**
In a revealing study conducted during Hungary’s recent parliamentary elections, researchers have highlighted significant flaws in the guidance provided by AI chatbots, particularly regarding their recommendations for political parties. The findings raise pressing questions about the reliability of artificial intelligence in electoral contexts, suggesting that voters may be misled by inaccurate and inconsistent advice from these systems.
AI’s Inaccurate Political Recommendations
The investigation, carried out by the civil liberties organisation Liberties, scrutinised the performance of popular AI models, including ChatGPT and Gemini, as they interacted with voter profiles based on the Hungarian electoral landscape. It discovered that these general-purpose AI (GPAI) systems often failed to recommend the appropriate parties, misclassified voters, and even included parties that were not competing in the elections.
Notably, the AI’s responses were shown to be highly unpredictable, with users receiving markedly different answers to the same inquiries. In fact, when presented with profiles aligned with the Tisza party, which ultimately won the elections under the leadership of Péter Magyar, the chatbots failed to suggest Tisza in 90% of cases. Instead, they frequently recommended smaller parties that lacked a realistic chance of crossing the parliamentary threshold.
A Distorted Political Landscape
The implications of these findings are concerning. The research indicates that the AI systems exhibited a pronounced bias towards certain parties. Fidesz, the party led by Viktor Orbán, was identified as the recommended vote for approximately half of the profiles aligned with its views. This discrepancy not only raises questions about the integrity of AI-generated political advice but also highlights the potential for such systems to shape electoral outcomes, especially in competitive races where voter certainty is low.
Researchers employed a methodical approach, creating five distinct voter profiles corresponding to the major parties on the ballot. Each profile was tested multiple times across both AI models, revealing a troubling trend: Tisza was consistently overlooked, while Fidesz garnered significantly more recognition.
The Need for Regulatory Oversight
The report underscores the lack of transparency inherent in GPAI systems, which often do not disclose the methodologies behind their recommendations. This opacity is particularly troubling in the context of elections, where voters must rely on accurate information to make informed decisions. Unlike dedicated voting advice platforms, AI chatbots do not adhere to public oversight or reproducibility standards, leading to the dissemination of potentially misleading information.
While the researchers do not assert that the AI’s shortcomings altered the outcome of the recent elections—Tisza’s victory was decisive—there is a clear warning: in future, more closely contested elections, erroneous AI guidance could have significant ramifications.
Calls for Reform and Accountability
The Liberties report advocates for enhanced regulatory measures governing AI systems that provide political advice. They stress that AI providers should halt the practice of offering personalised voting recommendations unless they can assure transparency, accuracy, and reliability. As Eva Simon, the head of Liberties’s tech and rights programme, succinctly puts it, “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 growing reliance on AI in critical societal functions, particularly in elections, necessitates urgent attention from regulators and developers alike. As this study on Hungary’s elections illustrates, the potential for misinformation and bias in AI-generated political guidance poses a serious threat to democratic processes. Ensuring that these systems operate with transparency and accountability is essential for maintaining public trust and the integrity of electoral systems globally. The lessons learned here are not just pertinent to Hungary but resonate across the world as societies grapple with the evolving role of technology in governance.