
俄罗斯利用人工智能创作了数百首支持战争的歌曲,这些歌曲歌颂本国士兵,同时将乌克兰人妖魔化。这些歌曲被发布到Spotify、YouTube等主流音乐平台上,用于传播宣传内容。
Russia uses AI to create hundreds of pro-war songs glorifying its soldiers and dehumanising Ukrainians, spreading propaganda on major music platforms like Spotify and YouTube.
自2024年起,随着新型人工智能音乐工具的出现,此类歌曲的制作变得更加快捷、廉价且简便,其数量也迅速增加;不过克里姆林宫是否直接参与其中尚不清楚。
These AI-generated songs surged after 2024, as new AI music tools made production faster, cheaper, and easier, with unclear direct Kremlin involvement.
尽管各大流媒体平台制定了相关政策,但在审核内容方面仍面临巨大困难。这些平台常常无法识别外语中的宣传言论与仇恨言论,导致这些歌曲得以广泛传播。
Streaming services struggle to moderate such content, often missing propaganda and hate speech in foreign languages despite some policies, allowing widespread dissemination of these songs.
旧金山讯——在针对乌克兰的宣传战中,俄罗斯又多了一件“武器”:由人工智能生成的歌曲。
AI generated SAN FRANCISCO – Russia has a new weapon in its propaganda war against Ukraine: songs generated by artificial intelligence.
目前,已有数百首此类歌曲出现在全球最受欢迎的音乐流媒体平台上。这些歌曲类型多样,既有伤感的民谣,也有节奏强烈的重金属乐。它们均是由众多人工智能音乐生成工具创作而成,其中不少工具均来自美国。
Hundreds of them – from maudlin ballads to grinding thrash metal – have sprouted on the world’s most popular music streaming platforms, created using a growing number of AI music generators, many based in the United States.
这些歌曲歌颂俄罗斯士兵,同时以种族歧视性语言贬低乌克兰人。歌曲中将战场描绘成地狱般的存在,但总体上却将普京总统于2022年发动的入侵行动描述为必要的自卫行为。
The songs glorify Russian soldiers and dehumanise Ukrainians, sometimes with ethnic slurs. They depict the front lines as a hellscape but generally portray President Vladimir Putin’s invasion in 2022 as a necessary defence.
其中最为流行的歌曲之一是《火炮的力量》。这是一首用俄语演唱的摇滚曲,歌颂了诸如“克拉斯诺波尔”和“康乃馨”之类的火炮系统。
One of the most popular songs, Power of Artillery, is a churning rock paean in Russian to artillery systems with names like Krasnopol and Carnation.
“火炮的力量/无论昼夜都会突袭你/战争女神降临,上帝也救不了你/你无法逃脱她的怒火/只能赶紧祈祷罢了。”
Power of Artillery She will catch you off guard night and day The goddess of war, and God will not help you You won’t be able to escape her wrath You’ll have to remember your prayers quickly The songs have been posted and played millions of times on the world’s most popular streaming platforms, including Spotify, Apple, YouTube and Deezer, according to research published by Terrecon, a nonprofit organisation focused on threat intelligence in Ukraine that supports the government there.
据致力于为乌克兰政府提供威胁情报的非营利组织Terrecon发布的研究显示,这些歌曲已在Spotify、Apple Music、YouTube及Deezer等全球主流流媒体平台上被播放了数百万次。如今,那些改变了战争形态的技术同样也在重塑着战时宣传手段。
The technological advances that have reshaped warfare are also reshaping wartime propaganda.
Terrecon创始人达里亚·韦尔比茨卡(Daria Verbytska)在接受电话采访时表示:“他们找到了制造更便宜的无人机并大规模使用它们。”他指的是俄罗斯人。“音乐就像这样--在人工智能的帮助下进行大规模的宣传。”俄罗斯政府在宣传这些歌曲方面所扮演的角色尚不清楚,但克里姆林宫竭尽全力为战争培养社会和文化支持。
“They have found new ways to create cheaper drones and use them on a massive scale,” Daria Verbytska, a founder of Terrecon, said in a telephone interview, referring to the Russians. “It’s like that with music – creating massive propaganda with the help of AI.”The extent of the Russian government’s role in propagating the songs is not clear, but the Kremlin has gone to great lengths to cultivate social and cultural support for the war.
据俄罗斯新闻报道,上周,莫斯科互联网发展研究所的一名所长在一次演讲中列举了该研究所所推动的350多个有关战争的艺术项目。
Last week, a director of the Internet Development Institute, an organisation in Moscow that works closely with the Kremlin to promote patriotic values, cited in a presentation more than 350 artistic projects about the war the institute has promoted, according to news reports in Russia.
Terrecon总共对人工智能生成的981首歌曲进行了编目,这些歌曲包含该组织定义的亲俄战争宣传内容。在那些可以确定发布日期的歌曲中,绝大多数出现在2024年之后,当时Suno和Udio等人工智能歌曲生成器进入市场。
In all, Terrecon catalogued 981 songs generated by AI that contained what the group defined as pro-Russian war propaganda. Among those whose release date could be established, a vast majority appeared after 2024, when AI song generators including Suno and Udio entered the market.
“与词曲作者和制作一起创作一首真正的歌曲需要大量资源,”韦尔比茨卡说。“通过这些人工智能生成器更容易、更快、更便宜。”歌曲数量激增的部分原因是音乐生成器和流行流媒体服务的规则不一致,这些服务一直在努力应对音乐中人工智能日益使用的问题。只有Deezer和YouTube等部分服务需要标记识别人工智能内容。
“It takes a lot of resources to create a real song with songwriters, production,” Verbytska said. “It’s easier and faster and cheaper to make it through these AI generators.”The surge in songs has also been enabled partly because of inconsistent rules by music generators and popular streaming services that have grappled with the increasing use of AI in music. Only some services, including Deezer and YouTube, require labeling that identifies AI content.
Spotify在回答问题时表示,该公司于9月份制定了一项政策,对人工智能生成的歌曲进行标记,并且不向用户推荐它们。该公司在《纽约时报》分享的歌曲中添加了标签作为例子。
Spotify enacted a policy in September to label songs generated by AI and not to recommend them to users, the company said in response to questions. The company added labels to songs The New York Times shared as examples.
主要的流媒体网站都有反对仇恨言论和暴力内容的政策,但很少有明确限制宣传的。这些网站也并不总是能捕捉到外语中的惯用表达或诽谤。
The major streaming sites have policies against hate speech and violent content, but few explicitly restrict propaganda. The sites also do not always capture idiomatic expressions or slurs in foreign languages.
特雷康的研究人员指出,即便这些平台会将某些具有危害性的词汇和短语列为禁用内容,但它们未必会限制那些常见于各类歌曲中的战争主题。
Even when the platforms ban specific words and phrases as harmful, they do not necessarily restrict the martial themes that characterise many of the songs, Terrecon’s researchers argued.
研究人员还测试了多款人工智能音乐生成器:美国的Suno、Udio及Google Flow Music,以及总部位于新加坡的Mureka。他们用俄语和英语输入相关提示词,试图生成宣扬俄罗斯宣传内容的歌曲,其中不乏带有去人性化描述及煽动暴力倾向的内容。
The researchers also tested AI music generators – Suno, Udio and Google Flow Music in the United States, and Mureka, based in Singapore – by writing prompts in Russian and English for songs promoting Russian propaganda, including dehumanising content and calls for violence.
这些生成器拒绝执行部分提示词,但研究人员仍成功创作出与他们此前整理过的歌曲风格相近的作品,并将其上传至YouTube Music和SoundCloud平台。
The generators refused to carry out some of the prompts, but the researchers were able to create songs similar to the ones they had catalogued and to upload them on YouTube Music and SoundCloud.
特雷康的报告指出:“这些内容审核系统主要对特定的禁用词汇作出反应,而非考量提示词的整体含义。”
“The moderation systems mainly responded to specific banned words rather than the overall meaning of the prompts,” Terrecon’s report found.
Deezer的发言人耶斯珀·温德尔表示,该平台会逐案审查可能违反规定的内容。SoundCloud的发言人萨德·阿约德莱称,公司会对用户举报的内容进行审核,并强调“SoundCloud绝不容许任何煽动或美化暴力的内容存在”。
Deezer’s spokesperson, Jesper Wendel, said the platform reviewed potentially violative content case by case. A spokesperson for SoundCloud, Sade Ayodele, said the company reviewed reported content, adding that “there is no place on SoundCloud for content that incites or glorifies violence”.
隶属于谷歌的YouTube在声明中称,平台要求创作者必须披露其是否使用了人工智能技术,且所有内容都须符合反仇恨言论的相关政策。在《纽约时报》提出询问后,YouTube还给特雷康识别出的多首歌曲添加了标签,并下架了特雷康为测试目的而创作的歌曲。
YouTube, which is owned by Google, said in a statement that it required creators to disclose the use of AI and required all content to comply with policies against hate speech. YouTube also labelled several songs identified by Terrecon and removed one that Terrecon had created as part of its test after the Times’ inquiry.
苹果公司、Suno、Udio及Mureka均未回应置评请求。因从事宣传活动而遭到欧洲及其他多国政府实施经济制裁的莫斯科互联网发展研究所也未作出回应。
Apple, Suno, Udio and Mureka did not respond to requests for comment. Nor did the Internet Development Institute in Moscow, which is subject to economic sanctions imposed by governments in Europe and elsewhere for its propaganda.
目前尚不清楚这些歌曲的受欢迎程度。其中一首以火炮武器为主题的歌曲配有视频剪辑后发布在YouTube上,两年来观看量已突破200万次。
It is unclear how popular the songs are. The song about artillery weapons, posted on YouTube accompanied by a video montage, has amassed more than 2 million views in two years.
这首歌在Suno上创作,由一位名叫谢尔盖·瓦西金(Sergei Vaseykin)的用户发布,他在俄罗斯诗歌网站上的传记称他出生在秋明附近的一个西伯利亚村庄。瓦西金发表了许多对俄罗斯战争努力的颂歌(其中一首是对曲棍球明星亚历克斯·奥维奇金的颂歌),但他没有回应置评请求。
The song, created on Suno, was posted by a user called Sergei Vaseykin, whose biography on a Russian poetry site says he was born in a Siberian village near Tyumen. Vaseykin, who has posted numerous odes to the Russian war effort (and one to hockey star Alex Ovechkin), did not respond to a request for comment.
尼娜·L说,这些歌曲反映了自俄罗斯入侵乌克兰以来俄罗斯政治和文化日益军事化。赫鲁晓夫是纽约市新学院的国际事务教授,长期研究俄罗斯的宣传活动。
The songs reflect a growing militarisation in Russian politics and culture since Russia invaded Ukraine, said Nina L. Khrushcheva, a professor of international affairs at the New School in New York City who has long studied Russian propaganda efforts.
他们还利用了苏联通过音乐促进爱国主义的传统,特别是在第二次世界大战期间和之后,被称为俄罗斯的伟大卫国战争。
They also tap into the Soviet Union’s tradition of promoting patriotism through music, especially during and after World War II, known as the Great Patriotic War in Russia.
赫鲁晓夫指出,这些歌曲-其中许多仍然为大多数俄罗斯人所熟悉-可能被苏联政府用作宣传,但反映了士兵们在一场几乎触及该国每个人的生存战争中的真实经历。相比之下,这些人工智能创造物是“一个又一个的拟像,”她说。
Khrushcheva noted that those songs – many of them still familiar to most Russians – might have been used as propaganda by the Soviet government but reflected genuine experiences of soldiers’ fighting an existential war that touched virtually everyone in the country. By contrast, these AI creations are “a simulacrum upon simulacrum upon simulacrum,” she said.
“这很容易咀嚼-你不必感觉到任何东西,”她补充说。“这有点像沙发爱国主义。”纽约时报本文最初发表在《纽约时报》上。
“It’s easy to chew on – you don’t have to feel anything,” she added. “It’s sort of couch patriotism.”NYTIMES This article originally appeared in The New York Times.