Every few weeks, we are met with a new publication that warns against the effects of AI on psychological functioning, from “blandification”, “cognitive offloading”, and “AI sycophancy”, to decreasing neurological activity and “colonizing” how we think. Around the same time I last published on AI here, I came across a beautiful intervention by AI, buried within an even more beautiful story about whales’ social behaviour. It got me thinking: is psychology biased towards problematizing AI?
A Little History
For starters, it is worth considering the field itself and its origins. Psychology has long been wed to subfields such as “abnormal psychology” and psychopathology that focus on pathology and problems. In his book on the history of psychoanalysis, George Makari notes that the field was initially developed through the study of animals, children, and psychopathology. Granted, figures such as Wundt were less focused on this, and the work of William James was even more positively inclined – asking more how psychology could help humans flourish – the field of ‘positive psychology’ did not emerge until the 1990s. The very fact that a subfield explicitly focused on the “positive” is arguably telling.
Critical psychologist Teo cites Foucault: “There are many examples demonstrating that psychology in fact did not solve problems but produced problematizing in which neutral issues were turned into highly problematized objects.”
While Foucault was speaking to issues such as race rather than technology, the point is relevant. Psychologists can counter, saying their cautions about AI are rooted in evidence-based findings.
Small and Big Changes
The relatively young field of behavioral economics has contributed several key ideas to psychology. Among them is the finding that apparently small factors can influence at scale, whether it is which option is a default among a set of choices, or the day of the week influencing political decision-making.
While behavioral economics highlights how small changes can impact, the work of those such as Jonathan Haidt makes a case that bigger changes, such as the mass rollout of the smartphone, can have a monumental impact. Haidt is one of the leading psychologists who is pushing for policy-driven restrictions on smartphone use. While the dynamics of smartphones and social media differ from AI and LLMs, they both represent technology that delivers profound psychological and social change.
Behavioral economics has also deepened our knowledge of cognitive biases that may contribute to how the field approaches this topic (or any other). Instinctively, a group of people who chose to pursue a field that focuses on human beings, and at the clinical end of the profession, on working collaboratively with other human beings to alleviate challenges and distress, are also likely to be distrustful of inanimate machines, especially when they feign animacy.
The (Mis)uses of Tech
LLMs are being touted as tools to promote psychological well-being, by way of serving as counsellors and caring “partners.” While the extreme end of this “partnership” has seen phenomena such as “AI psychosis,” at the less critical end are reports that LLMs cannot match the standard of human-to-human engagement. Li and colleagues (2026) report that messaging another human significantly decreased loneliness (p=0.006) and isolation (p=0.001). A control group that wrote in a journal and another experimental group that messaged a chatbot did not experience this benefit. Those who messaged another human also experienced significant benefits to their mood.
An unsettling element lies in the propensity for options initially framed as a last resort to eventually becoming the default for purposes of cost and convenience. In his history of the psychiatric asylum, Davies mentions that they were initially seen as last ports of call, but quickly became the go-to option. In our own lifetimes, we see psychoactive medication being more readily dispensed than talking therapy, likely due to cost-effectiveness rather than considerations of treatment efficacy. Once “chatbots as psychological treatment” becomes an option, it could disincentivize states and policymakers to find more robust solutions. That LLMs have been made available en masse without a full account of their psychological impact, let alone robust longitudinal evidence, is relevant.
Back to the communicative behavior of whales and AI-enabled discoveries, perhaps this is where AI can make a hugely positive impact. AI trawled through masses of data to identify patterns that would have been difficult for a team of humans to find independently. It shines when processing inanimate, vast data. Discovering animal behavior can be genuinely enriching for humans and can catalyze nurturing environmental attitudes. As for therapy and human-to-human collaborations, AI could play a supporting role on several levels, but centering it as the heart of the solution seems misguided.