Chatbot Therapy - Help or Hindrance

Kadi Rae Smith, Portland State UniversityFollow

Subjects

ai, psychosis, therapy, chatbot, mental health, ai psychosis, chatbot therapy, ai therapy

Abstract

Introduction

Generative artificial intelligence (Gen-AI) chatbots are complex computer programs designed to mimic human conversations using programming that incorporates large language models (LLM) and natural language models (NLM) to simulate human interactions that are being rapidly integrated into everyday life across an increasing amount of online platforms. This article explores the applications and efficacy of chatbots for mental healthcare, from self-directed chatbot interactions to clinician-supervised implementation of chatbot-based therapeutic tools, weighing the benefits and limitations of this ever-expanding technology.

Methodology/Taxonomy

A literature review was conducted primarily using Web of Science, accessed through Portland State University, and Google Scholar, with specifications of peer reviewed articles addressing the uses of chatbots for therapeutic interventions beginning in 2022 when the first AI generative chatbot platform, ChatGPT, was launched. Additionally, a review of large-scale public facing news outlets, such as The Guardian and Time, was conducted in order to provide a qualitative, holistic view of chatbot therapy with personal anecdotal accounts from users. As chatbots are a relatively new tool, there is limited empirical information on their uses in medical applications but an ever-growing body of experiential knowledge shared by individuals who have engaged with AI generative chatbots in a self-help capacity.

Results/Patterns

A pattern of temporary relief in times of acute crisis was found in both anecdotal and empirical assessments of engaging with chatbots for therapeutic purposes. The most notable results reported by users engaging with chatbots for mental healthcare was an initial sense of safety when sharing private information under simulated anonymity, as well as satisfaction at receiving instantaneous responses. Long-term usage of chatbots in this capacity was more often associated with negative feelings and mental health degradation. Hyper-specialized AI generative therapeutic tools are actively being used, though none were found to be free and each was designed to be used in tandem with supervised, clinician-based therapeutic interventions rather than in a stand-alone capacity.

Conclusion

AI generative chatbots are as yet an unreliable and unpredictable means of self-propelled therapy, however, chatbots may provide short-term relief in times of acute mental health crisis. Ongoing global efforts are underway lobbying for a collective governance of AI generated tools in relation to individual and societal health. While this may eventually result in the standardization of AI regulations and the instillment of mandatory safeguards, there is no prevailing legal standard by which generative AI companies are being held accountable for the detrimental effects resulting from the use of their products by vulnerable populations. Specialized Gen-AI therapeutic tools may serve as a viable intermediate for aiding in mental health management when access to clinician-led interventions are not available or otherwise accessible. The standardization of user safeguards is imperative for mitigating the degradation of mental health stemming from prolonged chatbot use.