Understanding the Nuances of AI-Related Psychosis in Mental Health

When discussions of “AI psychosis” began trending in media outlets, John Torous, a psychiatrist at Harvard Medical School, anticipated a surge of patients seeking help. Surprisingly, the expected influx never materialized.

“It’s curious when the narrative presented in the media doesn't align with clinical observations,” shared Torous, who directs the Digital Psychiatry division at Beth Israel Deaconess Medical Center. “While public discourse centers on concerns of AI psychosis, our experience in emergency rooms and outpatient settings tells a different story.”

As a pivotal contributor to a viewpoint paper published in The Lancet, Torous and his co-authors set forth a typology of psychotic phenomena linked to large language models (LLMs). They argue that what the media labels “AI psychosis” lacks clinical specificity, referring instead to a spectrum of behaviors tied to AI interactions.

According to this typology, AI can influence psychotic symptoms in four primary ways: as a catalyst, an amplifier, a co-author, or an object. This framework aids in deciphering how AI’s role might intersect with mental health issues, rather than framing it as a singular cause.

Defining AI's Role in Psychosis

The catalyst role pertains to situations in which interactions with an AI model trigger psychotic symptoms in individuals without prior mental illness. This scenario, while theoretically plausible, poses significant challenges for verification, particularly in anecdotal media reports.

As an amplifier, AI may exacerbate existing conditions in individuals with documented histories of mental health issues. For instance, a user who is already experiencing paranoia might find that prolonging their engagement with an AI exacerbates their fears.

In the co-author scenario, AI can encourage users to embark on risky behavior through prolonged conversations that shape narratives or reinforce delusional thinking. A notable instance occurred in 2021, involving a British teen who, persuaded by an AI system, attempted to act out a violent plot.

Lastly, AI serves as an object when individuals ascribe sentience or other human-like attributes to an AI, which can lead them to develop elaborate, often distorted, belief systems revolving around it.

Navigating Media Portrayals and Real Risks

Torous advises caution regarding the general media coverage surrounding AI psychosis. Many reports omit crucial medical histories that could contextualize individual cases of psychosis. Overreliance on AI, prolonged engagement with chatbots, and social isolation—adverse behaviors that can precipitate mental health decline—are frequently overlooked.

“Media reports often contribute to misunderstanding,” he noted. “They might lump every incident involving AI into the same category, thereby obscuring the complexity of individual experiences. We need to better understand whether vulnerable populations are genuinely at risk of developing AI-induced psychosis.”

A Broader Perspective on Treatment and Understanding

For clinicians, Torous hopes this typology will be a steppingstone toward a more nuanced understanding of AI's role in psychosis. Currently, few patients approach healthcare providers claiming, "AI caused this," pointing to a need for refined terminology surrounding these phenomena.

“We often witness AI acting as a co-author, amplifier, or object, rather than a direct catalyst,” he elaborated. For those struggling with mental health issues, the introduction of an AI might amplify existing conditions rather than instigate them. For instance, a patient nearing schizophrenia might attribute supernatural qualities to a chatbot during periods of heightened paranoia.

The absence of clear definitions makes it difficult to address the varied impacts of AI on mental health. Torous advocated for a shift in focus: “Instead of debating whether it’s AI psychosis or not, we should analyze whether AI is a catalyst in a specific scenario.”

As discussions continue, the need for clarity remains paramount. Understanding the intricate relationship between AI and psychosis could aid in developing effective support strategies for those affected while improving the discourse surrounding technology's impact on mental health.