内容来源：https://www.theverge.com/entertainment/897923/ghost-in-the-machine-valerie-veatch-interview

内容总结：

纪录片导演揭露生成式AI背后的种族主义幽灵：技术狂欢下的优生学阴影

2024年，当OpenAI首次向公众发布文生视频模型Sora时，导演瓦莱丽·维奇与许多人一样被其吸引。尽管不完全理解这项技术，她仍好奇其潜力，并看到艺术家们组建在线社区分享AI创作。然而，踏入AI领域后，维奇震惊地发现，这项技术频繁生成充满种族主义和性别歧视的图像，而周围的AI爱好者们对此却漠不关心。

这一现象促使维奇远离早期的AI实验，并推动她执导纪录片《机器中的幽灵》，深入探究生成式AI背后的技术基础与思想根源。影片没有聚焦于AI加速主义者鼓吹的“未来福祉”，而是追溯历史，揭示该技术为何以当前形态存在。

维奇指出，“人工智能”一词自1956年被计算机科学家约翰·麦卡锡提出以来，始终是一个模糊的营销术语。影片强调，AI的发展脉络实际可追溯至维多利亚时代的优生学。优生学创始人弗朗西斯·高尔顿（查尔斯·达尔文的表弟）提出的种族优劣论，虽已被科学界摒弃，但其通过多维建模量化人类特征的方法，影响了卡尔·皮尔森等统计学家的发展，而皮尔森的逻辑回归等成果正是现代机器学习的基础组成部分。

“当我以纪录片作者身份研究‘超级智能’时，很快便撞上了种族科学的低门框——它已深植于技术之中。”维奇表示，这些概念浸透着优生学思维。

影片通过历史分析解释，为何开发AI技术的公司对解决其现有问题兴趣缺缺。维奇分享亲身经历：她在艺术家Slack群组中使用早期Sora时，一位有色人种女性成员发现，模型总会将她的形象“白化”。维奇向群组指出这是软件本身的问题，却无人回应。她随后联系OpenAI，反馈模型输出中存在种族主义、性别歧视内容（例如女性在生成场景中长出多余乳房或突然跳电臀舞），但公司仅回应称“提出这些令人尴尬，我们无法改变”。

这一经历促使维奇深入挖掘生成式AI屡现问题的根源。她发现，从高尔顿的优生学统计到现代生成式AI机构，存在一条清晰的历史连线。《机器中的幽灵》采访的AI研究员、历史学家和批判理论家共同指出，AI领域的几乎每个层面都深受其历史上与歧视性科学领域的关联影响。

当被问及是否考虑采访被影片批评的AI公司负责人时，维奇笑道：“他们已拥有足够的媒体发言权。难道我要在镜头前拥抱萨姆·奥特曼吗？那将成为宣传片，而非关于技术真相的纪录片。”

《机器中的幽灵》将于3月26日至28日在Kinema平台限时放映，并于今年秋季在PBS播出。影片试图唤醒公众：在技术狂欢背后，隐藏着一段从未被彻底清算的历史。

中文翻译：

与许多人一样，导演瓦莱丽·维奇在2024年OpenAI首次向公众发布Sora文生视频生成式人工智能模型时便被深深吸引。尽管她并不完全理解这项技术，却对其功能充满好奇，并注意到许多艺术家正组建线上社区分享自己的AI创作。与同好交流的期望将维奇引入了AI领域，但踏入其中后，她却震惊地发现这项技术频繁生成充满种族主义与性别歧视的图像。

生成式AI的迷魂汤泛着优生学的馊味

《机器幽灵》导演瓦莱丽·维奇希望人们理解种族科学如何塑造了此刻的科技景观。

更令维奇不安的是，身边这些AI狂热爱好者似乎毫不在意他们追捧的机器会未经明确指令就肆意喷吐充满仇恨与偏见的糟粕。这种荒诞境况迫使维奇停止了早期对生成式AI的探索，却也激发她创作了纪录片《机器幽灵》，旨在揭示奠定生成式AI存在基础的技术与思想源流。

《机器幽灵》并未聚焦于生成式AI加速主义者所鼓吹的、那些近在咫尺却遥不可及的社会效益，而是通过梳理技术史来阐释其当下运作方式的根源。维奇近期接受采访时表示，她希望记录生成式AI的起源，让人们看清当前行业炒作狂潮的实质。但首先，她必须穿透AI公司对核心概念的有意混淆。

"要使用'人工智能'这个词，我们首先得明白这玩意儿到底指什么，"维奇在视频通话中直言，"事实上它毫无意义，从来就是个营销术语。这个完全误导人且愚蠢的词汇已经形成了特定的文化内涵，我认为厘清用词及其含义至关重要。"

正如《机器幽灵》反复强调的，"人工智能"一词最初由计算机科学家约翰·麦卡锡于1956年创造，目的是为项目争取更多资金。但纪录片揭示这仅是时间轴上的一个节点，而这条时间轴实际始于维多利亚时代英格兰优生学的诞生。弗朗西斯·高尔顿不仅是查尔斯·达尔文的表弟，更是优生学的奠基人——这套种族主义且早被证伪的理论认为，通过系统清除"劣等"（即非白人）种族便能改良人类。

维奇在采访中强调，尽管高尔顿确实为学术界作出过有益贡献，但绝不能淡化其根深蒂固的白人至上主义思想对当时社会科学的影响。高尔顿与其优生学同僚卡尔·皮尔逊虽未直接参与早期计算机开发，但高尔顿在测量非洲与欧洲女性吸引力时采用的多维建模方法，深刻影响了皮尔逊开发逻辑回归等统计工具的思路，而逻辑回归正是现代机器学习的基础构件。

"难道我要在镜头前拥抱萨姆·奥尔特曼吗？那能算关于这项技术的真实纪录片吗？那是宣传片。"

高尔顿和皮尔逊助推了"不同种族存在可量化本质差异"观念的常态化。这种种族主义思维使高尔顿及其追随者相信人类智力可被测量，人脑运作方式酷似机器。维奇指出，正是这种观念跃迁为人工智能的虚幻概念铺平了公众认知道路。

"最初深入调查时最令我震惊的是，当你以纪录片导演或记者身份探究超级智能问题时，很快就会撞上种族科学的低矮门框——这门科学已彻底渗透进技术底层，"维奇解释道，这些概念无不"浸透着优生学思维"。

《机器幽灵》没有试图反驳"生成式AI因受训数据包含仇恨意识形态而输出糟粕"（即"垃圾进，垃圾出"）的观点，而是通过历史分析阐释为何开发这些技术的公司对解决当下问题如此漠然。这段历史背景帮助维奇理解了自身使用生成式AI的不快经历：当初在艺术家Slack群组试用早期版Sora时，她记得这个原本友善的社群，直到一位有色人种女性成员开始质疑模型总是将她照片生成的图像"漂白"时才发生变化。

"模型保留了她的辫子和衣着风格，但当她提示生成自己在艺术馆的形象时，程序将艺术馆理解为'白人空间'，"维奇描述道，"我的反应是'这太离谱了'，并试图向群组解释这是软件本身的问题。"然而她的发帖无人回应。"那个Slack群往常每条帖子都有几十个尖叫考拉表情回复，但这次什么都没有。"

维奇主动联系OpenAI，警告他们注意"所见输出内容中严重的种族主义、性别歧视和厌女倾向——比如女性角色生成两次后就开始长出额外乳房或跳电臀舞"。她原以为公司会将其视为需要紧急修复的关键缺陷，不料对方却漠然置之。

"得到的反馈大概是：'提这个真尴尬，我们无能为力。'"维奇回忆道。

此事点燃了维奇探究各类生成式智能体为何总表现出丑陋行径的决心。起初她认为，与相关技术白皮书作者进行Zoom通话难以构成引人入胜的纪录片，但随着从高尔顿的优生学统计工作到现代生成式AI机构之间清晰脉络的浮现，她的想法彻底改变。

《机器幽灵》汇集了AI研究员、历史学家与批判理论家的声音，有力论证了AI领域的每个侧面如何深受其历史渊源影响——这些科学领域本就是为支撑歧视性世界观而建立的。当被问及是否考虑直接采访片中批评的AI公司负责人时，维奇笑了。她说获得这种访问需要完成各种思想体操并作出妥协，这会使她的影片沦为生成式AI危害的共犯。

"这些人不会随便信任别人，这我理解。废话，我当然希望他们不信任我。我不需要他们出现在影片里，他们在媒体上说得够多了。难道我要在镜头前拥抱萨姆·奥尔特曼吗？那能算关于这项技术的真实纪录片吗？那是宣传片。"

《机器幽灵》将于3月26日至28日通过Kinema平台限时放映，并于今年秋季在PBS电视台播出。

更正声明（3月21日）：本文前版本误将"逻辑回归"写作"逻辑回归"，特此更正。

英文来源：

Like many people, director Valerie Veatch was intrigued when OpenAI first released its Sora text-to-video generative AI model to the public in 2024. Though she didn’t fully understand the technology, she was curious about what it could do, and she saw that other artists were building online communities to share their new AI creations. The hope of connecting with people drew Veatch into the AI space, but once she was there, she was shocked to see how often the technology would generate images dripping with racism and sexism.
The gen AI Kool-Aid tastes like eugenics
Ghost in the Machine director Valerie Veatch wants you to understand how race science has shaped this moment in tech.
The gen AI Kool-Aid tastes like eugenics
Ghost in the Machine director Valerie Veatch wants you to understand how race science has shaped this moment in tech.
Veatch was even more unsettled by the way her new AI-enthusiast peers did not seem to care that the machine they rallied around spewed out hateful, bigoted garbage without being explicitly prompted to do so. The bizarre situation drove Veatch away from her early experimentation with gen AI. But it also inspired her to make Ghost in the Machine, a new documentary about the technologies and schools of thought that laid the groundwork for gen AI’s existence.
Instead of focusing on the potential (if highly improbable) benefits to society that gen AI accelerationists swear are just around the corner, Ghost in the Machine explores the technology’s history to explain why it works the way it does now. When I recently spoke with Veatch about the film, she told me that she wanted to chronicle gen AI’s genesis to give people a clear view of the very intense cycle of industry hype we’re currently living through. First, however, she had to cut through AI firms’ purposeful obfuscation of the entire concept.
“In order to use the phrase ‘artificial intelligence,’ we have to know what the fuck that phrase means,” Veatch told me over a video call. “The truth is, it doesn’t mean anything; it’s a marketing term and always has been. It’s a completely misleading, stupid phrase that has taken on its own cultural meaning, and I think being really clear about the words we use and the meaning of those words is essential.”
As Ghost in the Machine repeatedly stresses, “artificial intelligence” was originally coined in 1956 by computer scientist John McCarthy when he was trying to secure more funding for his projects. But the documentary presents the term’s coinage as just one of many important points on a timeline that actually begins in Victorian-era England with the birth of eugenics. In addition to being Charles Darwin’s cousin, Francis Galton was the originator of eugenics — the racist and discredited belief that humanity can be improved through the systemic elimination of “inferior” (read: non-white) races.
While Galton definitely made some useful contributions to academia, in our interview, Veatch explained that it is important to not minimize the fact that his deeply held white supremacist beliefs informed the era’s social sciences. Galton and his fellow eugenicist / protegee Karl Pearson were not directly involved in the development of early computational machines. But Galton’s foundational work with multidimensional modeling — a technique he used while measuring the attractiveness of African and European women — shaped Pearson’s thinking as he developed statistical tools like logistic regression, which is one of the fundamental components of modern machine learning.
“Am I going to hug Sam Altman on camera? Is that a truthful film about this technology? That’s propaganda.”
Galton Pearson helped normalize the idea that people of various races were fundamentally different in quantifiable ways. This kind of racist thinking is what led to Galton and his peers believing that human intelligence could be measured, and that human brains function very much like machines. That jump, Veatch says, played a major role in selling the public on the fantastical idea of artificial intelligence.
“What was really surprising to me during my initial dive into all of this was how, when you look at the question of superintelligence as a documentarian or journalist, it doesn’t take long before you smack your forehead into the low doorframe of race science because it’s baked into this technology,” Veatch said, explaining that these concepts are “soaked” in eugenic thinking.
Rather than trying to disprove the idea that gen AI models produce hateful ideology because they’ve been trained on it (a concept commonly known as “GIGO” — garbage in, garbage out), Ghost in the Machine uses its historical analysis to explain why the companies building this technology seem so disinterested in addressing its present-day issues. This historical context helped Veatch make sense of some of her own troubling experiences with gen AI, back when she was playing around with an early version of Sora in an artists’ Slack. Veatch remembers the group as being a friendly, welcoming place right up until another member — a woman of color — began voicing concerns about the way the model whitewashed her every time she prompted it to generate images based on photos of herself.
“It kept her braids and it kept her fashion, but she was prompting herself into an art gallery, which the program understood to be a ‘white space,’” Veatch explained. “My reaction was ‘what the fuck,’ and I tried explaining to the group how this was really a problem with the software itself.” No one else in the group engaged with her post. “This was a Slack where, normally, there are always like dozens of screaming koala emoji reactions on every post. But this time, there was nothing.”
Veatch took it upon herself to get in contact with OpenAI directly to alert the company about “how racist, sexist, and misogynistic the outputs [she] was seeing were — outputs where women would start growing extra tits and twerking after like two rounds of generating a scene.” Veatch thought OpenAI would see this as a critical bug worth fixing before encouraging more people to adopt Sora into their lives; instead the company brushed her concerns aside.
“The feedback I got was basically, ‘This is very cringe to be bringing up; there’s nothing we can do to change it,’” Veatch recalled.
That situation lit a fire within Veatch to learn about why so many different forms of generative intelligence consistently behave in such ugly, troublesome ways. At first, she didn’t really think that having Zoom calls with the authors of white papers about the technology could be turned into a compelling documentary, but that changed as she began to see a clear line from Galton’s eugenic statistics work to modern gen AI outfits.
The voices featured in Ghost in the Machine — a blend of AI researchers, historians, and critical theorists — make a compelling case that basically every facet of the AI space has been profoundly influenced by its historical connections to fields of science built to support discriminatory world views. When I asked Veatch if she had ever been interested in speaking directly with the heads of the companies Ghost in the Machine takes to task, she laughed. Getting that kind of access, she said, would require her to go through all kinds of ideological gymnastics and make compromises that would make her film complicit in gen AI’s harms.
“There’s the idea, you know, these people won’t trust just anyone,” Veatch said. “Yeah, no shit, and I certainly hope they wouldn’t trust me. I don’t want them in the film and they already speak enough to the media. Am I going to hug Sam Altman on camera? Is that a truthful film about this technology? That’s propaganda.”
Ghost in the Machine will be available to stream via Kinema from March 26th to March 28th before it airs on PBS some time this fall.
Correction, March 21st: An earlier version of this story used the phrase “logical regression” when “logistic regression” is correct.