In a new viral video, Senator Bernie Sanders tried to show how the AI industry is a threat to the privacy of Americans, but ended up showing how the tendency of AI chatbots to agree and flatter their users can lead chatbots to become a mirror of users’ beliefs rather than a tool for discovery.
We’ve seen this problem among a growing number of people suffering from “AI psychosis,” which is where an AI chatbot reinforces the irrational thoughts and beliefs of a mentally unstable person. In some cases, this dark pattern has even led users to take their own lives, several cases have alleged.
In the Sanders case, the AI sycophancy was seen as an AI interviewer tailoring its responses to suit the politician.
It’s important to note that the interview begins with Sanders introducing himself to Claude (who he mistakenly calls an AI “agent”) – a move that could help influence the chatbot’s responses.
Then, as Sanders asks questions about AI companies’ data collection practices and other privacy concerns, Claude answers according to what the politician wants to hear. In part, that’s because of the way Sanders framed his questions, asking things like, “What would surprise the American people to know how that information is collected?” or “How can we trust AI companies to protect our privacy when they use people’s data to make money?” These leading questions force the chatbot to accept the premise of the question and come up with the right answer. That’s just how these things work.
And when Claude’s answer suggested that the topic was more complex or nuanced than Sanders had put it, Sanders would not agree, pushing the chatbot to admit, with an AI condescension, that the senator was “absolutely right.”
The sycophantic nature of AI is what can lead people down dangerous paths if they think that a chatbot is a source of universal truth, rather than a tool that can be influenced by its user.
It’s not clear if Sanders knows this is true and doesn’t care (because this is just an ad, after all!), or if he thinks he’s tricked Claude into being a whistleblower in the AI industry.
And of course, there’s also the question of whether Sanders’ team initiated the conversation in order to respond in a certain way, given that this was an on-stage “interview.”
While there are real concerns about data collection and privacy, things aren’t as bad as the AI responses in this video suggest.
We already live in a world where companies collect and sell online user data on a scale – and have for years. We know that social media giants like Meta have turned personalized ads into a multi-billion dollar money press. And thanks to the regular transparent reports of the tech giants, we know that governments around the world often request access to user data for their own purposes.
AI may represent a new way for lawmakers to regulate, but personal data has long fueled the digital economy. (Ironically, Anthropic is an AI company that has promised not to use personalized ads to make money, despite what its responses to Sanders might suggest.)
While the overall conversation between Sanders and Claude misses the mark for anyone who understands how AI chatbots work, we can at least credit it with giving us some great new memes.
