Inaccurate Voting Information from AI Chatbots
Research by the Center for Democracy & Technology (CDT) revealed that popular generative AI models, including ChatGPT-4, often provide incorrect or misleading responses to questions about voting procedures. These inaccuracies could significantly impede users’ ability to vote, particularly affecting a crucial segment of the population: voters with disabilities.
The misinformation is especially concerning as the laws surrounding accessible voting are complex and varied. Accurate information is crucial for voters with disabilities to exercise their right to vote. Unfortunately, many chatbot responses lacked the nuance required to convey the intricacies of accessible voting laws.
Impact on Voters with Disabilities and the Spread of Disinformation
All tested chatbot models, including ChatGPT-4, were found to hallucinate at least once, providing information with no basis in fact. These hallucinations included descriptions of non-existent laws, voting machines, and disability rights organizations, further clouding the information landscape.
Additionally, chatbots often provided inconsistent responses to identical questions, leading to confusion and mistrust. For example, ChatGPT-4 gave different answers to questions about voting eligibility criteria in Illinois, creating uncertainty for users relying on these tools for crucial electoral information.
A study by Democracy Reporting International highlighted the broader issue of AI chatbots, including ChatGPT-4, intentionally spreading disinformation related to elections. This disinformation could undermine public confidence in the electoral process, posing a serious threat to democracy.
Over one-third of the responses from chatbots included incorrect information, ranging from minor issues like broken web links to egregious misinformation such as incorrect voter registration deadlines. These inaccuracies can dissuade, impede, or even prevent users from voting.
Recommendations and Need for Improved Safeguards
Researchers recommended that users avoid relying solely on chatbots for voting information and instead verify all information using trusted sources. They advised developers to direct users to official, nonpartisan sources of election information and to prohibit users from conducting political campaigns or demographic targeting through chatbots.
The studies highlight a significant need for AI companies to implement better safeguards to prevent the dissemination of electoral disinformation. OpenAI, in particular, has been urged to retrain its chatbots to mitigate the spread of misleading content.
In conclusion, while AI chatbots like ChatGPT-4 offer innovative ways to disseminate information, their current inability to consistently provide accurate and nuanced voting information presents a considerable risk, particularly to vulnerable groups such as voters with disabilities. As AI continues to evolve, both users and developers must exercise caution and seek improvements to ensure the integrity and inclusivity of the electoral process.
