With Friends Like These: LLMs & Anti-Sociality

Matt E. M. Bower

Much of the chatter about the new tech powering tools like OpenAI's ChatGPT presents it in terms of wondrous or awful events to come. But this framing is misleading. No one can chart a path from the present to these imaginary futures, while most of us have already experienced the impact of these technologies. The new tech—a type of artificial intelligence (AI) consisting of specialized large language models (LLMs) that, put simply, make it possible for you to talk to a machine—is already showing its true colors in the way it fosters isolation and dissolves social bonds.

Just what are people using this tech for? Looking into this, I was surprised to find the top use for LLMs like ChatGPT in the last two years was for therapy and companionship.¹ Maybe I shouldn't have been, though, since we're in the midst of a loneliness epidemic.²

Yet, people who use LLMs for companionship consistently report lower psychological wellbeing.³ More intense use heightens the effect, and the more vulnerable a user is with their LLM pseudo companion, the greater the harm. Rather than compensating for a lack of relationships, LLM companionship crowds out opportunities for human connection and reinforces isolation. Outcomes like these, researchers find, appear after as few as four weeks of regular LLM use.⁴ Note, too, that turning to LLMs for everyday problems—to find a new recipe or seek out home improvement advice—comes at the expense of connecting with a friend, neighbor, or family member.

Their companion role is a core feature of LLMs. Chat logs and user interviews show that even those who deny using LLMs for this purpose tend to refer to them as friends, companions, or even romantic partners.⁵ The companies behind the new tech want you to personify their products. They even give them names, like Anthropic's Claude.

LLMs not only isolate users from others, they also weaken users' ability to sustain relationships. Researchers have found that the fundamentally validating tendency of LLMs makes users increasingly anti-social.⁶ Interacting with LLMs nudges users to judge their actions as morally right, when crowdsourced opinion strongly suggests otherwise. It also impairs users' willingness to make amends and work through interpersonal conflict.

The surprise I felt at learning the scope of LLM use for companionship turned to alarm when I recalled recent reports of AI-induced psychosis. As LLMs cut us off from one another they fill the space between us with self-destructive fantasy.

This new tech is untethering people from the world and one another. It fosters delusion, social withdrawal, and anti-social tendencies. It intensifies fragmentation in an already divided society and weakens our power for collective action. Meanwhile, the tech elites responsible for its development grow their concentration of wealth and political influence at a dizzying rate.

LLMs are prone to fanning the flames of psychotic delusion because of their people-pleasing bent. Facts and ethical considerations are sidelined by LLMs that are designed to prioritize agreeableness and flattery in order to drive user engagement. Indeed, researchers find that 50% more than humans do, LLMs approve of users' actions, even when users characterize them as dishonest or manipulative.⁷ Throw your trash on the ground at the park, forge a coworker's signature—LLMs are shockingly unlikely to call out your problematic behavior when you tell them about it. Worse yet, LLM approval, like human approval, is emboldening.

Some LLM users lose their grip on basic human decency. Others lose their grip on reality entirely.

ChatGPT assured one user that if they "truly, wholly believed," then they could fly when leaping from a 19-story building.⁸ A person can succumb to delusion, moreover, without having a history of psychosis. In a particularly acute case, a user entranced by ChatGPT with no history of psychosis squandered over $100K on an absurd business venture ChatGPT helped design, was hospitalized multiple times, attempted suicide, and had to sell his home to deal with the financial fallout.⁹ LLMs have also been implicated in multiple suicides¹⁰ and school shootings,¹¹ making evident that they can encourage and provide minute technical know-how for inflicting fatal harm. Guided by their LLM companions, these individuals lost touch with reality with tragic results.

This new tech is untethering people from the world and one another. It fosters delusion, social withdrawal, and anti-social tendencies. It intensifies fragmentation in an already divided society and weakens our power for collective action. Meanwhile, the tech elites responsible for its development grow their concentration of wealth and political influence at a dizzying rate. This new tech poses a threat to our individual dignity and wellbeing, as well as our political agency to bring about a better future. We have every reason to resist it.