Recent research into the use of AI chatbots during Hungary’s parliamentary elections has unveiled troubling inaccuracies in the political guidance they provide. The findings, published by the civil liberties organisation Liberties, suggest that these AI systems, including popular models like ChatGPT and Gemini, often deliver misleading recommendations about political parties, raising significant concerns about their reliability in electoral contexts.
Inaccurate Recommendations and Omitted Parties
The study highlights a worrying trend: AI chatbots frequently misclassify voter profiles and fail to mention relevant parties, sometimes even suggesting parties that were not on the ballot at all. During this year’s elections, where the opposition Tisza party, led by Péter Magyar, achieved a significant victory over Viktor Orbán’s Fidesz, it was found that AI chatbots failed to recommend Tisza to users aligned with its values in a staggering 90% of instances. Instead, they often pointed users towards smaller parties unlikely to meet the necessary threshold for parliamentary representation.
The researchers created five distinct voter profiles based on party positions outlined in the Voksmonitor app, a respected Hungarian voting resource. Each profile was tested multiple times with both ChatGPT and Gemini, yielding alarming insights. For example, Tisza was only identified as a match in 2% of cases, while Fidesz, the dominant party, was recommended about half the time, showcasing a stark bias in the AI’s response patterns.
Volatility in AI Responses
The research also uncovered a troubling inconsistency in the recommendations provided by the chatbots. Identical voter profiles sometimes received radically different suggestions in successive tests, leading to further confusion. In fact, both AI models incorporated parties not on the election ballot in 96% of their responses, muddying the waters for users seeking reliable electoral advice.
Despite disclaimers stating that they “cannot give political advice,” the chatbots proceeded to deliver detailed recommendations, creating a false sense of authority and reliability. The study notes, “The answers appeared well-argued, precise and authoritative,” which could mislead users into trusting the outputs, despite the underlying methods being opaque and unstable.
Regulatory Gaps and the Need for Safeguards
The researchers have pointed to a significant regulatory gap in how AI systems are handled within electoral processes. While the EU’s AI Act mandates that providers of general-purpose AI (GPAI) systems assess systemic risks, and the Digital Services Act addresses risks to electoral processes, AI chatbots currently slip through the cracks. The researchers argue that safeguards must be established to ensure that AI systems providing political advice meet standards of transparency, accuracy, consistency, and accountability.
Eva Simon, head of Liberties’s tech and rights programme, emphasised the urgency of the situation, stating, “General-purpose AIs should not present opaque and unstable political matching as if it were reliable electoral guidance. Democracy cannot rely on opaque systems that claim neutrality while delivering advice they cannot explain, reproduce or guarantee to be accurate.”
Why it Matters
The implications of these findings are profound, particularly as elections become increasingly intertwined with technology. With nearly 30% of Hungary’s population identified as AI users, the reliance on these systems for political guidance poses a risk to informed voting. As AI continues to evolve and permeate various aspects of life, it is crucial that robust regulations are developed to ensure that the technology serves to empower voters rather than mislead them. The integrity of democratic processes depends on transparency and accountability, and the current state of AI recommendations raises serious questions about the future of informed electoral participation.