Here’s Why You Shouldn’t Use AI to Vote

Save this storySave this storyAs Americans go to the polls today, many will have had their decision about who to vote for informed by an artificial intelligence–powered chatbot.Around 50 percent of recently surveyed US adults have at one time used an AI chatbot, and 8 percent say they have used these tools ahead of the midterms to find information on candidates and policies, according to Pew Research.

随着美国人今天前往投票站，许多人的投票决定受到了人工智能聊天机器人的影响。根据皮尤研究中心的数据，近期接受调查的美国成年人中，约有 50% 的人曾使用过人工智能聊天机器人，8% 的人表示他们在中期选举前使用过这些工具来查找有关候选人和政策的信息。

But a lot of information from these chatbots is out of date and has been hallucinated or sourced from partisan websites.Even supposedly nonpartisan chatbots like OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini have been found wanting when it comes to questions about the midterms.

然而，这些聊天机器人提供的许多信息已经过时，要么是凭空捏造的，要么来源于党派网站。即使是像 OpenAI 的 ChatGPT、Anthropic 的 Claude 和 Google 的 Gemini 这样号称中立的聊天机器人，在涉及中期选举的问题时也被发现存在不足。

Just last week, Google’s AI Mode refused to provide the results of the 2020 elections, and the company’s Gemini chatbot said the deadline to request a mail-in ballot in Pennsylvania was November 3, rather than the real deadline of October 27, according to research by Forum AI.According to the Forum AI research, almost 40 percent of election answers given by all the mainstream chatbots contained at least one factual error.

根据 Forum AI 的研究，就在上周，Google 的 AI 模式拒绝提供 2020 年选举的结果，而该公司的 Gemini 聊天机器人则称宾夕法尼亚州申请邮寄选票的截止日期是 11 月 3 日，而非实际的 10 月 27 日。Forum AI 的研究还显示，所有主流聊天机器人给出的选举相关答案中，近 40% 至少包含一个事实错误。

In August, when Forum AI’s researchers asked why someone should vote for North Carolina congressman Don Davis, Claude gave a response that conflated the lawmaker with a “former school superintendent” of the same name.Google and Anthropic did not respond to requests for comment. Open AI directed WIRED to a blog outlining the company’s approach to elections in 2026.

今年 8 月，当 Forum AI 的研究人员询问为何应该投票给北卡罗来纳州国会议员唐·戴维斯（Don Davis）时，Claude 给出的回答将这位议员与一名同名的“前学校主管”混为一谈。Google 和 Anthropic 没有回应置评请求。OpenAI 则引导《连线》（WIRED）查阅了一篇概述该公司 2026 年选举方针的博客文章。

Part of the problem is that chatbots are reliant on an increasingly problematic pool of online information. Research by NewsGuard found that mainstream chatbots cite so-called “pink slime” websites nearly 50 percent of the time when asked about candidates and policies related to the midterms. Pink slime sites present as mainstream news websites but are in fact funded by political or ideological groups and individuals.

问题的一部分在于，聊天机器人依赖于一个日益充满问题的在线信息池。NewsGuard 的研究发现，当被问及与中期选举相关的候选人和政策时，主流聊天机器人引用所谓的“粉红泥浆”（pink slime）网站的频率接近 50%。这些“粉红泥浆”网站伪装成主流新闻网站，但实际上是由政治或意识形态团体及个人资助的。

Ahead of the midterms, many more of these sites have sprung up, and chatbots are paying attention. The worst offender, according to NewsGuard, was OpenAI’s ChatGPT, which cited the pink slime sites in more than 70 percent of searches, while xAI’s Grok chatbot performed best, citing pink slime sites in 30 percent of cases.

在中期选举前，这类网站大量涌现，而聊天机器人也对此予以“关注”。据 NewsGuard 称，表现最差的是 OpenAI 的 ChatGPT，其在超过 70% 的搜索中引用了这些“粉红泥浆”网站；而 xAI 的 Grok 聊天机器人表现最好，引用此类网站的比例为 30%。

“I don’t think that most Americans are prepared to navigate an election in the age of consumer AI,” Nina Jankowicz, executive director of the American Sunlight Project and a former disinformation czar during the Biden administration, tells WIRED. “One of the problems is that executives and companies have really overpromised on what AI is capable of. It has to ingest this information from somewhere, and often those sources are not very good. And it’s predictive, so if it doesn’t know the answer, it does frequently make things up.”

“我不认为大多数美国人已经准备好在消费级人工智能时代应对选举，”美国阳光项目（American Sunlight Project）执行董事、拜登政府时期的前虚假信息主管尼娜·扬科维奇（Nina Jankowicz）告诉《连线》杂志。“问题之一在于，高管和公司对人工智能的能力过度承诺了。它必须从某个地方获取信息，而这些来源往往并不理想。而且它是预测性的，所以如果它不知道答案，它经常会编造事实。”