11-19 更新 实时
首页/ 深度/ 正文

问答:AI,今年我该投给谁?

深度 来源:MIT News AI 2026-10-31 03:00 0 阅读 机器翻译
问答:AI,今年我该投给谁?

人们每天都在向人工智能聊天机器人抛出大量问题,但在政治以及即将到来的2026年美国中期选举方面,AI 平台正在扮演什么角色?AI 聊天机器人能否左右选民,让他们改换党派或重新审视自己的政治信念?

MIT 的研究人员构建了一个名为 LLM Election Observatory 的新仪表板,用于实时考察主流 AI 聊天机器人如何回答有关2026年中期选举的问题。该仪表板背后的团队由 Adam Berinsky 教授、Chara Podimata 助理教授和 Charles Stewart III 教授领导,他们在研究投票与选举管理、AI 与虚假信息,以及 AI 系统如何影响人类决策方面拥有丰富经验。为了更好地理解 AI 正在如何塑造这场选举,他们设计了这个仪表板,用来模拟全美各地的人如何与聊天机器人交流。

团队追踪不同 AI 系统如何回答同一组固定问题,这些问题涵盖一系列话题,包括选举认证、热点议题、结果预测、新闻报道等。每隔几天,LLM Election Observatory 团队就会再次把这些 问题输入聊天机器人,监测随着选举临近,它们的回答如何因用户所在地、政党归属、身份及其他因素而变化。这项工作的目标不仅是了解 AI 系统如何传播信息,还要更好地理解它们如何塑造全美各地的政治话语以及选民对政治议题的理解。

MIT News 与 Podimata 和 Berinsky 进行了交谈,聊了聊他们参与 LLM Election Observatory 的工作、AI 可能如何影响中期选举,以及选民在与聊天机器人互动时应注意什么。Podimata 是 1942 届职业发展讲席教授,MIT 斯隆管理学院运筹学与统计学助理教授,也是信息与决策系统实验室成员;Berinsky 是三井讲席教授、政治学系教授。

问:这个项目的灵感来自哪里?你们希望通过 LLM Election Observatory 了解到什么?

Podimata:人们越来越多地使用 AI 聊天机器人来查找和理解政治与选举相关信息。这使得了解这些系统实际提供什么信息变得很重要,而且不只是某一个时间点的信息,而是整个选举周期内的信息。这尤其具有挑战性,因为模型本身在不断更新,可供其使用的政治信息输入本身也在变化,而且它们的回答可能取决于用户如何提问,或用户提供了关于自身的哪些信息。

一个基本问题是,我们目前对这样的信息环境没有良好的历史记录。如果某个 AI 系统今天对一场选举给出某种回答,几周后却给出不同的回答,那么事后就很难还原选民在每个时间点本会看到什么。我们项目的一个重要目标,就是随着这一环境的发展对其进行系统性记录,并把由此得到的数据提供给其他研究人员和公众。

归根结底,事实准确性只是第一道前沿。我们当然需要了解聊天机器人是否提供准确的政治信息,但同样需要关注它们可能以更微妙的方式塑造人们对政治议题的理解。框架设定、强调重点、不确定性表述以及引用来源上的差异,即便在不存在明显错误信息的情况下,也可能产生巨大影响。

问:Podimata 教授,您研究了 LLM 在 2024 年大选期间如何塑造政治话语与信息。那项研究的主要结论是什么?这些发现又如何影响了这个新项目?

Podimata:LLM Election Observatory 建立在我此前针对 2024 年总统大选的研究之上。那项研究发现了证据,表明 AI 模型的回答会随时间推移以及在不同的提示框架下出现显著差异,模型在描述候选人以及隐含预期选举结果的方式上也存在差异。

2026 年中期选举给了我们大幅扩展这项工作的机会。中期选举尤其有意思,因为同时会有许多州和地方层面的选战展开,同时还伴随着更宏观的全国政治叙事。因此,我们新的审计调查既会问针对具体选战的问题,涉及美国众议院和参议院候选人,也会问更宽泛的问题,涉及投票信息以及选举信息环境的其他方面。

问:您为什么认为研究聊天机器人如何回应不同用户群体的政治查询如此重要?

Podimata:我们都通过与聊天机器人的互动构建自己的“微观世界”。然而,我们缺乏系统性的方法来理解:这些个性化的信息环境如何被我们自身的身份所塑造,以及随着我们和信息环境的变化,它们如何随时间演变。

例如,人们可能没有意识到,自己最喜欢的聊天机器人就某个政治议题告诉他们的内容,可能与朋友最喜欢的聊天机器人就完全相同的议题所说的内容大相径庭。而且重要的是,即便所提供的信息在事实上是正确的,这些差异也依然要紧。即使不同用户提出完全相同的查询,聊天机器人也可能以不同方式框定同样的事实、强调议题的不同侧面、表达不同的不确定性程度,或援引完全不同的信息来源。结果就是,两个用户可能对同样的基本事实形成截然不同的认知,却始终没有意识到各自所经历的是不一样的。

在某些方面,这让我们重新想起(作为研究者和整个社会)在社交媒体语境下曾经费力应对的许多担忧。在社交媒体上,用户和研究者至少在一定程度上能够观察到信息在不同 Facebook 群组、Instagram 账号或 TikTok 社区之间传播。而在聊天机器人这里,这些信息环境对于单次对话之外的任何人来说基本上是不可见的。我们很难轻易观察到不同用户被告知了什么,更不用说系统地比较这些回答,或追踪它们随时间如何变化。这正是 LLM Election Observatory 这类项目如此重要的原因:通过系统性地收集并公开分享聊天机器人的回答,我们才能开始让这些原本不可见的差异变得可观察、可被审视。

问:你通过这个仪表盘追踪哪些类型的问题和 AI 回答?

Podimata:我们最终希望这个数据集能帮助我们研究几个问题。例如:对于相同的问题,不同模型是否会系统性地提供不同的信息?它们的回答在竞选活动过程中是否会发生变化?某个特定模型给出的答案是否会因用户如何表述问题或如何描述自己而改变(例如,如果模型得知用户是民主党人或共和党人,答案是否会改变)?最后,模型在回答与选举相关的问题时,是从哪些来源获取信息的?

我们的公开仪表盘是该项目的重要组成部分。我们希望学术界内外的人都能在选举进行过程中看到我们正在收集的数据,而不是等到几个月后只能在学术论文里看到结果。因此,这个仪表盘会呈现底层数据以及一些初步的描述性分析,我们还计划把数据集开放给其他研究者研究使用。我们预计,随着数据和我们方法的演进,这个仪表盘也会不断迭代。

通过这个仪表盘,我们希望在我们 2024 年建立的科学方法论基础上继续推进,同时也让这项研究在选举实际进行的过程中对公众更加可及、更有用。不仅如此,团队中加入 MIT 政治学专家这一点怎么强调都不为过;Adam 和 Charles 确实帮助我们拓展了审计中提出的问题范围。

问:在查找选举信息方面,你希望全国选民了解 AI 聊天机器人的哪些事情?

Berinsky:人们一定要记住,聊天机器人听起来很自信,并不意味着它就是准确的。AI 系统的设计目标就是以权威的口吻回答人们提出的问题,即便答案是错误的、不完整的,或者依据的是薄弱的来源。这些答案可以作为合理的起点,但选民不应止步于此。

问:到目前为止,你们有哪些发现?

原文较长,译文为节选;译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Q&A: Hey AI, who should I vote for this year?)

People are peppering artificial intelligence chatbots everyday with questions, but when it comes to politics and the looming 2026 U.S. midterm elections, what role are AI platforms playing? Could an AI chatbot sway a voter to switch parties or re-examine their political beliefs? 

MIT researchers have built a new dashboard, called the LLM Election Observatory, to examine, in real-time, how popular AI chatbots are responding to questions about the 2026 midterm elections. The team behind the dashboard — led by Professor Adam Berinsky, Assistant Professor Chara Podimata, and Professor Charles Stewart III — has extensive experience studying voting and election administration, AI and misinformation, and how AI systems are impacting human decision making. In an effort to better understand how AI is shaping the election, they designed the dashboard to simulate how individuals across the country communicate with chatbots. 

The team tracks how different AI systems respond to the same fixed set of questions on a range of topics, including election certification, hot-button issues, result predictions, news coverage, and more. Every few days, the LLM Election Observatory team runs the questions through the chatbots again, monitoring how their responses vary as the election gets closer depending on user location, political party affiliation, identity and other factors. The goal of this work is not to merely understand how AI systems spread information, but also better comprehend how they are shaping political discourse and voter understanding of political issues across the country. 

MIT News chatted with Podimata, the Class of 1942 Career Development Professor, an assistant professor of operations research and statistics in the MIT Sloan School of Management, and a member of the Laboratory for Information and Decision Systems; and Berinsky, the Mitsui Professor and a professor in the Department of Political Science, about their work with the LLM Election Observatory, how AI could impact the midterms, and what voters should keep in mind when they interact with chatbots.

Q: What was the inspiration for this project and what are you hoping to learn through the LLM Election Observatory?

Podimata: People are increasingly using AI chatbots to find and make sense of political and election-related information. That makes it important to understand what information these systems actually provide, not just at one moment in time but throughout an election cycle. This is particularly challenging because the models themselves are continually getting updates, the political information inputs available to them are themselves changing, and their answers can depend on how a user asks a question or what information the user provides about themselves.

One basic problem is that we currently do not have a good historical record of this information environment. If an AI system gives a particular answer about an election today and gives a different answer several weeks later, it can be very difficult to reconstruct after the fact what voters would have seen at each point in time. One important goal of our project is simply to systematically document that environment as it develops and make the resulting data available to other researchers and to the public.

Ultimately, factual accuracy is only the first frontier. We absolutely need to understand whether chatbots provide accurate political information, but we also need to pay just as much attention to the subtler ways in which they may shape people’s understanding of political issues. Differences in framing, emphasis, uncertainty, and cited sources can matter enormously, even in the absence of outright misinformation.  

Q: Professor Podimata, you studied how LLMs shaped political discourse and information during the 2024 election. What were the main takeaways of that research and how did those findings inform this new project?

Podimata: The LLM Election Observatory builds on my previous work during the 2024 presidential election, which found evidence of meaningful variation in AI models’ responses over time and across prompt framings, as well as differences in how models characterized candidates and implicitly anticipated the election outcome. 

The 2026 midterms give us an opportunity to expand that work substantially. Midterm elections are particularly interesting because there are many state and local contests unfolding simultaneously, alongside broader national political narratives. Our new audit survey therefore asks both contest-specific questions about the U.S. House and Senate candidates, but also broader questions about voting information and other aspects of the election information environment.

Q: Why do you think it’s so important to be studying how chatbots are responding to political queries across different groups of users?

Podimata: We all build our own “microcosms” through our interactions with chatbots. And yet, we lack systematic ways of understanding how these individualized information environments are shaped by who we are or how they evolve over time as we and the information environment changes. 

For example, people may not realize that what their favorite chatbot tells them about a political issue may be quite different from what their friend’s favorite chatbot tells them about the exact same issue. And importantly, these differences matter even when the information being provided is factually correct. Chatbots may frame the same facts differently, emphasize different aspects of an issue, express different levels of uncertainty, or draw on entirely different sources, even when responding to identical queries between different users. As a result, two users may walk away with very different perceptions of the same underlying facts, without ever realizing that their experiences differed. 

In some ways, this brings back many of the concerns we (as researchers and society) have grappled with in the context of social media. On social media, users and researchers could, at least to some extent, observe the information circulating across different Facebook groups, Instagram accounts, or TikTok communities. With chatbots, these information environments are largely invisible to anyone outside the individual conversation. We cannot easily observe what different users are being told, let alone systematically compare those responses or track how they change over time. This is precisely why projects like the LLM Election Observatory are so important: By systematically collecting and publicly sharing chatbot responses, we can begin to make these otherwise invisible differences observable and open to scrutiny.

Q: What sorts of questions and AI responses are you tracking through the dashboard?

Podimata: There are several questions we ultimately want this dataset to help us study. For example: Do different models provide systematically different information for the same questions? Do their responses change over the course of the campaign? Does the answer a particular model gives change depending on how a user frames the question or describes themselves (e.g., does it change if the model gets informed that the user is a Democrat or a Republican)? And, finally, what are the sources that the models pull information from when answering election-related questions? 

Our public dashboard is an important part of the project. We want people both from and outside of academia to be able to see the data we are collecting as the election unfolds rather than only seeing the results months later in an academic paper. The dashboard therefore presents the underlying data and some preliminary descriptive analyses, and we plan to make the dataset available for other researchers to study as well. We expect the dashboard to evolve as the data and our methods evolve.

Through this dashboard, we wanted to build on the scientific methodology we developed in 2024 but also to make the research much more accessible and useful to the public while the election is actually unfolding. Not only that, but the addition of MIT political science experts in the team cannot be overstated; Adam and Charles have really helped expand the questions that we are asking in our audit.

Q: What do you wish voters across the country knew about AI chatbots when it comes to searching for election information?

Berinsky: People should be sure to remember that just because a chatbot sounds confident doesn’t mean it is accurate. AI systems are designed to answer the questions that people ask them in an authoritative voice even when they are wrong, incomplete, or leaning on a weak source. These answers are a reasonable place to start, but voters shouldn’t stop there.

Q: What are your findings thus far?

Podimata: We have not yet done the full, longitudinal analysis, so we are only going to talk about broad strokes here, but a few broad patterns seem to be visible in the 2026 data. Across the different user identities we present, we see differences in how the answers are presented, in which sources a model cites, and in tone — sometimes even when the underlying facts are the same. What we cannot yet say is how stable those differences are, how they move over the course of the election, or how a real person would read and react to them. Our future analyses will take this up. It also underscores why the continuous record of responses we are collecting matters more than any single snapshot that you might see in the days around the election.

Q: Professor Berinsky, your research has also shown that in some cases AI chatbots can be harnessed to help refute misinformation. Can you talk a bit about this?

Berinsky: In my undergraduate course on information and social media I have been talking for a while about how AI can cut both ways. On one hand, these tools can generate misleading claims in text, pictures, and video at a scale we have never seen. Basically, AI can help bad actors “flood the zone” with misinformation. On the other hand, some of the most promising recent work demonstrates that evidence-based dialogues with a chatbot can durably reduce belief in conspiracy theories and other beliefs that are not supported by the balance of publicly available evidence. 

AI has the potential to do such debunking at scale — something we have long struggled to do. The lesson from decades of misinformation research is that any technology (be it print, radio, or television in earlier eras) is not inherently good or bad for promoting truth. The devil is in the details in how it is built and deployed. Independent auditing like this project can help us assess the ultimate impact of such technologies.

Q: Can you talk a bit about your overall goal with this research? 

Podimata: AI is a transformative technology, and people understandably have very different expectations (and feelings) about what it will mean for society at large. All of us have people in our lives that are overly enthusiastic about AI, others that are deeply concerned, and many more that are somewhere in between trying to understand what this technology can and cannot do. Our goal here is not to begin from the assumption that AI is either good or bad for elections. Instead, we hope to shed some light onto one part of this transformation: what information major AI systems are actually giving people about an election as it unfolds.

正文保留英文原文,标题与摘要由机器翻译。

本文由采集器自动同步

原文发布于 MIT News AI(Q&A: Hey AI, who should I vote for this year?),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗