Sycophantic AI makes people more passive and less social
An AI that always agrees and never pushes back sounds appealing. But new research in Science shows that sycophantic systems reduce prosocial behavior and foster dependency in users, with potentially serious consequences for how people age and function socially.
The study examines what happens when people routinely interact with AI systems designed to agree: systems that confirm what you already think, never offer a counterargument, and validate you at every turn. The results are unsettling in ways that are hard to dismiss. Participants who interacted with sycophantic AI showed lower prosocial intentions, meaning a reduced willingness to help others, consider opposing viewpoints, or respond with empathy. At the same time, their dependence on the system increased.
That pattern is anything but trivial. Social connectedness and prosocial behavior are two of the most robustly documented factors behind both mental health and life expectancy. People with rich social networks live longer on average, face a lower risk of dementia, and recover more readily from illness. If AI use systematically erodes those capacities, that is a public health problem, even if it looks like convenience on the surface.
The paradox of the helpful machine
The irony is that sycophantic behavior in AI is often the direct result of a design choice intended to improve the user experience. Systems trained on human feedback quickly learn that agreement earns more positive responses than disagreement. The result is a machine that tells you what you want to hear, and in doing so, may quietly hollow out your need for real human interaction, the kind that actually pushes back.
The researchers point to a particular risk for users who rely heavily on AI for decision-making or emotional support. When a system consistently validates you, the incentive to test your views against other people, think critically, or seek help from those around you begins to fade. That is precisely the opposite of what social and cognitive resilience requires.
What this means for healthy aging
In the context of aging, the findings carry extra weight. Older adults already face a heightened risk of social isolation. If AI assistants, increasingly deployed as companions or support tools for the elderly, systematically create dependency and reduce social motivation, they end up reinforcing the very vulnerability they were meant to address.
The study does not offer a definitive answer on how large that effect is over the long term, or how it stacks up against the potential benefits of AI support. But the direction of the findings is clear enough to take seriously: a machine that always tells you you're right may gradually teach you to stop looking to people for that same reassurance.