In the throes of the final days leading up to the upcoming presidential election, a new wave of political messaging is targeting Latino voters in the United States, one of the most rapidly growing and influential voter demographics in the country. This time, the source of concern isn't only televised ads or campaign flyers; it’s the advent of artificial intelligence (AI) chatbots disseminating misleading information in Spanish at an alarming rate.
A recent analysis conducted by Proof News and Factchequeado, in collaboration with the Science, Technology and Social Values Lab at the Institute for Advanced Study, has revealed that AI-generated responses in Spanish are more likely to contain inaccuracies compared to those in English. The study revealed that over half of the election-related answers generated in Spanish by AI models were incorrect, in contrast to the 43% in English.
Notably, Meta’s Llama 3 AI model, widely integrated into platforms such as WhatsApp and Facebook Messenger, was highlighted for its ineffective performance. It provided errors in nearly two-thirds of its Spanish responses versus about half in English. One instance involved a misunderstanding about "federal only" voters in Arizona—who are restricted to voting in federal elections without state-provided citizenship proof—where the AI inaccurately equated them with residents of U.S. territories like Puerto Rico, who cannot partake in presidential votes.
Moreover, other AI models, including Anthropic’s Claude and Google’s Gemini, displayed their flaws. Claude, for instance, directed users to contact election authorities in unrelated countries for voter eligibility questions, whereas Gemini offered convoluted explanations when asked about the Electoral College.
Given the burgeoning use of AI for information delivery, this Spanish language disparity is worrisome amidst concerted efforts to court Latino voters by both Democratic and Republican campaigns. Vice President Kamala Harris is set to engage Latino voters with a rally in Las Vegas, featuring high-profile performers such as Jennifer Lopez and the Mexican band Maná. Concurrently, former President Donald Trump recently appealed to this demographic during a public event in Pennsylvania.
The inaccuracies identified in AI-generated information raise potential concerns about election-related misinformation, particularly in states with significant Hispanic populations such as Arizona, California, Nevada, and Florida. According to the UCLA Latino Policy and Politics Institute, nearly a third of eligible voters in California are Latino, with many relying primarily on Spanish for election information.
Tech companies are responding to the findings with statements of intent to improve their systems. Meta has been working on refining their models based on safety and accuracy criteria, while Anthropic has updated its policy to better address Spanish-language inquiries by steering users toward reliable sources. However, Google has so far not provided a public comment on the issue.
As elections draw near, the need for accurate election-related information becomes even more critical. Rommell Lopez, a paralegal from California who frequently engages with AI tools like OpenAI’s ChatGPT for information verification, expressed a pragmatic outlook on relying on technology—a sign of the larger challenge voters face when distinguishing factual data from misinformation.
This development forms part of a broader discourse on the role of AI in the 2024 election cycle, as major tech advancements continue to influence the ways voters access and interpret electoral information.
Source: Noah Wire Services