How disinformation works—and how to counter it

Did you know that the wildfires which ravaged Hawaii last summer were started by a secret “weather weapon” being tested by America’s armed forces, and that American ngos were spreading dengue fever in Africa? That Olena Zelenska, Ukraine’s first lady, went on a $1.1m shopping spree on Manhattan’s Fifth Avenue? Or that Narendra Modi, India’s prime minister, has been endorsed in a new song by Mahendra Kapoor, an Indian singer who died in 2008?
您是否知道去年夏天肆虐夏威夷的野火是由美国武装部队测试的秘密“气象武器”引发的，而美国非政府组织正在非洲传播登革热？乌克兰第一夫人奥琳娜·泽连斯卡 (Olena Zelenska) 在曼哈顿第五大道疯狂购物 110 万美元？或者印度总理纳伦德拉·莫迪 (Narendra Modi) 在一首新歌中得到了 2008 年去世的印度歌手马亨德拉·卡普尔 (Mahendra Kapoor) 的认可？

These stories are, of course, all bogus. They are examples of disinformation: falsehoods that are intended to deceive. Such tall tales are being spread around the world by increasingly sophisticated campaigns. Whizzy artificial-intelligence (AI) tools and intricate networks of social-media accounts are being used to make and share eerily convincing photos, video and audio, confusing fact with fiction. In a year when half the world is holding elections, this is fuelling fears that technology will make disinformation impossible to fight, fatally undermining democracy. How worried should you be?
当然，这些故事都是假的。它们是虚假信息的例子：旨在欺骗的谎言。这些荒唐的故事正通过日益复杂的活动而在世界各地传播。巧妙的人工智能（ai）工具和错综复杂的社交媒体账户网络被用来制作和分享令人难以置信的令人信服的照片、视频和音频，混淆了事实与虚构。今年，世界上一半的国家都在举行选举，这加剧了人们的担忧，即技术将使虚假信息变得无法对抗，从而致命地破坏民主。你应该有多担心？

Disinformation has existed for as long as there have been two sides to an argument. Rameses II did not win the battle of Kadesh in 1274bc. It was, at best, a draw; but you would never guess that from the monuments the pharaoh built in honour of his triumph. Julius Caesar’s account of the Gallic wars is as much political propaganda as historical narrative. The age of print was no better. During the English civil war of the 1640s, press controls collapsed, prompting much concern about “scurrilous and fictitious pamphlets”.
只要争论双方存在，虚假信息就一直存在。拉美西斯二世在公元前1274年的卡迭石战役中并未获胜。最好的情况下，这只是一场平局；但你永远不会从法老为纪念他的胜利而建造的纪念碑中猜到这一点。尤利乌斯·凯撒对高卢战争的描述既是历史叙述，也是政治宣传。印刷时代也好不到哪儿去。 1640 年代的英国内战期间，新闻控制崩溃，引发了人们对“粗俗和虚构的小册子”的担忧。

The internet has made the problem much worse. False information can be distributed at low cost on social media; AI also makes it cheap to produce. Much about disinformation is murky. But in a special Science & technology section, we trace the complex ways in which it is seeded and spread via networks of social-media accounts and websites. Russia’s campaign against Ms Zelenska, for instance, began as a video on YouTube, before passing through African fake-news websites and being boosted by other sites and social-media accounts. The result is a deceptive veneer of plausibility.
互联网使问题变得更加严重。虚假信息可以在社交媒体上以低成本传播；人工智能还使得生产成本低廉。关于虚假信息的很多内容都是模糊的。但在一个特殊的科学与技术部分，我们追踪了它通过社交媒体帐户和网站网络传播和传播的复杂方式。例如，俄罗斯针对泽连斯卡女士的运动最初是在 YouTube 上发布的一段视频，然后通过非洲假新闻网站传播，并得到其他网站和社交媒体帐户的推动。结果是表面上看似合理，但具有欺骗性。

Spreader accounts build a following by posting about football or the British royal family, gaining trust before mixing in disinformation. Much of the research on disinformation tends to focus on a specific topic on a particular platform in a single language. But it turns out that most campaigns work in similar ways. The techniques used by Chinese disinformation operations to bad-mouth South Korean firms in the Middle East, for instance, look remarkably like those used in Russian-led efforts to spread untruths around Europe.
传播者账户通过发布有关足球或英国王室的信息来吸引追随者，在掺入虚假信息之前获得信任。许多关于虚假信息的研究往往集中在特定平台上使用单一语言的特定主题。但事实证明，大多数活动的运作方式都是相似的。例如，中国在中东的虚假信息行动中诽谤韩国公司所使用的技术，与俄罗斯领导的在欧洲各地传播谎言的行动所使用的技术非常相似。

The goal of many operations is not necessarily to make you support one political party over another. Sometimes the aim is simply to pollute the public sphere, or sow distrust in media, governments, and the very idea that truth is knowable. Hence the Chinese fables about weather weapons in Hawaii, or Russia’s bid to conceal its role in shooting down a Malaysian airliner by promoting several competing narratives.
许多行动的目标不一定是让您支持一个政党而不是另一个政党。有时，其目的只是为了污染公共领域，或者在媒体、政府以及真相可知这一观念上播下不信任的种子。因此，中国关于夏威夷气象武器的传说，或者俄罗斯试图通过宣传几种相互竞争的叙述来掩盖其在击落马来西亚客机中所扮演的角色。

All this prompts concerns that technology, by making disinformation unbeatable, will threaten democracy itself. But there are ways to minimise and manage the problem.
所有这些都引发了人们的担忧，即技术使虚假信息变得无与伦比，从而威胁民主本身。但有一些方法可以最小化和管理这个问题。

Encouragingly, technology is as much a force for good as it is for evil. Although AI makes the production of disinformation much cheaper, it can also help with tracking and detection. Even as campaigns become more sophisticated, with each spreader account varying its language just enough to be plausible, AI models can detect narratives that seem similar. Other tools can spot dodgy videos by identifying faked audio, or by looking for signs of real heartbeats, as revealed by subtle variations in the skin colour of people’s foreheads.
令人鼓舞的是，技术既是一种行善的力量，也是一种作恶的力量。尽管人工智能使虚假信息的生产成本大大降低，但它也可以帮助跟踪和检测。即使活动变得更加复杂，每个传播者帐户的语言都各不相同，足以显得可信，人工智能模型也可以检测到看似相似的叙述。其他工具可以通过识别伪造的音频或寻找真实心跳的迹象（如人们额头肤色的细微变化所揭示的那样）来发现可疑视频。

Better co-ordination can help, too. In some ways the situation is analogous to climate science in the 1980s, when meteorologists, oceanographers and earth scientists could tell something was happening, but could each see only part of the picture. Only when they were brought together did the full extent of climate change become clear. Similarly, academic researchers, ngos, tech firms, media outlets and government agencies cannot tackle the problem of disinformation on their own. With co-ordination, they can share information and spot patterns, enabling tech firms to label, muzzle or remove deceptive content. For instance, Facebook’s parent, Meta, shut down a disinformation operation in Ukraine in late 2023 after receiving a tip-off from Google.
更好的协调也有帮助。在某些方面，这种情况类似于 20 世纪 80 年代的气候科学，当时气象学家、海洋学家和地球科学家可以判断正在发生的事情，但每个人只能看到图片的一部分。只有将它们放在一起，气候变化的全部范围才变得清晰起来。同样，学术研究人员、非政府组织、科技公司、媒体和政府机构也无法独自解决虚假信息问题。通过协调，他们可以共享信息并发现模式，使科技公司能够标记、限制或删除欺骗性内容。例如，Facebook 的母公司 Meta 在收到谷歌的举报后，于 2023 年底关闭了在乌克兰的虚假信息业务。

But deeper understanding also requires better access to data. In today’s world of algorithmic feeds, only tech companies can tell who is reading what. Under American law these firms are not obliged to share data with researchers. But Europe’s new Digital Services Act mandates data-sharing, and could be a template for other countries. Companies worried about sharing secret information could let researchers send in programs to be run, rather than sending out data for analysis.
但更深入的理解还需要更好地访问数据。在当今算法提要的世界中，只有科技公司才能知道谁在阅读什么。根据美国法律，这些公司没有义务与研究人员共享数据。但欧洲新的数字服务法案强制要求数据共享，并且可以成为其他国家的模板。担心共享秘密信息的公司可以让研究人员发送要运行的程序，而不是发送数据进行分析。

Such co-ordination will be easier to pull off in some places than others. Taiwan, for instance, is considered the gold standard for dealing with disinformation campaigns. It helps that the country is small, trust in the government is high and the threat from a hostile foreign power is clear. Other countries have fewer resources and weaker trust in institutions. In America, alas, polarised politics means that co-ordinated attempts to combat disinformation have been depicted as evidence of a vast left-wing conspiracy to silence right-wing voices online.

One person’s fact...

The dangers of disinformation need to be taken seriously and studied closely. But bear in mind that they are still uncertain. So far there is little evidence that disinformation alone can sway the outcome of an election. For centuries there have been people who have peddled false information, and people who have wanted to believe them. Yet societies have usually found ways to cope. Disinformation may be taking on a new, more sophisticated shape today. But it has not yet revealed itself as an unprecedented and unassailable threat. ■

For subscribers only: to see how we design each week’s cover, sign up to our weekly Cover Story newsletter.