Rua Hygino Muzy Filho, 737, MARÍLIA - SP contato@latinoobservatory.org
IMG-LOGO
Home / News

News

Voting rights groups warn that AI models are generating electoral misinformation in Spanish

Editores | 04/11/2024 10:36 | POLITICS AND THE ECONOMY
IMG https://apnews.com/article/ai-chatbots-chatgpt-falsehoods-spanish-latino-voters-d4ffabdc006a7c70eece3e28efa3de90

On the eve of the U.S. presidential election, Latino voters are facing an increase in Spanish-language political ads and messages, including unsubstantiated claims created by Artificial Intelligence (AI) chatbots. According to an analysis by Proof News and Factchequeado, AI models are producing electoral disinformation in Spanish more frequently than in English, which hinders access to accurate information for this growing and influential voting bloc, according to the Associated Press.


The analysis shows that more than half of the answers in Spanish contained incorrect information, while in English the percentage was 43%. Meta's Llama 3 model performed particularly poorly, with nearly two-thirds of responses in Spanish incorrect, including a wrong definition for "federal voter only." Other models, such as Anthropic's Claude and Google's Gemini, also had errors, with confusing or incorrect answers about the American electoral system.


The companies, including Meta and Anthropic, are trying to tweak their models to reduce inaccuracies, but voting rights advocates warn that Spanish-speaking voters should look to multiple sources of reliable information. According to the article, Lydia Guzman, an advocate for electoral rights, recommends voters to do careful research and cross-check information to avoid being harmed by misinformation, according to the publication.


These inaccuracies mainly affect states with large Hispanic populations, such as Arizona, Nevada, Florida and California, where Latino voters are decisive. One-third of eligible voters in California are Latino, and one in five Latino voters speak only Spanish, adding to the potential impact of misinformation.


A practical example was given by Rommell Lopez, a paralegal in California, who reports finding conflicting answers when trying to verify claims about immigrants. He points out that while the technology is useful, it cannot be completely trusted. This phenomenon reflects the current challenges of using AI in electoral contexts, especially among communities that rely heavily on information in diverse languages.

Search for a news: