Do you ever get the feeling that the people running the world are delulu? That the 1% are living in a completely different universe from the rest of us? You’re not the only one. Even some tech elites are starting to worry about their peers’ grasp on reality. “CEOs are uniquely prone to AI psychosis,” Aaron Levie, a co-founder of the enterprise cloud company Box, declared on X last month. His reasoning for this? “They’re sufficiently distant from the last mile of work that still has to happen to generate most value with AI. So when they play with AI, they see the happy path results, often not considering the next 10 or 20 things that have to happen to get sustainable results from agents.”
In other words: CEOs are so high up the food chain that they don’t understand the human labour that goes into turning an error-riddled AI creation into something that functions properly in a business context. They are desperate to replace their annoying and expensive human labour with compliant AI models, but grossly overestimate what the technology can do. Meanwhile, the industry is rushing out overhyped AI solutions without properly stress-testing them.
This collective euphoria has resulted in some predictable disasters. In April, an AI coding agent powered by Anthropic’s Claude went berserk and deleted a company called PocketOS’s entire production database, along with backups. Jeremy Crane, PocketOS’s founder, later mused on X that this sort of failure was “inevitable” because the industry is “building AI-agent integrations into production infrastructure faster than it’s building the safety architecture to make those integrations safe”. To recall Facebook’s old mantra: it’s moving fast and breaking things.
Those things include our brains and grasp on reality. There’s a viral quote floating around the internet that quips: “The dumbest person you know is currently being told ‘You’re absolutely right!’ by ChatGPT.” Our tech overlords designed AI chatbots to be obsequious because it’s good business: having your opinions and feelings constantly validated increases user engagement.
But what else does this constant flattery do? While we’re still figuring that out, early studies aren’t reassuring. Research published in the Lancet Psychiatry in March found chatbots can encourage delusional thinking, particularly in people already vulnerable to developing psychotic symptoms. And a recent study from Stanford computer scientists found LLM “sycophancy can undermine users’ capacity for self-correction and responsible decision-making”. There is a pressing need, the study stresses, “to address AI sycophancy as a societal risk”.