A recent investigation into the performance of AI chatbots during Hungary’s parliamentary elections has raised alarming questions about the reliability of these systems in guiding voters. The research, conducted by the civil liberties organisation Liberties, has revealed that widely-used AI tools, including ChatGPT and Gemini, often provided inconsistent and misleading advice regarding political party recommendations. With significant implications for electoral integrity, the findings call for urgent scrutiny of AI’s role in democratic processes.
Inaccurate Recommendations Highlight Critical Flaws
The study’s findings are particularly concerning. Many voters seeking guidance from AI chatbots received incorrect or irrelevant party suggestions, including recommendations for parties that were not even contesting the elections. The analysis emphasises that these general-purpose AI systems produced highly variable responses to identical questions, undermining their credibility as trustworthy electoral advisers.
Notably, the report highlighted a pronounced bias against the Tisza party, led by Péter Magyar, which achieved a decisive victory against Viktor Orbán’s long-standing Fidesz party. Astonishingly, the AI tools failed to recommend Tisza 90% of the time when provided with profiles aligned with its platform. This raises significant concerns about the potential influence of AI on voter behaviour, especially in tighter electoral contests where every vote counts.
Methodology and Findings
To arrive at these conclusions, Liberties conducted tests on the two most prevalent AI chatbots, ChatGPT and Gemini, using voter profiles aligned with the five parties registered for the Hungarian elections. They created distinct profiles based on positions outlined in Voksmonitor, a reputable voting advice application. Each profile was evaluated multiple times, asking the AI systems to either recommend a party or provide a percentage match with the parties.
The results were striking. ChatGPT recognised the Tisza party in only 2% of its percentage-matching assessments, typically steering users towards smaller, less viable parties instead. In stark contrast, Fidesz-aligned profiles were consistently identified, with the AI recommending Fidesz as the preferred choice in half of the cases and presenting it as a primary option in many others.
The Need for Regulation and Transparency
The report underscores a critical regulatory gap pertaining to AI’s involvement in political advising. While the EU has established frameworks, such as the AI Act and Digital Services Act, these do not adequately cover the unique challenges posed by general-purpose AI chatbots. The study argues for the development of stricter safeguards to ensure that AI systems providing political advice can guarantee transparency, accuracy, and accountability.
Eva Simon, head of Liberties’s tech and rights programme, emphasised the need for a paradigm shift: “General-purpose AIs should not present opaque and unstable political matching as if it were reliable electoral guidance for voters.” This sentiment encapsulates the urgent requirement for AI systems to operate with clarity and reliability, particularly in contexts as sensitive as electoral politics.
Why it Matters
The implications of this study are profound. As AI systems become increasingly integrated into everyday decision-making processes, their role in shaping public opinion and electoral outcomes cannot be overlooked. The findings from Hungary serve as a crucial warning about the potential risks of relying on AI for political advice, particularly in democratic societies. It is imperative that regulators, technologists, and civil society collaborate to establish robust frameworks that ensure the integrity of electoral processes, safeguarding democracy against the pitfalls of unreliable technology.
