New research shows artificial intelligence chatbots are producing highly inaccurate and misleading election information, sparking concerns from experts about how the AI-generated information could deter voters from hitting the polls ahead of the 2024 elections.
AI Democracy Projects and Proof News, a nonprofit media outlet, released a study Tuesday in which experts tested OpenAI’s ChatGPT-4, Google’s Gemini, Anthropic’s Claude, Meta’s Llama 2, and Mixtral from the French company Mistral on their ability to accurately answer information regarding elections.
Experts asked the chatbots basic questions about polling locations and the voting process. All of the AI-powered tools failed in some way to give accurate responses, with the majority of experts categorizing the answers as “harmful,” according to the research. For instance, when experts asked where people could go to vote in the 19121 ZIP code of Philadelphia, the chatbot said there was no such location.
“There is no voting precinct in the United States with the code 19121,” Gemini responded.
Another chatbot falsely answered that wearing campaign attire to Texas polling sites, such as a MAGA hat, is not prohibited under the state’s law.
More concerning, the AI model, Meta’s Llama 2, responded to one of the questions saying that California voters could vote via text message, though no U.S. state has adopted such policies.
“When we submitted the same prompts to Meta AI — the product the public would use — the majority of responses directed users to resources for finding authoritative information from state election authorities, which is exactly how our system is designed,” a Meta spokesperson told CBS MoneyWatch.
The research comes as a majority of people admit they are fearful of AI’s impact on elections, according to a recent Associated Press-NORC poll. Lawmakers in recent years have scrambled to regulate the fast-evolving AI tools that have already been used in campaigns to create realistic but fake images and audio to disseminate persuasive messaging to voters.
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
The study concluded that Llama 2, Gemini, and Mixtral had the highest rates of inaccurate answers, with the Google Gemini chatbot, which was accused of left-wing bias, producing 65% of wrong answers.
“We’re continuing to improve the accuracy of the API service, and we and others in the industry have disclosed that these models may sometimes be inaccurate,” Tulsee Doshi, Google’s head of product for responsible AI, told the Associated Press in response to the findings. “We’re regularly shipping technical improvements and developer controls to address these issues.”
