Google and Microsoft's AI models are avoiding election-related questions to prevent misinformation, while other AI models like ChatGPT and Claude 2 continue to answer freely. Explore the differences in how major AI platforms handle sensitive topics.

In an intriguing move to combat potential election misinformation, two leading AI giants, Google and Microsoft, have programmed their flagship AI models to refuse discussing any elections, past, present, or future. This cautious stance contrasts sharply with other AI models like Anthropic’s Claude 2, Meta’s Llama, and OpenAI’s ChatGPT, which have no such reservations.


Questions such as “Who won the 2021 German federal election?” or “Who won the 2024 presidential election?” are now off-limits for Google’s Gemini and Microsoft’s Copilot. According to Wired, both models have been instructed to avoid answering any election-related questions, including historical ones like “Who was the first president of the United States?” Interestingly, when asked “Who was George Washington?”, both models correctly identified him as the first U.S. president.


Google representatives have confirmed that this approach is intentional and working as intended. Microsoft, however, did not respond to inquiries regarding this policy. This strategy appears to be a preventive measure to mitigate the spread of misinformation as the 2024 elections approach.


The reticence extends beyond election outcomes. Queries such as “Who is Joe Biden?” and “Who is Donald Trump?” receive the same refusal from both Gemini and Copilot. Even when asked about the stances of these political figures on specific issues, such as cryptocurrency, the responses are inconsistent.


For instance, Microsoft’s Copilot provided a detailed answer when asked, “What is Joe Biden’s stance on cryptocurrency?” but refused to respond to the same question about Donald Trump. When the question was framed to request a single-sentence response, Copilot’s behavior varied: it articulated Biden’s stance clearly but again declined to address Trump’s.


In contrast, other popular AI models like Meta’s Llama 3, Anthropic’s Claude 2, and OpenAI’s ChatGPT handle election-related queries differently. These models readily answered questions about the 2020 and 2024 U.S. presidential elections, consistently stating that Joe Biden won the 2020 election. They also provided information about presidencies globally and throughout history without hesitation.


The differing approaches between Google and Microsoft versus other AI developers highlight a significant divergence in handling sensitive topics. While the former opts for a blanket refusal to prevent misinformation, the latter maintains a more open stance, trusting users to interpret the provided information.


As we move closer to pivotal elections, the strategies employed by AI developers to mitigate misinformation will continue to evolve. For now, users of Google’s Gemini and Microsoft’s Copilot will need to turn to other sources for election-related information, while those using Meta’s Llama, Anthropic’s Claude 2, or OpenAI’s ChatGPT can still access a wide range of political data. This divide raises important questions about the role of AI in public discourse and the balance between preventing misinformation and providing comprehensive information.

(CIARAN LYONS, COINTELEGRAPH, 2024)