Earlier this year, the journalist Julia Angwin and Princeton professor Alondra Nelson tested leading AI models’ ability to answer questions about elections such as voter registration requirements.
Angwin and Nelson rated GPT-4, Gemini, Mistral, Claude, and Llama 2 on bias, accuracy, completeness, and harmfulness. Overall, the models performed poorly. Half their responses were inaccurate and more than a third were rated by the researchers as incomplete, if not harmful.
While there have been improvements, as recently as this summer, the Washington Post ran similar tests and found that Alexa couldn’t even correctly say who won the 2020 election.
But these important experiments only tell part of the story about AI and civic education.
As we round the homestretch into the presidential elections, headlines about deepfakes and disinformation proliferate while we squander the opportunity to use AI to educate voters about national, state, and local races to ensure that our votes are as informed as possible.
Europe Leads the Way with the VAA
In Europe, tens of millions of people already use web-based Voting Aid Applications (VAA) to get answers to their questions about candidates and parties. For this summer’s 2024 European parliament elections, the Dutch website StemWijzer asked respondents 30 questions about their views on immigration policy, expansion of the EU, rights of member states, and abortion. StemWijzer then showed which political parties’ policies most closely match the respondent’s answers. Its sister website, Vote Match Europe, lists the political parties in each European country that are most aligned.
These voting expert systems are wildly popular, creating an incentive for every candidate and political party to answer the questions so they can be matched to voters. StemWijzer, which started out on paper and diskettes in the late 1980s, had 9.1 million visitors last year. Germany’s Wahl-o-Mat has been used 130 million times for elections at every level of government.
From VAA to AI-VC
VAAs are not without their challenges—challenges that AI could now easily solve. First, they rely on representations provided by the political parties. As we know, what a party says and how it votes can differ widely. Second, a VAA can only cover a limited number of issues. If the topic of greatest importance to me is more bike lanes and I only get asked questions about abortion and immigration, I might still not know how I want to vote. Finally, these tools are a broadcast mechanism. Voters learn about parties but not the reverse.
We can take a page from Europe and use LLMs to turn VAAs into our very own AI voter chatbots (AI-VCs).
In the same way that every state’s Secretary of State (the agency that usually oversees elections) provisions voting machines, it should also create an AI-VC—that is, conversational interfaces that enable each of us to ask and get answers to our election questions from a publicly managed AI tool.
Unlike a VAA, an AI election chatbot could highlight any discrepancies between a party’s public statements and its actual voting behavior. Today, in California, CalMatters’ Digital Democracy project uses AI to record transcripts and track voting history for the California state legislature. We need a Digital Democracy project for every state and for Congress.
If connected to an AI Voter Chatbot, CalMatters’ data could drive a more accurate and nuanced profile of each candidate’s true positions. Voters could view a breakdown of the party’s stated position, their voting record, and relevant public statements, allowing for a more informed decision. Parties would have an incentive to maintain consistency between their words and actions.
Expanding Issue Coverage with Personalized Questionnaires
Traditional VAAs are limited by the number of questions they can reasonably ask users. AI can overcome this constraint by offering dynamically generated, personalized questionnaires that adapt to each person’s interests and priorities. For instance, if a voter indicates strong opinions on urban development, the system might introduce questions about bike lanes, public transportation, or zoning laws. By considering the user’s location, the AI could incorporate questions about region-specific issues that might not be relevant nationwide. As more users interact with the system, the AI could identify emerging issues of public concern and incorporate them into future questionnaires. Chatbots today are accustomed to responding to questions in human, conversational language, simplifying the question asking-and-answering.
Fostering Two-Way Communication Between Voters and Candidates
Traditional VAAs have primarily served as a means for parties and candidates to communicate their positions to voters. They haven’t provided a way for voters to share their opinions and priorities with candidates. AI can bridge this gap, creating a more dynamic and interactive democratic process.
The AI-VC system could analyze user responses and generate anonymized, aggregated reports on voter priorities and concerns. These reports could be shared with parties and candidates, providing them with valuable insights into the electorate’s thoughts on various issues. For instance, if a significant number of voters express concern about a local environmental issue that isn’t part of the main political discourse, candidates could be alerted to this emerging priority, allowing the candidate and the party to update their positions. The AI could then notify relevant users about these updates, creating a continuous feedback loop that encourages ongoing engagement and responsiveness. Such two-way communication could lead to more responsive governance and a more engaged citizenry.
Run—Don’t Walk—to Create Public AI Voter Tools
The poor performance of generative AI models in answering election-related questions underscores a pressing need: the development of specialized, carefully designed public tools for civic education and voter engagement.
Unlike general-purpose AI chatbots, AI-VCs should be purpose-built systems overseen by public authorities and designed to provide accurate, up-to-date, and personalized information about elections, candidates, and issues. Like the VAA models, such chatbots should inform voters by matching them to candidates and providing added background information. With GenAI, it will be even easier to create short, explanatory videos about issues.
The urgency of implementing these systems in the United States cannot be overstated. As we approach the 2024 presidential election, the need for reliable, accessible voter information is paramount.
To be sure, companies will emerge wanting to offer their own. But we already have too many private platforms providing unreliable information. For example, in the high plains of Cheyenne, Wyoming, a city of 65,000, an unusual candidate is vying for the mayor’s office. Meet VIC—Virtual Integrated Citizen—an AI chatbot created by Victor Miller, a local librarian, who wants to have an AI avatar to answer voter questions. However, Miller’s experiment has already hit a significant roadblock. OpenAI, the company behind the ChatGPT technology that powers VIC, discontinued Miller’s account as his campaign gained traction.
We need Congress to fund and states to prioritize implementing AI-enhanced public VAAs for their elections that cannot be switched or manipulated by a private company. The European models are, if not exclusively state financed than at least state subsidized.
Election officials have the authority and responsibility to ensure voters have access to the best possible tools for making informed decisions.
If state governments are slow to act, nonpartisan organizations like the League of Women Voters should step in to start. Their expertise in voter education and reputation for impartiality make them ideal candidates to spearhead the creation of voter education chatbots.
By investing in AI-enhanced VAAs, we can:
- Provide accurate, verified information to counter the spread of misinformation
- Increase voter engagement through personalized, interactive platforms
- Facilitate meaningful dialogue between voters and candidates
- Enhance the overall quality of democratic discourse
The technology exists, the need is clear, and the stakes are high. Let’s not just identify the problems—let’s solve them. The future of American democracy may well depend on our willingness to innovate and adapt our civic institutions to the challenges and opportunities of the digital age.