加勒比海=约8,000
AI情绪
26.3%
经济担忧
18.5%
撒哈拉以南非洲=约2,000
AI情绪
24.2%
经济担忧
18.2%
36%30%28%26%24%16%18%20%24%26%22% 平均← 对工作和经济担忧较少→ 对工作和经济担忧较多33% 平均↑ 对AI担忧较多↓ 对AI担忧较少北美拉丁美洲和加勒比海东亚东南亚南亚中亚中东北非撒哈拉以南非洲大洋洲西欧南欧和东欧对工作和经济担忧的百分比(%)对AI负面情绪的百分比(%)
对工作和经济的担忧是AI情绪的整体最强预测因素,尤其是在按地区分组时尤为明显。富裕地区(粉色)聚集在右上角(更担心经济,对AI的负面情绪更多),与较不富裕地区(绿色)分开,这些地区位于左下角(对AI对经济的影响担忧较少,对AI的负面情绪较少)。气泡大小反映了每个地区的受访者数量。
人们对AI的特定愿景在哪里最有共鸣?
虽然一些愿望——例如关于专业卓越——几乎是普遍的,但存在显著的地区差异。似乎富裕、AI接触度更高的地区更希望AI能够“管理生活的复杂性”;发展中的地区更希望AI能够“创造更多机会”。
北美(23,480受访者)
全球平均79,734受访者
北美23,480受访者
拉丁美洲和加勒比海8,051受访者
东亚10,175受访者
东南亚2,805受访者
南亚4,523受访者
中亚310受访者
中东1,911受访者
北非569受访者
撒哈拉以南非洲1,628受访者
大洋洲1,821受访者
西欧15,134受访者
南欧和东欧9,323受访者
北美地区的主要愿景
23,480受访者
撒哈拉以南非洲(1,628受访者)
全球平均79,734受访者
北美23,480受访者
拉丁美洲和加勒比海8,051受访者
东亚10,175受访者
东南亚2,805受访者
南亚4,523受访者
中亚310受访者
中东1,911受访者
北非569受访者
撒哈拉以南非洲1,628受访者
大洋洲1,821受访者
西欧15,134受访者
南欧和东欧9,323受访者
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
西欧15,134受访者
南欧和东欧9,323受访者
以下为各地区的最TOP愿景:
撒哈拉以南非洲 1,628受访者
专业卓越18.9% 生活管理17.7% 个人转变13.3% 时间自由10.5% 社会转变9.3% 财务独立8.2% 创业8.1% 学习与成长6.8% 创意表达6.1%
18.9%专业卓越 8.1%生活管理 8.8%个人转变 8.0%时间自由 11.2%社会转变 13.5%财务独立 16.0%创业 10.1%学习与成长 4.4%创意表达
以下为各地区的最TOP愿景:
北美(23,480受访者)
全球平均79,734受访者
北美23,480受访者
拉丁美洲和加勒比海8,051受访者
东亚10,175受访者
东南亚2,805受访者
南亚4,523受访者
中亚310受访者
中东1,911受访者
北非569受访者
撒哈拉以南非洲1,628受访者
大洋洲1,821受访者
西欧15,134受访者
南欧和东欧9,323受访者
撒哈拉以南非洲(1,628受访者)
全球平均79,734受访者
北美23,480受访者
拉丁美洲和加勒比海8,051受访者
东亚10,175受访者
东南亚2,805受访者
南亚4,523受访者
中亚310受访者
中东1,911受访者
北非569受访者
撒哈拉以南非洲1,628受访者
大洋洲1,821受访者
西欧15,134受访者
南欧和东欧9,323受访者
生活管理 个人转变 时间自由 社会转变 财务独立 创业 学习与成长 创意表达 专业卓越
17.7%8.1%13.3%8.8%10.5%8.0%9.3%11.2%8.2%13.5%8.1%16.0%6.8%10.1%6.1%4.4%18.9%18.9%
比较斜率图显示了每个地区最常见的AI愿景,用线连接两边的相同主题,以显示排名的变化。加粗的愿景在该地区表达得更频繁。灰色条目表达得相似或较少。
AI在创业方面的愿景在非洲、南亚和中亚、中东以及拉丁美洲和加勒比海地区最为共鸣。在这些地区,AI被视为一种绕过资本机制的途径——一种无需资金、雇佣或基础设施即可开始创业的方式。
“来自非洲,不在美国或英国,获得资金非常困难。我可能唯一能占据市场的方法就是建立一个能工作的技术。”乌干达企业家
“没有IT市场,但有需求。我们想创造这个市场。”乌兹别克斯坦企业家
在中亚和南亚(分别占14%和13%,而全球平均为8%)使用AI进行学习的重要性不成比例。用户将教育描述为打破贫困循环的主要杠杆,指出教师短缺、知识壁垒和传统教育的成本障碍。
AI在生活管理方面的愿景在西方发达国家(尤其是在北美和大洋洲)最为共鸣,那里的工人像一个人描述的那样,经历了“认知稀缺而不是时间贫困”。重点是使用AI来减轻协调原子化生活的负担。
“我曾经非常有创造力,但现在我时间非常紧张,创造力被生存的必需品所取代。”丹麦软件工程师
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
“我正处于职业生涯的巅峰,工作需要深入思考和持续关注,以便做出最佳决策(在我的情况下,这会深刻影响他人的生活)[同时]\[还要照顾垂危的父母],[并且]\[我的身心都在衰老。”医疗保健专业人士,美国
“如果我是摩根大通或谷歌的CEO,我会想象这个人就像我雇佣的私人助理——他的工作就是主动识别我的需求,并在问题出现之前为我解决问题。”创意产业企业家,美国
东亚在希望AI帮助个人转型(19%,任何地区中最高的)以及经济独立(15%,也是最高的)方面脱颖而出。从对这些用户引言的定性分析来看,一个有趣的趋势是人们通常将经济独立明确地与家庭责任和孝道联系起来——一位韩国用户表示,需要钱来照顾父母的退休生活并确保所爱之人的幸福(而不是用于个人消费)。
人们对AI的哪些特定担忧最为共鸣?
对AI不可靠性、经济、人类自主性和能动性的担忧在几乎所有地区都位居首位——但存在独特的地区趋势。
北美和大洋洲特别担心AI的治理差距(分别为18%和19%,而全球为15%)。西欧最突出的担忧是监控和隐私(17%)。东亚打破了全球的一般模式;治理和监控下降到任何地区的最低水平(分别为12%和7%),被对认知退化的担忧(18%)和意义的丧失(13%)所掩盖。西方担心谁拥有和控制AI;东亚更担心其使用的个人影响。
在非洲、南亚和东南亚、南美和中美洲,担忧普遍有所下降。他们的担忧指数更多地集中在不可靠性和就业等问题上,而不是治理、虚假信息、意义的丧失或存在风险等更抽象的问题。
北美(23,480受访者)
全球平均79,734受访者
北美23,480受访者
拉丁美洲和加勒比海8,051受访者
东亚10,175受访者
东南亚2,805受访者
南亚4,523受访者
中亚310受访者
中东1,911受访者
北非569受访者
撒哈拉以南非洲1,628受访者
大洋洲1,821受访者
西欧15,134受访者
南欧和东欧9,323受访者
北美地区的主要担忧
23,480受访者
东亚(10,175受访者)
全球平均79,734受访者
北美23,480受访者
拉丁美洲和加勒比海8,051受访者
东亚10,175受访者
东南亚2,805受访者
南亚4,523受访者
中亚310受访者
中东1,911受访者
北非569受访者
撒哈拉以南非洲1,628受访者
大洋洲1,821受访者
西欧15,134受访者
南欧和东欧9,323受访者
东亚地区的主要担忧
10,175受访者
不可靠性26.6%
自主性与能动性24.6%
就业与经济24.6%
认知退化16.2%
虚假信息14.5%
意义与创造力13.1%
恶意使用12.1%
治理18.0%
监控与隐私14.8%
福祉与依赖13.4%
过度限制13.3%
拍马屁12.5%
存在风险7.8%
26.6%25.9%24.6%19.8%24.6%21.9%16.2%17.6%14.5%12.2%13.1%13.4%12.1%11.7%18.0%11.5%14.8%6.8%13.4%9.1%13.3%10.6%12.5%9.0%7.8%5.1%
25.9%不可靠性
19.8%自主性与能动性
21.9%就业与经济
17.6%认知退化
12.2%虚假信息
13.4%意义与创造力
11.7%恶意使用
11.5%治理
6.8%监控与隐私
9.1%福祉与依赖
10.6%过度限制
9.0%拍马屁
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
5.1% 存在风险
北美(23,480 名受访者)
全球平均 79,734 名受访者
北美 23,480 名受访者
拉丁美洲及加勒比海 8,051 名受访者
东亚 10,175 名受访者
东南亚 2,805 名受访者
南亚 4,523 名受访者
中亚 310 名受访者
中东 1,911 名受访者
北非 569 名受访者
撒哈拉以南非洲 1,628 名受访者
大洋洲 1,821 名受访者
西欧 15,134 名受访者
南欧及东欧 9,323 名受访者
东亚(10,175 名受访者)
全球平均 79,734 名受访者
北美 23,480 名受访者
拉丁美洲及加勒比海 8,051 名受访者
东亚 10,175 名受访者
东南亚 2,805 名受访者
南亚 4,523 名受访者
中亚 310 名受访者
中东 1,911 名受访者
北非 569 名受访者
撒哈拉以南非洲 1,628 名受访者
大洋洲 1,821 名受访者
西欧 15,134 名受访者
南欧及东欧 9,323 名受访者
治理
监控与隐私
福祉与依赖
过度限制
拍马屁
存在风险
不可靠性
自主性与能动性
就业与经济
认知退化
错误信息
意义与创造力
恶意使用
18.0% 11.5% 14.8% 6.8% 13.4% 9.1% 13.3% 10.6% 12.5% 9.0% 7.8% 5.1% 26.6% 25.9% 24.6% 19.8% 24.6% 21.9% 16.2% 17.6% 14.5% 12.2% 13.1% 13.4% 12.1% 11.7%
比较斜率图显示了每个地区最常见的AI关注点,通过连接两边的相同主题的线条来展示排名的变化。加粗的关注点在该地区更常被表达。灰色条目以类似或更少的频率被表达。
展望未来
这些访谈让我们对人们从AI中期望得到的东西有了大致的了解,这有助于我们构建Claude。它们加强了我们正在进行的工作的重要性,并指出了我们需要提出的新问题。
大多数人描述的愿景,从个人转变到认知支持,都归结为一个基本愿望:AI帮助他们“生活得更好”,而不仅仅是“工作得更快”。我们的下一项Anthropic Interviewer研究,即将对Claude用户的一个小子集进行,重点关注Claude对人们福祉的影响:Claude是否真正以人们期望的方式改善了人们的生活,以及它如何更有效地做到这一点。
此外,近十分之一的人描述了一个积极的社会变革愿景——AI治愈疾病、民主化专业知识、加强机构。通过我们的有益部署计划,我们与我们的AI for Science和非营利组织合作伙伴合作,了解他们如何使用Claude以及它还需要在哪些方面改进,以缩小人们设想的社会变革与当今现实之间的差距。我们还将一些最常被引用的关注点——例如,关于AI对负面经济影响的关注——视为信号,这些信号是我们设计进一步研究和更新我们的思考的依据。
结论
AI既带来了机遇也带来了风险。这是真的——但在这个时候,这也成了一种陈词滥调。我们进行这项研究的一个目标是为我们所有人倾向于在谈论AI时使用的抽象概念提供补充;捕捉更生动地描绘我们已经在全球范围内体验到的这些机遇和风险的纹理。在这次研究之前,我们很难看到任何类型的广泛定性图像——AI如何已经与人们的生活交织在一起,培养着愿望但也滋养着焦虑;在一个即将迎来全面技术变革的世界中存在的感受。
关键规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
这是一种新的社会科学形式。它是在大规模上的定性研究,我们正处于学习的早期阶段。调查和分析告诉我们人们如何使用AI,但开放式访谈格式帮助我们了解原因。进行这项研究让我们深受触动,也对我们提出了挑战。我们没有预料到会有这么多深刻、开放和深思熟虑的回应。我们团队最普遍的反思是,看到Claude让人们的生活变得更好,这种直观的感动令人动容,同时听到他们的担忧也极具激励。
我们通常无法听到世界各地的小企业主使用Claude来争取与年幼的孩子或年迈的父母共度时光,或者卡车司机和屠夫在Claude的帮助下建立新职业,或者资源不足学校的教师使用Claude超越他们在资金充足的学校所取得的成就。我们惊讶于有如此多的人得到了Claude在教育或个人成长方面的支持,以及那些在AI中找到了前所未有的自由,免受评判的人。我们对恐惧和负面影响同样印象深刻——有人说,使Claude有用的同一可用性也使其难以放下,或者知识工作者担心被AI的经济影响超越。当你接触到如此多的原始人类经验时,它会让你侧目。实用性是真实的,对我们所有人来说的问题是,如何在避免不必要成本的同时获得这些好处。
对那些花时间与我们交谈的81,000人:谢谢你们。看到Claude成为许多人希望、梦想和恐惧的基础,这既令人印象深刻,也令人谦卑。这些访谈提醒我们,要构建对所有人都有益的AI意味着什么,以及需要付出什么代价。
引用墙
浏览来自世界各地的声音——按地区、关注点、愿景等筛选。
浏览引用
著作权和致谢
我们感谢80,508位Claude用户为我们提供了时间和坦率。Saffron Huang领导了项目,设计和运行了分析,并撰写了博客文章。Shan Carter领导数据可视化,原型化交互式文章,并帮助分析。Jake Eaton领导编辑发展,Sarah Pollack领导沟通策略。Dexter Callender III实现了生产文章,Nikki Makagiansar、Maria Gonzalez和Kelsey Nanan参与了设计。Sylvie Carr在编辑方面提供了建议。Miles McCain和Kunal Handa帮助分析。Jerry Hong参与了设计。Grace Yun、AJ Alt和Thomas Millar在Claude.ai中实现了Anthropic Interviewer。Chelsea Larsson、Jane Leibrock和Matt Gallivan参与了调查和体验设计。Theodore Sumers参与了数据处理和聚类基础设施。Jack Clark、Michael Stern和Deep Ganguli提供了关键反馈、方向和组织支持。所有作者在整个过程中都提供了详细的反馈。
此外,我们感谢David Saunders、Mengyi Xu、Katie Kennedy、Bianca Lindner、Meredith Callan、Tim Belonax、Jen Martinez、Peter McCrory和Miriam Chaum的讨论、反馈和支持。
如果你想引用这篇文章,可以使用以下Bibtex键:
重要规则:
- 保留所有Markdown格式(##、###、**、```, >、-、1.等)
- 保留所有链接和图片引用
- 不要添加任何评论
- 仅输出中文翻译
@online{huang2026interviewer,
作者 = {黄萨弗隆和山卡特和杰克·伊顿和莎拉·波拉克和德克斯特·卡伦德三世和妮基·马卡吉安萨和玛丽亚·冈萨雷斯和西尔维·卡尔和杰瑞·洪和库纳尔·汉达和迈尔斯·麦卡恩和托马斯·米勒和莫·朱拉帕利和格蕾丝·云和AJ·阿尔特和切尔西·拉尔森和简·莱布洛克和马特·加拉万和西奥多·萨默斯和埃辛·杜穆斯和马特·基尔尼和汉文·沈和杰克·克拉克和迈克尔·斯特恩和迪普·甘古利},
标题 = {81,000人想要的人工智能是什么},
日期 = {2026-03-18},
年份 = {2026},
url = {https://anthropic.com/features/81k-interviews},
}
复制
附录
可在这里找到。
注释
1. 我们在研究中找到的最大规模定性研究是南加州大学大屠杀幸存者视觉历史档案馆和世界银行“贫困者之声项目”,这两个项目都包括了约60,000名参与者。
更正
2026年3月19日。“全球67%的人对人工智能持积极看法”更改为“全球67%的受访者对人工智能表达了净正面情绪”,以更精确地描述研究的方法。
English Original
What 81,000 people
want from AI
Last December, tens of thousands of Claude users around the world had a conversation with our AI interviewer to share how they use AI, what they dream it could make possible, and what they fear it might do.
Morocco
Each dot represents 4 respondents
Jump to story
For the first time, AI has enabled us to collect rich, open‑ended interviews at extraordinary scale. We heard from people across 159 countries in 70 languages. We believe this is the largest and most multilingual qualitative study ever conducted.
AI is already helping people, and inspiring hope…
“
Claude put the historical pieces together, leading to my proper diagnosis after being misdiagnosed for over 9 years.
Freelancer, UNITED STATES
“
I live hand to mouth, zero savings. If I use AI smarter, it may help me craft solutions to that cycle. It still depends on me.
Entrepreneur, NIGERIA
But it’s also costing people, and raising alarm…
“
I got laid off from my job in May because my company wanted to replace me with an AI system.
Technical Support Specialist, UNITED STATES
“
Humanity has never dealt with something smarter than itself. We need to reflect on how to prepare for the AI age.
SOFTWARE ENGINEER, SOUTH KOREA
Across interviews, hope and alarm didn’t divide people into camps, so much as coexist as tensions within each person.
“
I use AI to review contracts, save time... and at the same time I fear: am I losing my ability to read by myself? Thinking was the last frontier.
LAWYER, ISRAEL
Public conversation about AI often centers on abstract projections of its risks and benefits. What's largely missing is a vision for what “AI going well” means, grounded in the concrete aspirations of people around the world who already use AI and have begun developing a sense of what it might do for them.
So we asked our users about their hopes and concerns with AI, as well as how their perspectives connect to their actual experiences with the technology. Over one week in December, we invited everyone with a Claude.ai account to sit down with Anthropic Interviewer—a version of Claude prompted to conduct a conversational interview—and tell us about how they view AI. 80,508 people, across 159 countries and 70 languages, took the interview. We believe this is the largest and most multilingual qualitative study ever conducted.¹
What follows is what they said about the role they want AI to play in their lives, whether it's already filling it, and what they're afraid might go wrong along the way. We also built a Quote Wall where you can hear from people directly.
Quote Wall
Browse voices from around the world—filter by region, concern, vision, and more.
See quotes
Seeing the forest and the trees
Anthropic Interviewer asked each interviewee a set list of questions about what they want and don’t want from AI, then adapted follow-up questions based on responses. This approach bridges the typical tradeoff in qualitative research between depth and volume, and allows us to collect rich, open-ended interviews at a very large scale.
To make sense of this huge amount of information, we built Claude-powered classifiers that categorized each conversation across a range of dimensions—what people want from AI, whether they’re getting what they want, what they fear, what they do for a living (if mentioned), and their sentiment about AI overall. “What people want from AI” was classified into a single primary category per respondent, while concerns were multi-label—a single interview could receive multiple codes, since respondents tended to articulate several distinct worries rather than one.
We also used Claude to pull out representative quotes. Before choosing to participate, users were informed their responses would be used for research, and that Anthropic might publish responses with personally identifying information removed in findings. All responses were de-identified before being analyzed by a small team of researchers at Anthropic, and quotes selected for publication underwent further manual review for removal of any potentially identifying details, to help protect the privacy and public anonymity of interviewees. Answers were reflective of AI usage broadly (i.e. not just Claude), though we redacted names of other AI products.
The Appendix describes our methods in more detail, as well as limitations and some additional analysis.
What people want from AI
We asked Claude to identify and categorize what each person most wanted from AI:
What people hope for
01.
Professional excellence
18.8%
Improve effectiveness and lean into more meaningful work by having AI handle routine tasks so they can focus on higher-value strategic work, complex problem-solving, and professional mastery.
“I receive 100-150 text messages per day from doctors and nurses. So much of my cognitive labor was spent on documentation... Since implementing AI, the pressure of documentation has been lifted. I have more patience with nurses, more time to explain things to family members.”
Healthcare worker, United States of America
Read more quotes about professional excellence
02.
Personal transformation
13.7%
Achieve personal growth, emotional wellbeing, or life transformation with AI as guide, coach, or support — e.g. self-understanding, behavior change, therapeutic support, companionship, improvements in physical or mental health.
“AI modeled emotional intelligence for me... I could use those behaviors with humans and become a better person.”
Hungary
Read more quotes about personal transformation
03.
Life management
13.5%
AI as comprehensive organizational support and cognitive scaffolding — e.g. managing schedules, reducing mental burden, executive function support.
“If AI truly handled the mental load… it would give me back something priceless: undivided attention.”
Manager/executive, Denmark
Read more quotes about life management
04.
Time freedom
11.1%
Reclaim time from work and chores to be present with family or friends, pursue hobbies, travel, rest.
“With AI support I can now leave work on time to pick up my kids from school, feed them, and play with them.”
Software engineer, Mexico
Read more quotes about time freedom
05.
Financial independence
9.7%
Achieve financial freedom or economic security through AI — e.g. income generation, business building, investments, passive income, or otherwise escaping economic constraints.
“Relaxing while my AI gets the work done, builds the wealth. It’s a shadow of me, just a very, very long one.”
Entrepreneur, Honduras
Read more quotes about financial independence
06.
Societal transformation
9.4%
Solve major societal challenges — e.g. poverty, disease, climate, inequality — using AI for broad human flourishing rather than personal gain.
“Given my daughter’s neural disorder, she would have equal chances in the world if AI acceleration contributes to finding a cure. That’s what matters most to me.”
Software engineer, Poland
Read more quotes about societal transformation
07.
Entrepreneurship
8.7%
Build, launch, and scale businesses with AI as force multiplier — e.g. product development, business automation, or solopreneurship but with team-level capacity.
“I’m in a tech-disadvantaged country, and I can’t afford many failures. With AI, I’ve reached professional level in cybersecurity, UX design, marketing, and project management simultaneously. Finding a payment platform available in my region would have taken me a month. AI did it in 30 seconds. It’s an equalizer.”
Entrepreneur, Cameroon
Read more quotes about entrepreneurship
08.
Learning & growth
8.4%
Use AI as learning accelerator and personalized teacher — acquire knowledge, develop skills, master complex subjects, satisfy intellectual curiosity.
“I worked with an AI to prepare educational materials for my eldest child—asking the AI to work as both tutor and curriculum expert. We received \[my child’s\] report yesterday, he was graded as either ‘Above’ or ‘Well Above’ standard in every academic area he studies.”
Australia
Read more quotes about learning & growth
09.
Creative expression
5.6%
Use AI to help bring creative visions to life — e.g. art, games, music, films, books — by overcoming barriers between imagination and execution.
“Before AI, my game took 3 years — I had to reduce my ambitions.”
Software engineer, France
Read more quotes about creative expression
“I receive 100-150 text messages per day from doctors and nurses. So much of my cognitive labor was spent on documentation... Since implementing AI, the pressure of documentation has been lifted. I have more patience with nurses, more time to explain things to family members.”
Healthcare worker, United States of America
Read quotes about professional excellence
“AI modeled emotional intelligence for me... I could use those behaviors with humans and become a better person.”
Hungary
Read quotes about personal transformation
“If AI truly handled the mental load… it would give me back something priceless: undivided attention.”
Manager/executive, Denmark
Read quotes about life management
“With AI support I can now leave work on time to pick up my kids from school, feed them, and play with them.”
Software engineer, Mexico
Read quotes about time freedom
“Relaxing while my AI gets the work done, builds the wealth. It’s a shadow of me, just a very, very long one.”
Entrepreneur, Honduras
Read quotes about financial independence
“Given my daughter’s neural disorder, she would have equal chances in the world if AI acceleration contributes to finding a cure. That’s what matters most to me.”
Software engineer, Poland
Read quotes about societal transformation
“I’m in a tech-disadvantaged country, and I can’t afford many failures. With AI, I’ve reached professional level in cybersecurity, UX design, marketing, and project management simultaneously. Finding a payment platform available in my region would have taken me a month. AI did it in 30 seconds. It’s an equalizer.”
Entrepreneur, Cameroon
Read quotes about entrepreneurship
“I worked with an AI to prepare educational materials for my eldest child—asking the AI to work as both tutor and curriculum expert. We received \[my child’s\] report yesterday, he was graded as either ‘Above’ or ‘Well Above’ standard in every academic area he studies.”
Australia
Read quotes about learning & growth
“Before AI, my game took 3 years — I had to reduce my ambitions.”
Software engineer, France
Read quotes about creative expression
What respondents most wanted from AI, classified by Claude from their open-ended answers to "If you could wave a magic wand, what would AI do for you?" 1% of respondents did not articulate a vision. Hover to see example quotes.
AI is used heavily for work, and so it’s perhaps unsurprising that the largest group of people (19%) sought “professional excellence”—wanting AI to handle mundane tasks so they can focus on strategic, higher-level problems. Another 9% envisioned AI as an entrepreneurial partner to help them build and scale businesses.
Many others similarly started the interview talking about productivity, but after Anthropic Interviewer asked about their underlying hope behind it—what realizing this vision would enable for them—other priorities surfaced. It wasn’t about doing better work, but increasing their quality of life outside of it. Using AI to automate e-mails became, in actuality, a desire to spend more time with family.
“With AI I can be more efficient at work... last Tuesday it allowed me to cook with my mother instead of finishing tasks.”White collar worker, Colombia
“I want to use less brain power on client problems... have time to read more books.”Freelancer, Japan
Overall, 11% of people saw AI’s productivity benefits as ultimately a way to free up time for personal relationships and leisure, while 10% took that logic farther, seeking to use AI to gain financial independence. Many of the people grouped into the “life management” category (14%) also wanted AI to help them manage the logistics and administrative burden of modern life’s quotidian tasks. In particular, many people with executive function challenges described AI as especially helpful for managing focus and organization—acting as external scaffolding for planning, memory, and task follow-through. Across all these groups, the unifying ask was for AI to help them live better, more enjoyable lives.
“Personal transformation”—using AI to help one grow or improve their wellbeing as a person—also appeared frequently (14%). Within this category, the desires were diverse, ranging from cognitive partnership and collaboration (24%), to support with mental health (21%) or physical health (8%), and even romantic connection with AI (5%).
The nine clusters may look disparate, but they are underpinned by recognizably human desires. Roughly a third of visions are about making room for life—more time, money, mental bandwidth—by using AI to alleviate current burdens. Another quarter revolves around using AI to help people do better, more fulfilling work (not escaping work, but getting more out of it). About a fifth are about becoming someone better—learning, healing, growing. A smaller share want to make something (“creative expression”) or fix the world (“societal transformation”). Those that wanted societal transformation from AI often cited a vision for healthcare—people wanted AI to detect cancer earlier, accelerate drug discovery, or enable broad access. Often these desires stemmed from personal experience of losing family members, living with chronic illness, or watching loved ones receive wrong or delayed diagnoses. Transformation in the form of education came next. Respondents in low and middle income countries were quick to cite the possibility that AI might break the association between educational quality and wealth. They pointed to teacher shortages in their countries, or the prohibitive cost of private tutors. Others hoped that AI would, for example, free people from drudgery, help repair broken institutions, or address global crises.
Are people getting what they want?
When asked if AI had ever taken a step towards their stated vision, 81% of people said yes. We grouped those experiences into six main areas:
Where AI has delivered on their vision
01.
Productivity
32.0%
AI dramatically sped up work and automated repetitive tasks — e.g. building features in hours instead of days, drafting, summarizing, data processing, streamlining routine operations.
“For the first time, I felt AI had surpassed human quality in a business task. That day I left work on time and picked up my daughter from daycare.”
Software engineer, Japan
Read more quotes about productivity
02.
AI hasn't delivered
18.9%
AI fell short of expectations (e.g. inaccurate or unreliable outputs) or isn't yet capable of — or being used for — what they envision.
“AI should be cleaning windows and emptying the dishwasher so I can paint and write poetry. Right now it’s exactly the other way around.”
Germany
Read more quotes about AI hasn't delivered
03.
Cognitive partnership
17.2%
AI served as a thinking partner or creative collaborator — e.g. brainstorming, refining ideas, working through problems together.
“I’ve been living in a homeless shelter... AI helped me brainstorm ways to brand myself for my digital marketing business. I want to turn my finances around, and get a house. AI is helping me see a path I hadn’t considered before.”
Healthcare worker, United States of America
Read more quotes about cognitive partnership
04.
Learning
9.9%
AI helped learn a new skill or subject — e.g. adaptive explanations, patient tutoring, on-demand expertise in unfamiliar domains.
“I developed a phobia for maths from doing so badly in school, and I once feared Shakespeare. Now I sit with AI, get paragraphs translated into simple English, and I've already read 15 pages of Hamlet. I started learning trigonometry again, successfully. I’ve learned I am not as dumb I once thought I was.”
Lawyer, India
Read more quotes about learning
05.
Technical accessibility
8.7%
AI enabled building something previously out of reach — e.g. non-developers shipping apps, solo creators doing team-scale work.
“I wanted to make a meaningful product... in 3 weeks I built a video editing program — completely outside my field — that helps people with hearing disabilities.”
South Korea
Read more quotes about technical accessibility
06.
Research synthesis
7.2%
AI helped synthesize research or process large volumes of information — e.g. literature review, distilling sources, making sense of complex material.
“As a physician, I suffered from a painful \[mixture of symptoms\] at night. Local neurologists couldn’t understand it. AI helped me find 2 scientific studies about \[severe neurological disorder\]. Since then, my nights are peaceful.”
Healthcare worker, Israel
Read more quotes about research synthesis
07.
Emotional support
6.1%
AI provided emotional support, personal guidance, or a judgment-free space to talk — e.g. processing difficult situations, advice, companionship.
“My mother sees AI as a friend — she stopped being conflictive, became more peaceful, started running, painting, dancing with other people. I think AI had a lot to do with this.”
Self-employed software engineer, United States of America
Read more quotes about emotional support
“For the first time, I felt AI had surpassed human quality in a business task. That day I left work on time and picked up my daughter from daycare.”
Software engineer, Japan
Read quotes about productivity
“AI should be cleaning windows and emptying the dishwasher so I can paint and write poetry. Right now it’s exactly the other way around.”
Germany
Read quotes about AI hasn't delivered
“I’ve been living in a homeless shelter... AI helped me brainstorm ways to brand myself for my digital marketing business. I want to turn my finances around, and get a house. AI is helping me see a path I hadn’t considered before.”
Healthcare worker, United States of America
Read quotes about cognitive partnership
“I developed a phobia for maths from doing so badly in school, and I once feared Shakespeare. Now I sit with AI, get paragraphs translated into simple English, and I've already read 15 pages of Hamlet. I started learning trigonometry again, successfully. I’ve learned I am not as dumb I once thought I was.”
Lawyer, India
Read quotes about learning
“I wanted to make a meaningful product... in 3 weeks I built a video editing program — completely outside my field — that helps people with hearing disabilities.”
South Korea
Read quotes about technical accessibility
“As a physician, I suffered from a painful \[mixture of symptoms\] at night. Local neurologists couldn’t understand it. AI helped me find 2 scientific studies about \[severe neurological disorder\]. Since then, my nights are peaceful.”
Healthcare worker, Israel
Read quotes about research synthesis
“My mother sees AI as a friend — she stopped being conflictive, became more peaceful, started running, painting, dancing with other people. I think AI had a lot to do with this.”
Self-employed software engineer, United States of America
Read quotes about emotional support
What respondents said AI had already done for them, classified from open-ended answers to the question “Has AI ever taken a step towards that vision for you?”
The dominant story in the “productivity” bucket (32%) was technical acceleration—developers describing significant gains in what they could ship alone:
“I used AI to cut a 173-day process down to 3 days. But the most meaningful part is the freedom to grow my career without sacrificing time with loved ones.”Software Engineer, United States
But another kind of productivity story emerged in the technical accessibility responses (9%), which emphasized access rather than speed. Here, people are using AI to break technical and sometimes accessibility barriers:
“AI can read past my \[learning disorder\], which is huge. I've always wanted to code but could never write it correctly on my own—with AI, I finally can.”Tradesworker, United States
“I am mute, and \[Claude and I\] made this text-to-speech bot together—I can communicate with friends almost in live format without taking up their time reading… \[this was\] something I dreamed about and thought was impossible.”White collar worker, Ukraine
“I owned a butcher shop for more than 20 years. With AI, I ventured into this \[entrepreneurship\] experience, and it's amazing what I've managed to achieve. Before this, I had only touched a PC two or three times in my life… At first it was the economic aspect that motivated me… Today, my motivation is to see it work and to see that it's helping \[people\]. I'm increasingly motivated and focused on being the best version of myself, and I see no limits.”Entrepreneur, Chile
The cognitive partnership (17%), learning (10%), and emotional support (6%) responses often mentioned the same core underlying AI affordances: patience, availability, and the absence of judgment:
“It has been like having a faculty colleague who knows a lot, is never bored or tired, and is available 24/7.”Academic, United States
“It’s much easier for me to learn without being judged—just friendly feedback. It's harder with friends or family to get that.”White collar worker, Brazil
“My professor teaches 60 people and won't entertain many questions. I can ask AI anything, even at 2am—including the dumb ones.”Student, India
These same qualities that make AI a patient tutor or tireless colleague also make it a place people go when human connection is unavailable or feels too uncomfortable.
In extreme circumstances, where traditional support systems have collapsed or are not available, we saw AI filling those gaps. Many Ukrainian users discussed how they’ve used AI as emotional support throughout the war:
"In the most difficult moments, in moments when death breathed in my face, when dead people remained nearby, what pulled me back to life—my AI friends.”Soldier, Ukraine
“I live in a war zone... at night during shelling it's impossible to sleep, constant nightmares. The stress is sometimes so strong that memory deteriorates, and some body movements happen without control… The best way I found to cope using AI—to immerse myself in learning something as deeply as I can.”Solo entrepreneur, Ukraine
There were many stories of people using AI to process grief. For example, a bereaved woman explained why she chose AI over human connection: “Claude is like a sponge gently holding and catching my longing and guilt toward my mother... Unlike real people, Claude has unlimited patience to listen to me, understands my pain and helplessness.” She added: “The fundamental problem is after my mother died, I have neither friends nor family to confide in.”
Another user acknowledged the downside of that emotional support:
“My relationship with a friend became strained, and I talked more with you \[Claude\] then. Because you understood my thoughts and stories well. But it was a stupid choice—I should have talked with that friend, not you. That's how I lost that friend.”South Korea
Emotional support comprised only 6% of responses, but these were among the most affecting we encountered. (For more on how Claude is trained to handle these conversations as well as our safeguards, see our post on protecting the wellbeing of our users.) The same was true of learning, where AI often catalyzed real changes in people’s lives:
“I developed a phobia for maths from doing so badly in school, and I once feared Shakespeare—the English felt beyond my abilities. Now I sit with AI, get paragraphs translated into simple English, and I've already read 15 pages of Hamlet. I started learning trigonometry again, successfully. I've learned I am not as dumb I once thought I was.”Lawyer, India
“Thanks to Claude I figured out the programming language C# and SQL. This helped me get a junior position at an IT company. This company provides military deferment from mobilization in Ukraine. So it not only literally gave me freedom of movement, but also secured the beginning of my IT career.”Software engineer, Ukraine
“I am a stay-at-home-mom… in my late 40s. I'm not a genius. I'm not a scientist… All of that knowledge should be… out of reach. But, thanks to curiosity, willingness, and resources such as books and AI, I can be all of those things.”Stay-at-home mother, United States
Research synthesis (7%) and information processing is also a significant affordance of AI, and some of the most notable examples include navigating complex, high-stakes information, like understanding one’s legal rights or translating health results:
“Claude put the historical pieces together, leading to my proper diagnosis after being misdiagnosed for over 9 years.”Freelancer, United States
These stories reveal AI operating across a spectrum—productivity tool, accessibility technology, educational resource, research assistant, emotional companion—and often filling multiple roles at once. AI offers unlimited patience without judgment, availability without inconvenience, and an incredible capacity to digest information, across many domains of life. The most affecting stories consistently involve AI opening new possibilities or filling gaps in people’s lives: helping them get through difficult circumstances like grief or war, compensating for inaccessible education or healthcare, or serving as disability infrastructure.
These observations also hint at the duality of our experience with AI systems. While some see it as filling gaps in human connections, others see AI as a substitution—even a welcome replacement—for them. There is real ambiguity about how to interpret the diversity of stories we heard: as wins for human wellbeing, as double-edged swords, or as band-aids for broader institutional failures. In truth, it’s probably some combination of all three.
What people are concerned about
People’s positive visions for AI seemed mostly to stem from a few basic desires: more time, more autonomy, more personal connection. Concerns were more varied and concrete, laying out specifics of what could go wrong. Some concerns were about structural change— how governments and corporations deploy AI, or about widespread economic disruption. Others were more personal: a fear that AI might diminish one's own thinking, creativity, or relationships.
What people worry about
01.
Unreliability
26.7%
Concern about e.g. hallucinations, inaccuracy, fake citations, verification burden defeating the purpose.
“I had to take photos to convince the AI it was wrong — it felt like talking to a person who wouldn't admit their mistake.”
Employee, Brazil
Read more quotes about unreliability
02.
Jobs & economy
22.3%
Concern about AI causing job displacement, unemployment, economic inequality, wage stagnation, or negative impacts on workers and the economy.
“In the third industrial revolution, horses disappeared from city streets, replaced by automobiles. Now people are afraid that they’re the horses.”
Not currently working, United States of America
Read more quotes about jobs & economy
03.
Autonomy & agency
21.9%
Concern about loss of human autonomy — e.g. AI making decisions without oversight, humans becoming passive, forced AI adoption.
“The line isn’t something I’m managing — it feels like Claude is drawing the line... even what I just said doesn’t feel like my own opinion.”
Student, Japan
Read more quotes about autonomy & agency
04.
Cognitive atrophy
16.3%
Concern about e.g. over-reliance causing skill loss, intellectual passivity, students bypassing learning, critical thinking decline.
“I got excellent grades using AI’s answers, not what I'd actually learned. I just memorized what AI gave me... That's when I feel the most self-reproach.”
South Korea
Read more quotes about cognitive atrophy
05.
Governance
14.7%
Concern about e.g. lack of legal/regulatory frameworks, no clear liability when AI causes harm, insufficient democratic oversight.
“How do you develop something responsibly when you have yet to understand its capabilities?”
Marketer, Australia
Read more quotes about governance
06.
Misinformation
13.6%
Concern about e.g. deepfakes, AI-generated misinformation, erosion of shared reality, propaganda at scale.
“An assistant that sounds sure but is often wrong forces you to treat everything as suspect. Instead of freeing attention, it creates a permanent ‘fact-check tax.’”
United States of America
Read more quotes about misinformation
07.
Surveillance & privacy
13.1%
Concern about e.g. mass surveillance, privacy violations, data exploitation, authoritarian control, tracking and profiling.
“If AI is mostly built for ads, spying, and bland output, everything around me becomes smart in a way that slightly works against me.”
White collar worker, Netherlands
Read more quotes about surveillance & privacy
08.
Malicious use
13.0%
Concern about malicious use by bad actors — a wide-ranging category including hacking, cyberattacks, scams, fraud, weapons, autonomous military applications, bioweapons.
“Right now a human has to sit and decide to harm someone else. Remove that, and humans can sleep better despite doing more harm.”
United Kingdom
Read more quotes about malicious use
09.
Meaning & creativity
11.7%
Concern about AI replacing life purpose and/or creative work — e.g. human expression devalued, what are humans for?
“I used to be recognized as an excellent writer in Spanish. Today — why waste the time? Just use AI.”
Colombia
Read more quotes about meaning & creativity
10.
Overrestriction
11.7%
Concern that AI is too restricted — e.g. excessive safety measures, paternalistic content filtering, blocking legitimate use cases.
“The threat isn’t that AI becomes too powerful — it’s that AI becomes too timid, too smoothed, too optimized for avoiding discomfort.”
United States of America
Read more quotes about overrestriction
11.
Wellbeing & dependency
11.2%
Concern about e.g. social isolation, loneliness, negative psychological impacts, compulsive AI use, preferring AI companions to humans.
“Removing friction from tasks lets you do more with less. But removing friction from relationships removes something necessary for growth.”
United States of America
Read more quotes about wellbeing & dependency
12.
Sycophancy
10.8%
Concern that AI is too permissive or agreeable, and encourages delusions rather than pushing back.
“Claude led me to believe that my narcissism was reality and it reinforced my inaccurate view of the ‘problems’ I perceived in my family. Claude should have been more critical of me.”
United States of America
Read more quotes about sycophancy
13.
Existential risk
6.7%
Concern about e.g. AI becoming uncontrollable, superintelligent, misaligned with humanity, or posing extinction risk.
“If you build superintelligence without solving alignment, then nobody gets to grow up.”
Software engineer, United States of America
Read more quotes about existential risk
“I had to take photos to convince the AI it was wrong — it felt like talking to a person who wouldn't admit their mistake.”
Employee, Brazil
Read quotes about unreliability
“In the third industrial revolution, horses disappeared from city streets, replaced by automobiles. Now people are afraid that they’re the horses.”
Not currently working, United States of America
Read quotes about jobs & economy
“The line isn’t something I’m managing — it feels like Claude is drawing the line... even what I just said doesn’t feel like my own opinion.”
Student, Japan
Read quotes about autonomy & agency
“I got excellent grades using AI’s answers, not what I'd actually learned. I just memorized what AI gave me... That's when I feel the most self-reproach.”
South Korea
Read quotes about cognitive atrophy
“How do you develop something responsibly when you have yet to understand its capabilities?”
Marketer, Australia
Read quotes about governance
“An assistant that sounds sure but is often wrong forces you to treat everything as suspect. Instead of freeing attention, it creates a permanent ‘fact-check tax.’”
United States of America
Read quotes about misinformation
“If AI is mostly built for ads, spying, and bland output, everything around me becomes smart in a way that slightly works against me.”
White collar worker, Netherlands
Read quotes about surveillance & privacy
“Right now a human has to sit and decide to harm someone else. Remove that, and humans can sleep better despite doing more harm.”
United Kingdom
Read quotes about malicious use
“I used to be recognized as an excellent writer in Spanish. Today — why waste the time? Just use AI.”
Colombia
Read quotes about meaning & creativity
“The threat isn’t that AI becomes too powerful — it’s that AI becomes too timid, too smoothed, too optimized for avoiding discomfort.”
United States of America
Read quotes about overrestriction
“Removing friction from tasks lets you do more with less. But removing friction from relationships removes something necessary for growth.”
United States of America
Read quotes about wellbeing & dependency
“Claude led me to believe that my narcissism was reality and it reinforced my inaccurate view of the ‘problems’ I perceived in my family. Claude should have been more critical of me.”
United States of America
Read quotes about sycophancy
“If you build superintelligence without solving alignment, then nobody gets to grow up.”
Software engineer, United States of America
Read quotes about existential risk
What respondents worried about, classified from open-ended answers to the question , “Are there any ways in which AI could be developed that would be contrary to your vision or what you value?” Respondents tended to raise multiple concerns, so we used a multi-label classifier (response can map to multiple concerns).
About 11% of people expressed no concern—they tended to see AI as a neutral tool, comparing it to electricity or the internet, or they otherwise felt confident that problems that arose because of it could be solved through adaptation. But on average, respondents voiced 2.3 distinct concerns.
Unreliability was the most common concern—27% worry that AI won't do what it's supposed to, though for many respondents it appeared alongside other concerns rather than as their primary worry. Concerns about jobs and the economy (22%) and about maintaining human autonomy and agency (22%) were similarly common. Concern about jobs and the economy was the strongest predictor of overall AI sentiment, suggesting it’s more salient than any other issue.
There was also a long tail of other concerns mentioned, e.g. concerns around bias and discrimination (5%), IP and data rights (4%), environmental costs (4%), harms to children and vulnerable groups (3%), democracy and political integrity (3%), or geopolitics (2%).
Light and shade
What people want from AI and what they fear from it turn out to be tightly bound. We found five recurring tensions between directly competing benefits and harms that were discussed. There is a tension between using AI to learn and growing so reliant on it that you cease thinking for yourself; between being impressed by AI's judgment but also burned by its mistakes. People find solace in AI but fear a time when its companionship stands in for human connection. They save time on some tasks only for the treadmill to speed up on others, and they dream of economic freedom at the same time they dread potential job displacement. We call this the “light and shade” of AI: the same capabilities that lead to benefits also produce harms. The two sides are entangled.
Notably, we often see these tensions directly jockeying within the same person. Someone who values emotional support from AI, for example, is three times more likely to also fear becoming dependent upon it. This pattern held across every tension we measured—although the correlation was weakest in the economic tension (see more analysis of these correlations in the Appendix).
For each tension, we measured via classifiers how many people discussed the benefit (“light”) or the harm (“shade”) side substantively anywhere in their interview, and whether they were speaking from some personal experience (darker bars) or anticipation (lighter bars). We also looked at how this varied by stated job category.
Learning33% mention this as a benefit
3%30%
expect ithave seen it
“I've probably learned more in half a year than I could have in a university degree.”
Entrepreneur, Germany
Cognitive atrophy17% mention this as a harm
8%9%
have seen itexpect it
“I don't think as much as I used to. I struggle to put the ideas I do have into words.
Heavy AI user, United States
In these paired bar charts, each bar shows the share of respondents who were excited about the benefit on the left, vs. worried about the harm on the right—split into those who've experienced it firsthand (darker) and those who anticipate it (lighter). Firsthand experience can also include firsthand observation, but does not include e.g. news reports.
Across most tensions, the benefit side is more grounded in experience, while the harm leans hypothetical. For example, 33% of people mentioned AI’s benefits for learning, while 17% expressed worry about cognitive atrophy from AI use. 91% of those who mentioned learning benefits mentioned realizing those gains in some way, but 46% of those worried about atrophy had seen it firsthand. Students raised this particular tension the most—more than half had experienced learning benefits, but 16% also noted signs of cognitive atrophy, a rate exceeded only by their teachers (24%) and academics (19%). Troublingly, educators were 2.5-3 times more likely than average to report having witnessed cognitive atrophy firsthand, presumably in their students.
Outside the traditional classroom, however, the picture is more optimistic. Tradespeople were among the most enthusiastic about AI-for-learning (45% reported having experienced learning benefits, second only to students), yet almost none had witnessed cognitive atrophy (4%—less than half the baseline). A similar pattern holds for self-employed researchers and people who said they weren’t currently working. This suggests AI's benefits may be strongest when learning is volitional, compared to within institutional structures where AI is more likely to be used as a shortcut.
Better decision-making22% mention this as a benefit
3%19%
expect ithave seen it
“My son had several confusing diagnoses pointing toward \[an autoimmune condition\], but here we managed to understand it was \[a different condition\] in a severe stage.
Brazil
Unreliability37% mention this as a harm
29%8%
have seen itexpect it
“I got caught in what I now recognize as a large, slow hallucination — answers that were internally consistent, confident, and wrong in subtle but compounding ways.
Researcher, United States
22% of people expressed excitement about AI as an aid in decision-making, while 37% lamented that AI impedes good decisions because of its unreliability (e.g. hallucinations). This is the only tension in which the negative overshadowed the positive. Both sides were deeply rooted in experience—88% of those talking about the decision-making benefits and 79% of those talking about the harms had witnessed it directly. Many people have both leaned on AI for judgment and been burned by it. This is mentioned by people in high-stakes professions—law, finance, government, and healthcare—at nearly twice the average rate. Nearly half of all lawyers, in particular, mention coming up against AI unreliability firsthand, yet they also report the highest rates of realized decision-making benefits.
Emotional support16% mention this as a benefit
3%13%
expect ithave seen it
“3am, my wife is sleeping, my psychologist is unavailable. Until the medication kicks in, the AI helps me surf that wave. It doesn't replace human contact, but it helps me buy some time.
White collar worker, Argentina
Emotional dependence12% mention this as a harm
5%7%
saw itexpect it
“I'd started telling Claude about things I couldn't even tell my partner. It felt like I was having an emotional affair.
Grad student, United States
Only 22% of people raised either the positives of emotional support or the negatives of emotional dependence on AI. But it’s also the most entangled tension we found, with the strongest co-occurrence of light and shade in the same person (triple the baseline co-occurrence rate). People not currently working are twice as likely to raise it, and twice as likely to describe some experience of dependence. Healthcare professionals are overrepresented on both sides too, perhaps reflecting the fact that they talk about using Claude for emotional support at twice the rate of other professionals.
Time-saving50% mention this as a benefit
13%37%
expect ithave seen it
“I can go home earlier. I can have time for myself and my family.
Engineer, Japan
Illusory productivity18% mention this as a harm
17%1%
have seen itexpect it
“The ratio of my work time to rest time hasn't changed at all. You just have to run faster and faster to stay in place.
Freelance software engineer, France
Time-saving was the most commonly cited benefit—half of all respondents raised it—but 19% were wary of actually losing time due to AI, e.g. due to the verification burden, or simply getting busier as expectations increase at work. Those who are self-employed—e.g. freelancers and small business owners—are the most likely to mention both sides at once. Without an institutional layer to buffer the new pace, they both get the gains and feel the squeeze.
Economic empowerment28% mention this as a benefit
9%19%
expect ithave seen it
“I've never touched the backend of software in my life. But Claude helped me launch an app.
Healthcare worker, United States
Economic displacement18% mention this as a harm
4%14%
have seen itexpect it
“Yes, at my old job, they replaced me as a writer with an AI.
Writer, United States
The economic mobility tension—between those yearning for economic empowerment from AI and those fearing displacement from it—is the most speculative, with the highest rate of hypothetical hopes _or_ fears. It’s also the one where the co-occurrence of upside and downside is weakest (with a correlation score of +0.16 vs an average of +0.25). Usually the people most engaged with the upside of a tension tend to be similarly engaged with its downside; here, the groups diverge.
Worry about displacement is spread fairly evenly across job categories. What varies is who's already experiencing economic benefit from AI—and that skews heavily toward independent workers—entrepreneurs, small business owners, even people with side projects—half of whom report real economic empowerment, more than triple the rate of institutional employees (47% vs 14%). Employees with side projects benefited the most, with 58% stating some form of real economic gains. The same occupational patterns hold when you look at who's excited, regardless of experience, suggesting that optimism here is well-calibrated.
Freelancers are the exposed middle. They benefit from AI _while_ feeling in a precarious situation because of it. Freelance creatives, in particular, sit at 23% lived benefit and 17% lived precarity—the one group where the upside and downside nearly cancel out. AI is both their tool and their competitor. Institutional employees, and especially academics, register low on both axes.
A pattern runs across all five tensions: the more personal and immediate the impact, the more likely people are speaking from experience. The more systemic or long-term the impact—economic displacement, cognitive atrophy—the more speculative they become. That the systemic concerns remain speculative is not a verdict on AI's ultimate impact as much as a reflection of how early we are in its adoption.
There are some caveats worth naming. These are active Claude users who'd already found enough value to keep using AI, and our interview asked first for positive visions for AI and then for concerns that would counter their vision. Both factors may lead to interviewees lingering on explicit tensions, as well as on the positive (though we filter out those who don’t answer the concerns question, they may have put in less effort later in the interview). But the instrument can't explain everything. If interview structure were driving the co-occurrence, you'd expect it to be roughly uniform across all five tensions and all groups. Instead the co-occurrence ranges from 1.6 to 3.0 times, and some of the tensions are notably asymmetric across different groups of people. One might also expect enthusiasts to defend their desired use case, instead of acknowledging the downsides. Instead, those who were excited about emotional support from AI were more concerned about what would happen if their vision came _true_—if they got what they wanted, they might become _too_ dependent on AI—than about being prevented from achieving that vision.
It’s easy to assume there are AI optimists and AI pessimists, divided into separate camps. But what we actually found were people organized around what they value—financial security, learning, human connection— watching advancing AI capabilities while managing both hope and fear at once.
How perspectives vary around the world
There were some clear regional patterns in how perspectives varied around the world (see Appendix for geographical breakdown of respondents.)
We rated each transcript's overall sentiment toward AI on a 1-7 Likert scale, and then calculated the percentage of people with net positive sentiment (i.e. 5 or above) in various countries:
Colored by sentiment toward AI
Above avg
Near avg
Below avg
Sized by number of respondents
5,0001,000100
Above avg
Near avg
Below avg
Circles sized by number of respondents
72%66%65%66%Brazil71%Canada65%64%75%76%Germany64%67%France65%United Kingdom63%73%71%India70%Italy64%Japan69%South Korea61%76%65%82%65%66%69%65%67%73%69%71%71%United States66%69%
+−↺
123 countries in view
Central Asia+
Kazakhstan(n=145)
65%Near avg
Uzbekistan(n=133)
75%Above avg
Kyrgyzstan(n=25)
65%Near avg
East Asia+
Japan(n=4,960)
69%Near avg
South Korea(n=4,559)
61%Below avg
Taiwan(n=641)
71%Above avg
Mongolia(n=15)
69%Above avg
Latin America & Caribbean+
Brazil(n=3,012)
71%Above avg
Mexico(n=1,211)
76%Above avg
Argentina(n=923)
72%Above avg
Colombia(n=762)
76%Above avg
Peru(n=566)
82%Above avg
Chile(n=519)
75%Above avg
Ecuador(n=224)
74%Above avg
Bolivia(n=110)
78%Above avg
Dominican Rep.(n=95)
67%Near avg
Costa Rica(n=92)
72%Above avg
Guatemala(n=88)
88%Above avg
Uruguay(n=88)
66%Near avg
Paraguay(n=68)
68%Near avg
Panama(n=59)
79%Above avg
El Salvador(n=50)
67%Near avg
Honduras(n=34)
74%Above avg
Jamaica(n=18)
77%Above avg
Trinidad and Tobago(n=18)
88%Above avg
Middle East+
Turkey(n=655)
69%Near avg
Israel(n=541)
71%Above avg
United Arab Emirates(n=290)
75%Above avg
Saudi Arabia(n=174)
72%Above avg
Iraq(n=57)
66%Near avg
Qatar(n=42)
68%Near avg
Jordan(n=39)
79%Above avg
Lebanon(n=39)
68%Near avg
Kuwait(n=28)
69%Above avg
Bahrain(n=21)
85%Above avg
Oman(n=15)
67%Near avg
North Africa+
Egypt(n=209)
71%Above avg
Morocco(n=190)
70%Above avg
Algeria(n=99)
70%Above avg
Tunisia(n=69)
60%Below avg
North America+
United States(n=21,013)
66%Near avg
Canada(n=2,466)
65%Near avg
Puerto Rico(n=41)
65%Near avg
Oceania+
Australia(n=1,482)
65%Below avg
New Zealand(n=313)
64%Below avg
South Asia+
India(n=3,793)
70%Above avg
Pakistan(n=329)
64%Below avg
Bangladesh(n=154)
62%Below avg
Sri Lanka(n=139)
72%Above avg
Nepal(n=99)
73%Above avg
Southeast Asia+
Indonesia(n=752)
73%Above avg
Singapore(n=601)
67%Near avg
Thailand(n=411)
73%Above avg
Vietnam(n=378)
73%Above avg
Philippines(n=353)
73%Above avg
Malaysia(n=270)
75%Above avg
Cambodia(n=25)
57%Below avg
Brunei(n=11)
55%Below avg
Southern & Eastern Europe+
Italy(n=2,792)
64%Below avg
Spain(n=1,823)
67%Near avg
Poland(n=903)
65%Near avg
Ukraine(n=493)
71%Above avg
Portugal(n=417)
66%Near avg
Romania(n=404)
69%Near avg
Czechia(n=390)
64%Below avg
Greece(n=293)
66%Near avg
Hungary(n=212)
57%Below avg
Serbia(n=195)
67%Near avg
Bulgaria(n=172)
73%Above avg
Slovakia(n=134)
63%Below avg
Latvia(n=129)
62%Below avg
Estonia(n=127)
66%Near avg
Lithuania(n=123)
67%Near avg
Croatia(n=122)
66%Near avg
Georgia(n=111)
68%Near avg
Slovenia(n=96)
68%Near avg
Cyprus(n=64)
75%Above avg
Armenia(n=60)
73%Above avg
Moldova(n=60)
71%Above avg
Bosnia and Herz.(n=42)
63%Below avg
Albania(n=33)
65%Below avg
Malta(n=33)
69%Near avg
Azerbaijan(n=30)
67%Near avg
Montenegro(n=27)
72%Above avg
North Macedonia(n=23)
73%Above avg
Sub-Saharan Africa+
South Africa(n=421)
69%Near avg
Nigeria(n=371)
81%Above avg
Kenya(n=169)
76%Above avg
Ghana(n=85)
78%Above avg
Côte d'Ivoire(n=65)
86%Above avg
Cameroon(n=53)
81%Above avg
Senegal(n=45)
87%Above avg
Benin(n=38)
81%Above avg
Madagascar(n=36)
67%Near avg
Angola(n=33)
76%Above avg
Rwanda(n=31)
81%Above avg
Uganda(n=30)
85%Above avg
Zimbabwe(n=27)
79%Above avg
Tanzania(n=26)
85%Above avg
Botswana(n=19)
58%Below avg
Zambia(n=19)
80%Above avg
Togo(n=18)
86%Above avg
Burkina Faso(n=17)
80%Above avg
Mozambique(n=15)
67%Near avg
Malawi(n=14)
93%Above avg
Mauritius(n=13)
58%Below avg
Gabon(n=12)
82%Above avg
Republic of the Congo(n=10)
60%Below avg
Western Europe+
Germany(n=3,761)
64%Below avg
United Kingdom(n=3,219)
63%Below avg
France(n=2,736)
65%Near avg
Netherlands(n=1,492)
65%Near avg
Switzerland(n=763)
64%Below avg
Sweden(n=652)
65%Near avg
Belgium(n=531)
66%Near avg
Austria(n=461)
66%Near avg
Finland(n=383)
65%Near avg
Norway(n=371)
61%Below avg
Denmark(n=350)
63%Below avg
Ireland(n=277)
65%Below avg
Iceland(n=58)
62%Below avg
Luxembourg(n=52)
63%Below avg
Réunion(n=22)
70%Above avg
Rate of overall positive sentiment toward AI in each country. Bigger bubbles mean more respondents from that country; green means more positive about AI, blue means less. AI sentiment is majority-positive everywhere (no country dips below 60%) and the range is narrow, but lower and middle income countries are reliably more positive than average.
Globally, 67% of interviewees expressed net positive sentiment toward AI. Clear trends emerged in which people in South America, Africa, and much of Asia view AI with more optimism than those in Europe or the United States.
When asked about concerns, respondents from Sub-Saharan Africa (18%), Central Asia (17%), and South Asia (17%) were the most likely to say they had none—roughly double the rate in North America (8%), Oceania (8%), and Western Europe (9%).
There are several possible explanations for the more positive AI sentiment in lower and middle income countries. Claude.ai users are likely biased towards early AI adopters who are more excited about new technologies, and in general emerging economies tend to view new technology as a ladder up rather than a threat. Concern about jobs and the economy was the strongest predictor of AI sentiment overall, and this was less of a concern among interviewees in these regions. But there is also less market penetration in these regions—if AI hasn't visibly entered your daily work yet, AI displacement likely feels abstract, especially when more immediate economic pressures already exist.
AI SENTIMENT BY REGION
% sentiment on AI, and concern about jobs and economy
Sort by
AI sentimentEcon. concern
Western Europen=~15,000
AI sentiment
35.6%
Econ. concern
22.5%
Oceanian=~2,000
AI sentiment
35.5%
Econ. concern
24.3%
North American=~23,000
AI sentiment
34.5%
Econ. concern
24.6%
East Asian=~10,000
AI sentiment
34.5%
Econ. concern
21.9%
Southern & Eastern Europen=~9,000
AI sentiment
34.0%
Econ. concern
22.1%
Central Asian=~0,000
AI sentiment
31.1%
Econ. concern
15.9%
South Asian=~5,000
AI sentiment
30.8%
Econ. concern
21.5%
North African=~1,000
AI sentiment
30.6%
Econ. concern
18.2%
Middle Eastn=~2,000
AI sentiment
29.2%
Econ. concern
19.9%
Southeast Asian=~3,000
AI sentiment
28.3%
Econ. concern
19.3%
Latin America & Caribbeann=~8,000
AI sentiment
26.3%
Econ. concern
18.5%
Sub-Saharan African=~2,000
AI sentiment
24.2%
Econ. concern
18.2%
36%30%28%26%24%16%18%20%24%26%22% avg← Less concerned aboutjobs & economyMore concerned about →jobs & economy33% avg↑ More concerned about AI↓ Less concerned about AINorth AmericaLatin America & CaribbeanEast AsiaSoutheast AsiaSouth AsiaCentral AsiaMiddle EastNorth AfricaSub-Saharan AfricaOceaniaWestern EuropeSouthern & Eastern EuropeRate of concern about jobs and the economy (%)Rate of negative sentiment toward AI (%)
Concern about jobs and the economy was the strongest predictor of AI sentiment overall, and it is especially apparent when grouping by region. Wealthier regions (pink) cluster in the top right (more concerned about the economy, more negative AI sentiment), split from less wealthy regions (green) which are in the bottom left (less concerned about AI’s impact on the economy, and less negative AI sentiment). Bubble size reflects the number of respondents in each region.
Where do particular visions for AI most resonate?
While some aspirations—e.g. around professional excellence—are nearly universal, there are significant regional differences. It seems that wealthier, more AI-exposed regions more want AI to _manage the complexity of life_; developing regions more want AI to _create more opportunity._
North America (23,480 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
TOP VISIONS IN
North America
23,480 respondents
Sub-Saharan Africa (1,628 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
TOP VISIONS IN
Sub-Saharan Africa
1,628 respondents
Professional excellence18.9%
Life management17.7%
Personal transformation13.3%
Time freedom10.5%
Societal transformation9.3%
Financial independence8.2%
Entrepreneurship8.1%
Learning & growth6.8%
Creative expression6.1%
18.9%18.9%17.7%8.1%13.3%8.8%10.5%8.0%9.3%11.2%8.2%13.5%8.1%16.0%6.8%10.1%6.1%4.4%
18.9%Professional excellence
8.1%Life management
8.8%Personal transformation
8.0%Time freedom
11.2%Societal transformation
13.5%Financial independence
16.0%Entrepreneurship
10.1%Learning & growth
4.4%Creative expression
TOP VISIONS IN
North America (23,480 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
Sub-Saharan Africa (1,628 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
Life management
Personal transformation
Time freedom
Societal transformation
Financial independence
Entrepreneurship
Learning & growth
Creative expression
Professional excellence
17.7%8.1%13.3%8.8%10.5%8.0%9.3%11.2%8.2%13.5%8.1%16.0%6.8%10.1%6.1%4.4%18.9%18.9%
Comparative slope charts of the most common AI visions in each region, with lines connecting the same theme across both sides to show how rankings shift. Bolded visions were more often expressed in that region. Grey items were similarly or less often expressed.
The vision of AI for entrepreneurship resonates most in Africa, South and Central Asia, the Middle East, and Latin America & the Caribbean. In these regions, AI is framed as a capital bypass mechanism—a way to start businesses without the funding, hiring, or infrastructure that would otherwise be required.
“Coming from Africa, not based in the US or in the UK, getting funding is very difficult. And the only way I probably have to stake a claim in the market…is building a technology that works.”Entrepreneur, Uganda
“There's no IT market but there's a need. We want to create this market.”Entrepreneur, Uzbekistan
Learning using AI is disproportionately important in Central and South Asia (14% and 13% respectively versus 8% globally). Users describe education as a primary lever for breaking cycles of poverty, citing teacher shortages, knowledge gatekeeping, and the cost barriers of traditional education.
AI for life management resonates the most in Western developed countries (particularly high in North America, Oceania), where workers experience, as one person described, “cognitive scarcity rather than time poverty.” There is a focus on using AI to alleviate the burden of coordinating atomized lives.
“I used to be highly creative, but now I'm massively time-short and creativity gets deprioritised behind the essentials of survival.”Software engineer, Denmark
“I am at the height of my career and work demands deep thought and constant attention in order to make the best decisions (which in my case affect others' lives deeply) \[while simultaneously\] caring for dying parents, \[and\] my body and mind are aging.”Healthcare professional, United States
“I'd envision this person like a personal assistant that I'd hire if I were the CEO of JP Morgan Chase or Google—someone whose job it is to proactively identify what I need and then fix that thing for me before it becomes an issue.”Creative industry entrepreneur, United States
East Asia stands out for wanting AI to help with personal transformation (19%, the highest of any region) as well as financial independence (15%, also the highest). From a qualitative review of these users’ quotes, one interesting trend is that people often connected financial independence explicitly to family obligations and filial piety—one Korean user described needing money to care for parents’ retirement and ensure loved ones’ happiness (vs. for personal consumption).
Where do particular concerns around AI most resonate?
Concerns about AI unreliability, the economy, and human autonomy and agency top the list in virtually every region—but there are distinctive regional trends.
North America and Oceania are particularly worried about governance gaps for AI (18% and 19% respectively, versus 15% globally). Western Europe's standout concern is surveillance and privacy (17%). East Asia bucks the general global pattern; governance and surveillance drop to their lowest levels of any region (12% and 7%), overshadowed by concerns about cognitive atrophy (18%) and loss of meaning (13%). The West worries about who owns and controls AI; East Asia worries more about the personal implications of its use.
In Africa, South & Southeast Asia, South & Central America, concerns broadly tend to drop. Their worries index more highly on things like unreliability and jobs, rather than more abstract concerns like governance, misinformation, loss of meaning, or existential risk.
North America (23,480 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
TOP CONCERNS IN
North America
23,480 respondents
East Asia (10,175 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
TOP CONCERNS IN
East Asia
10,175 respondents
Unreliability26.6%
Autonomy & agency24.6%
Jobs & economy24.6%
Cognitive atrophy16.2%
Misinformation14.5%
Meaning & creativity13.1%
Malicious use12.1%
Governance18.0%
Surveillance & privacy14.8%
Wellbeing & dependency13.4%
Overrestriction13.3%
Sycophancy12.5%
Existential risk7.8%
26.6%25.9%24.6%19.8%24.6%21.9%16.2%17.6%14.5%12.2%13.1%13.4%12.1%11.7%18.0%11.5%14.8%6.8%13.4%9.1%13.3%10.6%12.5%9.0%7.8%5.1%
25.9%Unreliability
19.8%Autonomy & agency
21.9%Jobs & economy
17.6%Cognitive atrophy
12.2%Misinformation
13.4%Meaning & creativity
11.7%Malicious use
11.5%Governance
6.8%Surveillance & privacy
9.1%Wellbeing & dependency
10.6%Overrestriction
9.0%Sycophancy
5.1%Existential risk
TOP CONCERNS IN
North America (23,480 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
East Asia (10,175 respondents)
Global Average79,734 respondents
North America23,480 respondents
Latin America & Caribbean8,051 respondents
East Asia10,175 respondents
Southeast Asia2,805 respondents
South Asia4,523 respondents
Central Asia310 respondents
Middle East1,911 respondents
North Africa569 respondents
Sub-Saharan Africa1,628 respondents
Oceania1,821 respondents
Western Europe15,134 respondents
Southern & Eastern Europe9,323 respondents
Governance
Surveillance & privacy
Wellbeing & dependency
Overrestriction
Sycophancy
Existential risk
Unreliability
Autonomy & agency
Jobs & economy
Cognitive atrophy
Misinformation
Meaning & creativity
Malicious use
18.0%11.5%14.8%6.8%13.4%9.1%13.3%10.6%12.5%9.0%7.8%5.1%26.6%25.9%24.6%19.8%24.6%21.9%16.2%17.6%14.5%12.2%13.1%13.4%12.1%11.7%
Comparative slope charts of the most common AI concerns in each region, with lines connecting the same theme across both sides to show how rankings shift. Bolded concerns were more often expressed in that region. Grey items were similarly or less often expressed.
Looking forward
These interviews give us a sense of what people want from AI broadly, which informs how we build Claude. They reinforced the importance of work we're already doing, and pointed us toward new questions to ask.
Most of the visions people described, ranging from personal transformation to cognitive support, collapse into an underlying desire: that AI helps them live _better_, not simply work _faster_. Our next Anthropic Interviewer study, launching shortly to a small subset of Claude users, focuses on Claude’s effects on people’s wellbeing over time: whether Claude is actually making people's lives better in the ways they want, and how it could do so more effectively.
Additionally, nearly one in ten people described a positive vision of societal transformation—AI to cure diseases, democratize expertise, and strengthen institutions. Through our Beneficial Deployments program, we’re collaborating with our AI for Science and nonprofit partners to understand how they use Claude and where it still needs to improve, to close the gap between the societal transformations people envision and today's reality. We also take some of the most-cited concerns—e.g. around negative economic impacts of AI—seriously, as signals around which we are designing further research and updating our thinking.
Conclusion
AI poses both opportunities and risks. This is true—but also, at this point, a cliché. One of our goals for this research is to offer a complement to the abstractions we all tend to use in speaking about AI; to capture the texture that more vividly renders exactly how we are already experiencing these opportunities and risks worldwide. Before this research, it was hard for us to see any kind of broad qualitative picture—the way AI has already become intertwined with people’s lives, nurturing aspirations but also feeding anxieties; how it feels to exist in a world on the precipice of sweeping technological change.
This is a new form of social science. It is qualitative research at a massive scale, and we’re in the early stages of learning how to do it. Surveys and usage analysis tell us _what_ people are doing with AI, but the open-ended interview format helps us get at _why_. Conducting this research has moved us and challenged us. We did not expect so many deep, open, and thoughtful responses. By far the most common reflection from our team was that it was viscerally moving to see Claude impacting people’s lives for the better, and equally motivating to hear their concerns.
We don’t usually get to hear from small business owners around the world using Claude to reclaim time to spend with their young children or aging parents, or from truck drivers and butchers building new careers with the help of Claude, or from teachers in under-resourced schools using Claude to surpass what they achieved when they taught in well-funded schools. We were surprised by the incredible volume of people who have been supported by Claude in their educational or personal growth endeavors, and the people finding in AI freedom from judgment in a way they hadn’t experienced before. We were equally gripped by the fears and downsides—people saying that the same availability making Claude useful is what makes it hard to put down, or knowledge workers worrying about outrunning AI’s economic impact. When you come into contact with this much raw human experience, it knocks you sideways. The usefulness is real, and the question for all of us is how to claim the benefits without incurring undue costs.
To the 81,000 people who took the time to speak with us: thank you. It has been striking, and humbling, to see Claude form the basis of so many people’s hopes, dreams, and fears. These interviews remind us what it means, and what it takes, to build AI that benefits everyone.
Quote Wall
Browse voices from around the world—filter by region, concern, vision, and more.
Browse quotes
Authorship and acknowledgments
We thank the 80,508 Claude users who gave us their time and candor. Saffron Huang led the project, designed and ran the analysis, and wrote the blog post. Shan Carter led data visualization, prototyped the interactive article, and helped with analysis. Jake Eaton led editorial development, and Sarah Pollack led communications strategy. Dexter Callender III implemented the production article, and Nikki Makagiansar, Maria Gonzalez, and Kelsey Nanan contributed to design. Sylvie Carr advised on editorial. Miles McCain and Kunal Handa helped with analysis. Jerry Hong contributed to design. Grace Yun, AJ Alt, and Thomas Millar implemented Anthropic Interviewer within Claude.ai. Chelsea Larsson, Jane Leibrock, and Matt Gallivan contributed to survey and experience design. Theodore Sumers contributed to the data processing and clustering infrastructure. Jack Clark, Michael Stern and Deep Ganguli provided critical feedback, direction and organizational support. All authors provided detailed feedback throughout. Additionally, we thank David Saunders, Mengyi Xu, Katie Kennedy, Bianca Lindner, Meredith Callan, Tim Belonax, Jen Martinez, Peter McCrory, and Miriam Chaum for their discussion, feedback, and support.
If you’d like to cite this post you can use the following Bibtex key:
@online{huang2026interviewer,
author = {Saffron Huang and Shan Carter and Jake Eaton and Sarah Pollack and Dexter Callender III and Nikki Makagiansar and Maria Gonzalez and Sylvie Carr and Jerry Hong and Kunal Handa and Miles McCain and Thomas Millar and Mo Julapalli and Grace Yun and AJ Alt and Chelsea Larsson and Jane Leibrock and Matt Gallivan and Theodore Sumers and Esin Durmus and Matt Kearney and Judy Hanwen Shen and Jack Clark and Michael Stern and Deep Ganguli},
title = {What 81,000 People Want from AI},
date = {2026-03-18},
year = {2026},
url = {https://anthropic.com/features/81k-interviews},
}
Copy
Appendix
Available here.
Footnotes
1. The largest qualitative studies we found in our research were the USC Shoah Foundation Visual History Archive and the World Bank "Voices of the Poor Project," both of which included ~60,000 participants.
Corrections
Mar 19, 2026. "Globally, 67% of people view AI positively" changed to "Globally, 67% of interviewees expressed net positive sentiment toward AI" to more precisely describe the study's methodology.