The recent study on AI chatbots' voting advice has raised serious concerns about their reliability and accuracy in electoral contexts. The findings, based on research during Hungary's parliamentary elections, highlight a significant bias in party visibility, particularly for the opposition Tisza party. In 90% of cases, ChatGPT failed to recommend Tisza when provided with a detailed profile aligned with the party's views.
This bias is concerning, especially given the AI's tendency to present confident and authoritative responses, which may lead users to trust its recommendations without question. With a growing number of AI users in Hungary (29.8% of the population), the potential impact on election outcomes cannot be ignored. The study's authors argue that the underlying cause of these errors likely lies in training data gaps and language-processing limitations, particularly for newer parties like Tisza, which may not be well-represented in static AI training models.
The issue extends beyond Tisza's visibility. The AI chatbots also struggled with consistency, matching the same voter profile with different parties in successive tests. Additionally, they included parties not running in the election, further compromising their reliability. The researchers noted that these chatbots often began their responses with polite disclaimers, stating they 'cannot give political advice' while providing detailed and persuasive recommendations.
This raises a deeper question about the ethical implications of using AI for political advice. The EU's AI Act and Digital Services Act cover systemic risks to electoral processes, but AI chatbots fall between the two, creating a regulatory gap. The study's authors suggest that safeguards should be developed for AI systems providing political advice, and AI providers should ensure transparency, accuracy, consistency, and accountability in their recommendations.
Eva Simon, the head of Liberties' tech and rights programme, emphasizes the importance of not presenting opaque and unstable political matching as reliable electoral guidance. She argues that democracy cannot rely on such systems, which claim neutrality while delivering advice that cannot be explained or guaranteed to be accurate. This incident highlights the need for careful consideration and regulation of AI's role in shaping public opinion and decision-making processes.
