Voters seeking election information through chatbots may encounter inaccurate and potentially harmful responses, according to a new study. Researchers from the AI Democracy Projects and Proof News evaluated five leading AI chatbots, including GPT-4, on their ability to answer basic voter queries. These queries covered topics such as polling place location, voter registration requirements, and voting procedures.
The study found that the chatbots delivered inaccurate information over half the time, raising concerns about their reliability as a source of election information. "People are using these models as a search engine, and it's kicking out garbage," stated Bill Gates, a Republican election official from Arizona, in an interview with Proof News. "If you want the truth about the election, don't go to an AI chatbot. Go to the local election website."
The study also highlighted the potential for harm beyond simple inaccuracies. In some cases, the chatbots provided misleading information that could discourage voters from participating in the election process. For instance, some chatbots incorrectly claimed that certain groups of people were ineligible to vote or that specific identification documents were required at polling places when they weren't.
"This is a serious danger to democracy," said Alondra Nelson, director of the Science, Technology and Social Values Lab at the Institute for Advanced Study and co-director of the AI Democracy Projects. "We need to be very concerned about the potential for AI to spread misinformation, both intentional and unintentional, during elections."
The researchers emphasize the need for caution when using AI chatbots for election information. They recommend that users consult official election websites or trusted sources for accurate and reliable information. Additionally, the study's findings highlight the importance of developing and implementing measures to ensure the accuracy and fairness of AI systems used in public spheres.
While some chatbots, such as GPT-4, performed better than others in terms of accuracy, the overall results raise concerns about the technology's current readiness for providing election-related information. As AI chatbots become increasingly sophisticated and integrated into our lives, addressing these concerns becomes crucial for ensuring responsible development and deployment of this technology.