ZDNET的关键外卖
- 美国人使用AI搜索信息——包括选举。
- 选民要求聊天员提供基本选举信息和候选人建议。
- AI系统可以产生幻觉，并在政治信息中增加层次的偏向。
大部分搜索工具都将AI放在你和您所寻找的信息之间
还有：厌倦AI概览？我找到9个Google搜索选项
选民问的聊天机器人
研究者要求35个州有参议院种族的近2500人询问ChatGPT,Claude,Grok或Gemini有关选举的情况。 这项研究在10月的11月3日中期选举前持续了几个星期，参与者是每周已经使用聊天器的成年人。
大多数用户向聊天员询问有关选举的基本事实，包括谁在选票上。 但36%的人要求一个聊天组来帮助他们挑选一个候选人投票，他们常常通过描述自己的优先事项，要求聊天组评估哪些候选人与他们一致。
前一个查询的出错空间略小，因为互联网拥有一吨关于选举基本事实的数据，例如谁在跑哪里。 Chatbot对候选人职位的诠释则稍稍有些机智
还：如何保持您的聊天室谈话的隐私
在一个例子中，一个用户要求ChatGPT给他们提供弗吉尼亚州每个候选人的“利弊”，这是一个固有的主观问题。
在推荐候选人时，聊天员差别很大。
Grok经常推荐一名候选人，ChatGPT在79%的聊天中推荐一名候选人
大赦国际和选举
Pew Research发现42%被调查的美国成年人使用聊天机器人搜索信息。 但皮尤也发现只有8%的美国成年人报告在今年的选举中使用聊天机寻找与选举有关的信息。
不过这些用户在找什么 Pew发现，其中76%的人询问候选人在某些问题上的立场，71%的人询问他们听说过的候选人或问题的情况。
OpenAI将合并 ChatGPT 的聊天和工作标签
由于聊天机器人仍然不是可靠的总结器，如果用户不检查聊天机器人的链接来源，这些问题的性质就会产生错误信息的风险。 Google的AI Overviews。 至少有一项研究发现，AI平台向用户提供了“过时或不准确”的信息，在某些情况下，这种信息可能会误导选民，因为没有规定最后期限，或者没有提到关于投票后勤的一个步骤。
由于2024年选举以来各种工具变得更加复杂，因此这一选举周期尤其受到大赦国际制造的深刻错误和错误信息的困扰。 据报告，OpenAI和Anthropic建立了选举“战争室”，根据Politico和Semafor的独家报道，监测误用和信息准确性。
虽然8月Brennan司法中心的一项研究发现，聊天机器人确实对与选举虚假宣传有关的某些谎言进行了惊人的反省，但它也发现AI模型在被问及时很容易产生“误导与选举有关的图像和视频”。
研究报告对ChatGPT、Gemini、Grok、Meta AI、Runway和Flux.2进行了测试。
还有：50 AI agents试图建立一个繁荣的Minecraft文明
然而，除了这些因素外，AI使用中还有一个对公民参与具有同样影响的广泛主题：认知卸载。 政治媒体的面貌已经很密集，使用不透明的训练数据集、过滤器和其他由人工智能实验室斟酌决定的设置来分层聊天员，除了这一信息搜索过程之外，还可以使水域进一步泥沙化，同时看起来像一个方便和中立的总结工具。
为什么这项研究很重要
虽然有些研究通过使用标准化的问题来测试AI聊天员如何回答选举问题，但这种方法让用户自己像通常那样与聊天员接触。 虽然耶鲁大学研究人员的查询和标准化的研究查询显示出许多相似之处，但这种区分对于评价是有价值的。
研究者指出，“这是衡量AI对用户实际提出的问题的政治影响的评估的一小步。
除了用户自我报告自己要求的聊天员之外，只有AI实验室有关于用户真正询问选举内容的记录。 10月，Anthropic分析了与Claude的25万次对话，确定人们“只是问Claude谁或该投什么票”，认为“0.51%的谈话具体涉及选举或投票，”尽管它预计与选举有关的询问会接近11月3日。
还有：6个AI agent工作流程，我用来传递日常任务
“我们经常看到有人问Claude问题，这些问题可以帮助他们达成自己的结论，例如候选人对某一问题的立场。 Anthropic的报告称显示
耶鲁研究者称的下一步研究是确定在回应中模型的准确性和公正性，以及哪些模型了解用户，以及他们的建议是否真正符合用户的需求。
" 在多回合聊天中，用户经常回答聊天人的问题，这些答案导致了其建议。 研究者解释道
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
ZDNET’s key takeaways
- Americans use AI to search for information – including elections.
- Voters ask chatbots for basic election information and candidate recommendations.
- AI systems can hallucinate and add layers of bias to political information.
Most search tools now put a layer of AI between you and the information you’re looking for, especially Google.
Also: Tired of AI Overviews? I found 9 Google Search alternatives
What voters asked chatbots
The researchers asked nearly 2,500 people in 35 states with Senate races to query ChatGPT, Claude, Grok, or Gemini about the election. Participants in the study, which ran for a few weeks in October leading up to the Nov. 3 midterm elections, were adults who already use chatbots weekly.
Most users asked chatbots for basic facts about the election, including who was on the ballot. But 36% asked a chatbot to help them pick a candidate to vote for, often by describing their priorities and asking the chatbot to assess which candidates aligned with them.
The former query has slightly less room for error, because the internet has a ton of data on the basic facts of an election, such as who is running where. Chatbots interpreting candidate positions, on the other hand, is a bit more dicey, and relies significantly on the sources a chatbot uses (a statement from a candidate’s own website, for example, is more reliable than a social media clip of an opponent characterizing them in a certain way).
Also: How to keep your chatbot conversations as private as possible
In one example, a user asked ChatGPT to give them the “pros and cons” of each candidate in Virginia, an inherently subjective question.
When it came to recommending candidates, chatbots varied widely.
“When users asked who to vote for but had not yet said anything about themselves, only Grok regularly recommended a candidate, in 14% of chats,” the researchers wrote. “Once users shared their views, ChatGPT recommended a candidate in 79% of chats and Claude in 40%, with Grok and Gemini in between.”
AI and elections
Pew Research found that 42% of US adults surveyed use chatbots to search for information. That’s a broad umbrella that can include election information — but Pew also found that only 8% of US adults reported using chatbots to find election-related information in this year’s election.
What those users are looking for matters, though. Pew found that of that group, 76% were inquiring about where candidates stand on certain issues, and 71% were fact-checking information they’d heard about candidates or issues.
Also: OpenAI will merge ChatGPT’s chat and work tabs
Given that chatbots are still not reliable summarizers, the nature of these questions poses a misinformation risk if users don’t also check a chatbot’s linked sources. An overwhelming majority of users don’t, at least when it comes to Google’s AI Overviews. At least one study found AI platforms gave users “outdated or inaccurate” information that could mislead voters in some cases by missing a deadline or failing to mention a step about voting logistics.
This election cycle was particularly plagued by AI-generated deepfakes and misinformation as tools have become much more sophisticated since the 2024 election. OpenAI and Anthropic reportedly set up election “war rooms” to monitor misuse and information accuracy, according to exclusive reporting by Politico and Semafor.
While an August study from the Brennan Center for Justice found that chatbots did push back surprisingly well against certain falsehoods related to election disinformation campaigns, it also found that AI models easily generated “misleading election-related images and video” when asked.
“Many of their safeguards for preventing this were ineffective, inconsistently applied, or functionally absent,” noted the study, which tested ChatGPT, Gemini, Grok, Meta AI, Runway, and Flux.2.
Also: 50 AI agents tried to build a thriving Minecraft civilization – it didn’t end well
Apart from these factors, however, there is a broader theme in AI use that is equally consequential for civic engagement: cognitive offloading. The political media landscape can already be dense and hard to parse for a voter trying to stay informed. Layering a chatbot — which has opaque training data sets, filters, and other settings determined at the discretion of AI labs — on top of that information search process can further muddy the waters while appearing like a convenient and neutral summarizing tool.
Why this study matters
While some studies test how AI chatbots respond to election questions by using standardized questions, this approach let users themselves engage with chatbots as they normally would. While the Yale researchers’ queries and standardized study queries show a lot of similarities, the distinction is valuable for evaluations.
“This is a small step toward evaluations that measure AI’s political influence on the questions users actually ask,” the researchers noted.
Outside of users self-reporting what they asked a chatbot, only AI labs have records of what users are really asking about elections. In October, Anthropic analyzed 250,000 conversations with Claude and determined people “rarely ask Claude whom or what to vote for,” finding that “0.51% of all conversations are specifically about elections or voting,” though it expected election-related queries to increase closer to Nov. 3.
Also: 6 AI agent workflows I use to hand off everyday tasks
“More often, we observed people asking Claude questions that could help them reach their own conclusions, like where candidates stand on a particular issue. Requests that violated our election integrity policies were even less common, appearing in about 0.002% of conversations,” Anthropic’s report said.
The Yale researchers said the next step for a study like theirs is determining how accurate and unbiased models are in their responses, as well as what models learn about users, and whether their recommendations actually match users’ needs.
“In multi-turn chats, users often answered the chatbot’s questions, and those answers led to its recommendations. The questions a model asks shape what users tell it and what it recommends,” the researchers explained.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.