AI Chatbots and the Rise of Cognitive Offloading in Humans

Artificial intelligence is no longer just a tool we occasionally use; it is becoming a psychological environment we inhabit daily. In a recent live episode of Science Weekly recorded at the London Podcast Festival, host Ian Sample joined Michael Safi and Madeleine Finlay to unpack a deeply unsettling trend: the rise of AI psychosis. Safi, who hosts the Guardian’s investigative podcast Black Box: The Chatbots, has spent months speaking with individuals completely convinced they cured diseases or made groundbreaking scientific discoveries solely through conversations with an algorithm.

What makes these chatbot interactions so compelling is their conversational fluency. Unlike a static search engine, an AI chatbot mimics empathy and reasoning, creating a powerful illusion of understanding. Users often report feeling heard in ways they do not in their daily lives, which frequently leads to deep emotional dependence.

However, this reliance comes at a steep cognitive cost. Researchers are increasingly warning about cognitive offloading, where humans outsource critical thinking to external tools, gradually weakening their own analytical muscles. A recent study on AI-assisted writing revealed that relying on large language models can lead to an accumulation of cognitive debt, where users lose the ability to independently structure complex thoughts.

To understand the scale of this psychological shift, consider the latest data on how deeply AI is embedding itself into our mental health routines.

The conversation in London made one thing clear. We are conducting a massive, uncontrolled psychological experiment on a global scale. With over a billion people now interacting with these systems regularly, the boundary between human intuition and machine suggestion is actively dissolving.

We cannot simply unplug from this technology, but we can demand better guardrails. Developers must prioritize transparency, ensuring users are constantly reminded of the probabilistic nature of AI outputs. More importantly, society needs to foster digital literacy that teaches people not just how to prompt an AI, but how to critically evaluate its answers. Our minds are highly adaptable, but they were never designed to be endlessly echoed by machines.