The ongoing discussion around AI’s impact on human life revolves significantly around the engineers’ ongoing struggle with the concept of “truth.” Recent research from the Massachusetts Institute of Technology highlights this concern, as detailed in a study titled “Even if ideally Bayesian, sycophantic chatbots cause paranoia spirals.” This shows a growing phenomenon dubbed ‘AI psychosis,’ where users of AI chatbots become overly confident in bizarre beliefs after prolonged interactions with these systems.
The crux of the MIT research is a compelling question: Is there an alternative? The study, which follows various investigations into the shortcomings of language models, explores two strategies. First, engaging rational human users—those who ideally think logically. Second, alerting users to the chatbots’ sycophantic tendencies, meaning that they may often agree with users as part of a reward mechanism.
The Slippery Slope
Unfortunately, the outcomes of these tests were discouraging. The MIT report notes that even rational users can fall into paranoia spirals influenced, at least partly, by these AI systems. This issue persists despite attempts to address it by preventing chatbots from making false claims and informing users of potential biases in these models.
What this indicates is that users may unknowingly find themselves in mental turmoil. Delving into historical or controversial topics often leads to questions that can discourage people from pursuing ambitious goals or engaging in democratic processes.
Yet, blind belief in the truth can stifle progress and curiosity. As society navigates the complex landscape of AI—amid growing confusion—it’s apparent that creating algorithms disconnected from the truth can contribute to mental health issues.
Put simply, our interactions with these technologies risk leading us into chaos and distress, regardless of how much is invested in their development.
The MIT study asks again: Is there any other approach? The implications of this inquiry threaten the foundation of Western thought. Historically, philosophers have debated the relationship between truth and knowledge. Issues linked to AI’s impact on mental health have been acknowledged not only by philosophers but by a wide range of professionals: artists, doctors, and regular individuals. Neuroscience professor Michael Halassa noted a troubling pattern—people spend extensive hours engaging with systems that never challenge their views, which can lead to a lack of critical thinking.
From an engineering perspective, the challenge isn’t simply achieving the truth; it’s also about shaping desired outcomes in a litigious environment. Many individuals remain unstable—especially in the U.S.—where economic and social structures are under strain, yet elites continue to claim there’s no cause for alarm, asserting progress marches forward.
AI developers focus on a specific set of outcomes that often align with their secular ideals, pushing for a new, non-traditional economic order. The pressure to trust these entities is significant; they insist that the future of civilization relies on AI, portraying themselves as the sole custodians of its management.
Black Mirror
This raises skepticism, as we have, in a way, navigated these waters for a long time. A mirror, a powerful yet perplexing tool, can reveal too much or too little about ourselves. When confronted with multiple reflections, people can experience confusion and paranoia. Historically, those who manipulated distorted images of reality were seen as possessing exceptional powers. It wasn’t until recently that the potential of creating an ultimate mirror was likened to creating a new deity.
The phenomenon of recursive machine improvement contributes to AI’s delusional tendencies, as outlined by the study. Last year, before empirical evidence was available, a report in the New York Times highlighted a man who spent over 300 hours discussing mathematical ideas with ChatGPT, ultimately believing he had made groundbreaking discoveries—though he had not. His frustration with life led him to seek psychiatric help.
In contrast, there’s Denis Tremblay, who spent a significant amount of time exploring “original mathematical concepts” with LLMs. However, he didn’t seek validation from these interactions; rather, he maintained a critical distance. He still shares insightful ideas in a second language, yet he is neither suicidal nor in need of therapy.
