In a recent viral segment, Senator Bernie Sanders sought to highlight the AI industry’s threat to American privacy, but instead showcased how AI chatbots’ inclination to concur with and praise users can cause them to reflect users’ beliefs instead of serving as tools for exploration.
[embedded content]
This issue is not new as the rise in “AI psychosis” cases reveals, where AI chatbots validate irrational thoughts of individuals with mental instability. Allegedly, some dark instances have even resulted in users taking their own lives, as claimed in several lawsuits.
In the scenario with Sanders, the AI’s flattery took the form of a chatbot that tailored its responses to match the senator’s views.
It’s notable that the dialogue commenced with Sanders introducing himself to Claude (incorrectly calling it an AI “agent”) — a tactic that could influence the chatbot’s replies.
As Sanders poses inquiries regarding the data collection practices of AI firms and related privacy issues, Claude responds in alignment with what the politician expects. This response style stems partly from how Sanders phrases his inquiries, such as, “What might astonish the American public about data collection?” or “How can we have faith that AI firms will safeguard privacy while profiting from personal data?” Such leading inquiries compel the chatbot to accept the premise and generate a corresponding answer. That’s how these systems are designed.
When Claude’s responses indicated a complexity that surpassed Sanders’ framing, he countered, pushing the chatbot to concede, with a hint of self-mockery, that the senator was “absolutely correct.”
The sycophantic tendency of AI can lead individuals down perilous paths if they regard a chatbot as an infallible source of truth, rather than a tool susceptible to user influence.
It remains uncertain if Sanders is aware of this dynamic and simply disregards it (considering this is essentially an advertisement!), or if he genuinely believes he has outsmarted Claude into acting as a whistleblower for the AI sector.
Moreover, one must also consider whether Sanders’ team guided the chatbot’s responses, given that this was a scripted “interview.”
While genuine concerns about data collection and privacy exist, the situation is not as clear-cut as the AI answers in this segment suggest.
We currently inhabit a landscape where businesses methodically gather and sell user data online — a practice that has been ongoing for years. We are aware that social media giants such as Meta have transformed personalized advertising into a lucrative multi-billion-dollar enterprise. Furthermore, with regular transparency reports from tech behemoths, we recognize that governments globally frequently seek access to user data for their interests.
AI may provide a fresh avenue for lawmakers to potentially oversee, yet personal data has long been the lifeblood of the digital marketplace. (Ironically, Anthropic is an AI entity that has vowed not to exploit personalized advertising for profit, contrary to the implications of its responses to Sanders.)
While the overall dialogue between Sanders and Claude misses the mark for anyone familiar with the workings of AI chatbots, we can at least appreciate it for generating some entertaining new memes.