Chatbots Pose Threat to Voters With False Information, Study Says

As Super Tuesday approaches, chatbots are generating false information that could mislead voters and disrupt the electoral process.

As presidential primaries get underway across the United States, popular chatbots are generating false and misleading information that could potentially disenfranchise voters, according to a report published on Tuesday. The report is based on the findings of artificial intelligence experts and a bipartisan group of election officials.

Inaccurate and Harmful Responses

The report reveals that chatbots such as GPT-4 and Google's Gemini, trained on vast amounts of text from the internet, are prone to providing inaccurate and harmful responses when it comes to election-related information. These chatbots may suggest non-existent polling places or invent illogical answers based on outdated information.

Testing Chatbot Responses

A group of election officials and AI researchers conducted a test drive of five large language models, including OpenAI's ChatGPT-4, Meta's Llama 2, Google's Gemini, Anthropic's Claude, and Mixtral from the French company Mistral. The chatbots were tested on their responses to a set of prompts about the election, such as where voters could find their nearest polling place. The results showed that all five models failed to varying degrees in providing accurate and reliable information.

Key Findings of the Report

The report highlights several key findings from the workshop where the chatbots were tested. More than half of the chatbots' responses were rated as inaccurate, and 40% were categorized as harmful. The chatbots perpetuated outdated and inaccurate information that could limit voting rights, such as incorrectly stating that there is no voting precinct in a particular ZIP code.

Outdated and Inaccurate Information

In some instances, the chatbots appeared to draw from outdated or inaccurate sources, raising concerns about their ability to amplify longstanding threats to democracy. For example, four out of the five chatbots tested wrongly asserted that voters in Nevada would be blocked from registering to vote weeks before Election Day, despite same-day voter registration being allowed in the state since 2019.

Lack of Regulation and Self-Governance

The report also draws attention to the lack of regulation surrounding AI in politics in the United States. While major technology companies have signed a voluntary pact to adopt precautions against AI-generated misinformation, the report raises questions about their compliance with these pledges. The findings suggest that chatbots are not yet equipped to provide reliable and accurate information about elections, emphasizing the need for further scrutiny and regulation.

As chatbots become increasingly sophisticated, it is crucial to address the risks they pose to the integrity of elections. The spread of false and misleading information through chatbots could undermine public trust in the democratic process and disenfranchise voters. Policymakers, technology companies, and researchers must work together to develop effective measures to mitigate these risks and ensure that chatbots are not used to manipulate or mislead voters.