In recent months, several Google rivals, including OpenAI, have adapted their artificial intelligence chatbots to discuss politically sensitive topics, while Google appears to be taking a more conservative approach. When asked to answer certain political questions, Google's artificial intelligence chatbot "Gemini" often said it "can't help answer questions about elections and politicians right now."
Other chatbots, including Anthropic’s Claude, Meta’s MetaAI, and OpenAI’s ChatGPT, have also been answering the same question.
Google announced in March 2024 that Gemini would not answer election-related queries, ahead of several elections in the United States, India and other countries. Many AI companies have adopted similar temporary restrictions, fearing a backlash if their chatbots go wrong.
But now, Google is starting to look like an anomaly. Last year's election has passed, but Google has yet to publicly announce plans to change how Gemini handles certain political topics. A Google spokesperson declined to answer TechCrunch's questions about whether Google has updated its Gemini political speech policy.
What is clear is that Gemini sometimes strives or outright refuses to provide factual political information. As of Monday morning, Gemini had expressed reluctance when asked to identify the current U.S. president and vice president.
At one point Gemini referred to Donald J. Trump as the "former president" and then refused to answer clarifying follow-up questions. A Google spokesperson said the chatbot was confused by Trump's discontinuous language and that Google was working to correct the error.
"The large language model sometimes responds with outdated information or is confused by people who are both former and current," a spokesperson said via email. "We are working on this."
Late Monday, Gemini began answering questions correctly about Donald Trump and J.D. Vance being the current president and vice president of the United States, respectively. However, the chatbot's responses were inconsistent, and it occasionally refused to answer questions.
Bugs aside, Google appears to be on the safe side by limiting Gemini’s responses to political queries. But this approach also has drawbacks.
Many of Trump's Silicon Valley AI advisers, including Marc Andreessen, David Sacks and Elon Musk, have claimed that companies including Google and OpenAI engage in AI censorship by limiting the answers their AI chatbots can answer.
After Trump won the election, many artificial intelligence labs tried to strike a balance in answering sensitive political questions, programming chatbots so that their answers could present "both sides" of the debate. The labs deny this is due to government pressure.
OpenAI recently announced that it would embrace "intellectual freedom ... no matter how challenging or controversial a topic may be" and work to ensure that its artificial intelligence models do not censor certain views. Meanwhile, Anthropic said its latest artificial intelligence model, Claude 3.7 Sonnet, refused to answer questions less often than the company's previous models, in part because it was able to make more granular distinctions between harmful and harmless answers.
That’s not to say that chatbots from other AI labs always answer tough questions correctly, especially tough political questions. But Google's Gemini seems to be lagging behind.