Balancing promise and concern in AI therapy: a critical perspective on early evidence from the MIT-OpenAI RCT

The emergence of AI therapy chatbots has the potential to reduce the widening
gap between the huge demand for psychological support today and the limited
availability of professional care. However, some scholars and clinicians are warning
that the integration of these chatbots could paradoxically lead to negative outcomes,
such as dependence, loneliness, and depression. Recently, a joint research team
from MIT and OpenAI conducted a four-week Randomized Controlled Trial (RCT),
reporting that “while participants on average were less lonely after the study…
extended daily interactions with AI chatbots can reinforce negative psychosocial
outcomes”. Considering the substantial public and academic attention that followed
the preprint publication of this RCT, it is crucial to examine the strength of the
evidence and the validity of its interpretation before drawing firm conclusions. In
this commentary, we offer a careful and appreciative review of this well-designed
and timely study. Nonetheless, we argue that due to key analytical limitations, the
findings do not substantiate claims of harmful effects. Given the transformative
potential of AI-based interventions, we urge caution in interpreting early findings
and offer practical guidance for future research.