Election voting advice from AI chatbots ‘inaccurate and unreliable’

Editor's take

AI chatbots offered demonstrably flawed electoral advice during a Hungarian election, recommending non-existent parties and providing wildly divergent suggestions for identical queries. This research highlights a critical vulnerability in AI's nascent role in civic discourse, particularly concerning its unvetted deployment in politically sensitive contexts where even minor inaccuracies can sow confusion. The implications extend beyond Hungary, raising serious concerns about the potential for AI to inadvertently or deliberately manipulate public opinion in democratic processes globally.

The immediate focus should be on rigorous, independent auditing of AI systems deployed for public information during elections, not just in Hungary but in any jurisdiction. Key questions remain about who bears responsibility for these inaccuracies—the AI developers, the platforms hosting them, or the users prompting them. Future developments to monitor include the establishment of clear regulatory frameworks for AI in political campaigning and the development of AI models specifically trained for factual accuracy and neutrality in civic information dissemination.