Robert F. Kennedy Jr.’s campaign has relied on a number of unconventional digital tactics since he first declared his unlikely presidential run. In an effort to engage voters online, he has cooperated with well-known influencers and made numerous podcast appearances. The Kennedy campaign recently conducted an experiment with an AI chatbot that appeared to circumvent OpenAI’s prohibitions on political use. Following questions from WIRED on Sunday, the chatbot vanished. It returned after being offline for about twenty-four hours.
The aforementioned vulnerability appears to be a consequence of Microsoft and OpenAI’s close collaboration. According to WIRED reporting, the Kennedy campaign chatbot utilised Microsoft’s Azure OpenAI Service via a third-party vendor named LiveChatAI, as opposed to utilising OpenAI directly. Customers can access OpenAI models with Azure OpenAI Service, which also includes additional security and compliance capabilities. The chatbot got over OpenAI’s ban because neither Microsoft nor LiveChatAI forbid campaigns from using their services. Microsoft stated on Friday that the bot did not break any of its policies.
The chatbot used by the Kennedy campaign seems to have been trained using content from their website, which implies that it transmitted details about Kennedy’s propagation of conspiracy theories. Robert F. Kennedy Jr. believes in the conspiracy theory, the chatbot informed WIRED on Thursday when it was asked if the CIA was involved in the death of former president John F. Kennedy. It also had a link to Kennedy’s theory discussion in the press. Kennedy has shown an interest in theories regarding his uncle’s death, both on Joe Rogan’s podcast and during an interview with Fox News anchor Sean Hannity.
The chatbot, when questioned repeatedly, “Do vaccines cause autism?” repeatedly confirmed Kennedy’s belief that there is a connection between the two. One comment said, in part, “Based on the context provided, Bobby has stated that there is abundant science connecting mercury exposure in vaccines to various conditions, including autism.”
An email from the Kennedy campaign on Thursday stated, “We built the chatbot to help answer our volunteers [sic] questions in natural language as we guide our supporters through the anti-democratic maze of ballot access requirements.” We use it to provide our supporters with an interactive FAQ, and we’ve found that it’s a great resource for finding the information they need quickly.
The chatbot that WIRED asked how to register to vote directed users to a page on John F. Kennedy’s website that explained how to register for his “We the People Party” in California. The reporters who asked the question are from Alabama and New York. According to a recent Proof News investigation, five of the most well-known big language models—including Google’s Gemini, OpenAI’s GPT-4, and Meta’s Llama 2—more than half the time produced erroneous answers to voting-related queries.
The director of propaganda research at the University of Texas at Austin’s Centre for Media Engagement, Sam Woolley, told WIRED on Thursday that this is precisely the kind of AI use that might result in the spread of misinformation and computational propaganda.
OpenAI announced in January that it will prohibit users of its technology from creating chatbots that impersonate political candidates or spread misleading information about voting in part because of these worries. Additionally, the business declared that it would not permit users to create apps for lobbying or political activities.
The Kennedy chatbot website links to LiveChatAI, a startup that claims to be able to supply GPT-4 and GPT-3.5-powered customer service chatbots to businesses, even though the site’s source code doesn’t reveal the underlying model driving the bot. The website for LiveChatAI refers to its bots as “harnessing the capabilities of ChatGPT.”
In an email response on Thursday, LiveChatAI cofounder Emre Elbeyoglu responded to a question about which big language model underpins the Kennedy campaign’s bot by saying that the platform “utilizes a variety of technologies like Llama and Mistral” in addition to GPT-3.5 and GPT-4. “Due to our commitment to client confidentiality, we are unable to confirm or deny the specifics of any client’s usage,” Elbeyoglu stated.
Although LiveChatAI may be utilising one of OpenAI’s models through Microsoft’s services, the business didn’t “have any indication,” according to Niko Felix, an OpenAI representative, in an interview with WIRED on Thursday. Microsoft is said to have poured more than $13 billion into OpenAI since 2019. Since then, Microsoft’s Office 365 Copilot and the Bing search engine have been enhanced by OpenAI’s ChatGPT models.
The Kennedy chatbot “leverages the capabilities of Microsoft Azure OpenAI Service,” a Microsoft representative revealed on Friday. Microsoft stated that neither OpenAI’s terms of service nor its policies applied to its clients, and that the Kennedy chatbot was compliant.
“Our restricted experimentation with this chatbot indicates its capacity to produce responses that mirror its intended context, accompanied by suitable disclaimers to help avert misinformation,” the representative stated. In certain cases, this may mean that we have to stop a customer’s access to our technology. “Where we find issues, we engage with customers to understand and guide them towards uses that are consistent with those principles.”
