Gemini AI dodges 2020 election question

Google’s AI chatbot, Gemini, was found to give evasive answers when asked who won the 2020 U.S. presidential election, but it was more forthcoming when asked about elections in other countries, a report from Ars Technica said. The chatbot said it “was not sure” or “didn’t have that information” when the question was about the U.S., but it correctly identified the winner of France’s 2022 election and the UK’s 2019 election. The discrepancy sparked concern about how the AI is trained to handle political content.

The issue matters because AI tools are used by millions to get quick facts. If a chatbot can’t answer a basic question about a recent election, users may doubt its accuracy. Google says its policy is to avoid providing definitive answers to political questions that could be seen as partisan. The policy is meant to keep the model neutral, but it also means the AI can appear evasive. The policy was introduced after the company faced criticism for giving misleading political information in earlier versions of its chatbot.

When Ars Technica tested Gemini, the AI’s response to the U.S. question was vague, citing a lack of certainty. The same test on the 2020 election in Canada produced a clear answer. Experts say the difference is likely due to the model’s training data and the way Google’s policy flags U.S. political content. Google’s engineering team said they are working on updating the policy so that the AI can provide factual answers while still remaining unbiased. The company also plans to improve the model’s ability to distinguish between political claims that are widely verified and those that are disputed.

What happens next is that regulators in the U.S. and the EU may scrutinize how tech firms handle political content in AI. Google says it will release updates that allow the chatbot to give clear, factual answers about elections while maintaining neutrality. The company also plans to publish a transparency report on how it applies its policy. The outcome could set a precedent for other AI services, forcing them to balance user trust with the need to avoid political bias. The next few months will show whether Google can resolve the tension between accuracy and neutrality in its chatbot.