Towards an Artificially Empathic Conversational Agent for Mental Health Applications: System Design and User Perceptions

290

198

Background Many potential benefits for the uses of chatbots within the context of health care have been theorized, such as improved patient education and treatment compliance. However, little is known about the perspectives of practicing medical physicians on the use of chatbots in health care, even though these individuals are the traditional benchmark of proper patient care. Objective This study aimed to investigate the perceptions of physicians regarding the use of health care chatbots, including their benefits, challenges, and risks to patients. Methods A total of 100 practicing physicians across the United States completed a Web-based, self-report survey to examine their opinions of chatbot technology in health care. Descriptive statistics and frequencies were used to examine the characteristics of participants. Results A wide variety of positive and negative perspectives were reported on the use of health care chatbots, including the importance to patients for managing their own health and the benefits on physical, psychological, and behavioral health outcomes. More consistent agreement occurred with regard to administrative benefits associated with chatbots; many physicians believed that chatbots would be most beneficial for scheduling doctor appointments (78%, 78/100), locating health clinics (76%, 76/100), or providing medication information (71%, 71/100). Conversely, many physicians believed that chatbots cannot effectively care for all of the patients’ needs (76%, 76/100), cannot display human emotion (72%, 72/100), and cannot provide detailed diagnosis and treatment because of not knowing all of the personal factors associated with the patient (71%, 71/100). Many physicians also stated that health care chatbots could be a risk to patients if they self-diagnose too often (714%, 74/100) and do not accurately understand the diagnoses (74%, 74/100). Conclusions Physicians believed in both costs and benefits associated with chatbots, depending on the logistics and specific roles of the technology. Chatbots may have a beneficial role to play in health care to support, motivate, and coach patients as well as for streamlining organizational tasks; in essence, chatbots could become a surrogate for nonmedical caregivers. However, concerns remain on the inability of chatbots to comprehend the emotional state of humans as well as in areas where expert medical knowledge and intelligence is required.

Section: Introductionmentioning

confidence: 99%

Physicians’ Perceptions of Chatbots in Health Care: Cross-Sectional Web-Based Survey

Palanica¹,

Flaschner²,

Thommandram³

et al. 2019

290

198

“…Some systems also leverage unconstrained natural language input to index health advice but do not frame the interaction as a conversation. Kokobot is a conversational agent that facilitates interactions among users of an online peer-to-peer social support platform designed to promote emotional resilience [ 27 ]. Users are prompted to describe stressful situations and associated negative thoughts, and Kokobot responds to these submissions by retrieving and repurposing statements from a corpus of supportive statements previously submitted to Koko by other users.…”

Section: Introductionmentioning

confidence: 99%

Patient and Consumer Safety Risks When Using Conversational Assistants for Medical Information: An Observational Study of Siri, Alexa, and Google Assistant

Bickmore¹,

Trinh²,

Ólafsson³

et al. 2018

211

145

BackgroundConversational assistants, such as Siri, Alexa, and Google Assistant, are ubiquitous and are beginning to be used as portals for medical services. However, the potential safety issues of using conversational assistants for medical information by patients and consumers are not understood.ObjectiveTo determine the prevalence and nature of the harm that could result from patients or consumers using conversational assistants for medical information.MethodsParticipants were given medical problems to pose to Siri, Alexa, or Google Assistant, and asked to determine an action to take based on information from the system. Assignment of tasks and systems were randomized across participants, and participants queried the conversational assistants in their own words, making as many attempts as needed until they either reported an action to take or gave up. Participant-reported actions for each medical task were rated for patient harm using an Agency for Healthcare Research and Quality harm scale.ResultsFifty-four subjects completed the study with a mean age of 42 years (SD 18). Twenty-nine (54%) were female, 31 (57%) Caucasian, and 26 (50%) were college educated. Only 8 (15%) reported using a conversational assistant regularly, while 22 (41%) had never used one, and 24 (44%) had tried one “a few times.“ Forty-four (82%) used computers regularly. Subjects were only able to complete 168 (43%) of their 394 tasks. Of these, 49 (29%) reported actions that could have resulted in some degree of patient harm, including 27 (16%) that could have resulted in death.ConclusionsReliance on conversational assistants for actionable medical information represents a safety risk for patients and consumers. Patients should be cautioned to not use these technologies for answers to medical questions they intend to act on without further consultation from a health care provider.

“…Their efficacy is limited because digital tools often fail to motivate and engage users [ 13 , 14 ]. They also currently lack the human-level intelligence required to address nuanced problems [ 15 , 16 ]. It appears that until realistic artificial intelligence is available, many people require human-delivered interventions to meet their preferences, engage them, and respond to their unique concerns.…”

Section: Introductionmentioning

confidence: 99%

Nonprofessional Peer Support to Improve Mental Health: Randomized Trial of a Scalable Web-Based Peer Counseling Course

Bernecker¹,

Williams²,

Caporale‐Berkowitz³

et al. 2020

Background Millions of people worldwide are underserved by the mental health care system. Indeed, most mental health problems go untreated, often because of resource constraints (eg, limited provider availability and cost) or lack of interest or faith in professional help. Furthermore, subclinical symptoms and chronic stress in the absence of a mental illness diagnosis often go unaddressed, despite their substantial health impact. Innovative and scalable treatment delivery methods are needed to supplement traditional therapies to fill these gaps in the mental health care system. Objective This study aims to investigate whether a self-guided web-based course can teach pairs of nonprofessional peers to deliver psychological support to each other. Methods In this experimental study, a community sample of 30 dyads (60 participants, mostly friends), many of whom presented with mild to moderate psychological distress, were recruited to complete a web-based counseling skills course. Dyads were randomized to either immediate or delayed access to training. Before and after training, dyads were recorded taking turns discussing stressors. Participants’ skills in the helper role were assessed before and after taking the course: the first author and a team of trained research assistants coded recordings for the presence of specific counseling behaviors. When in the client role, participants rated the session on helpfulness in resolving their stressors and supportiveness of their peers. We hypothesized that participants would increase the use of skills taught by the course and decrease the use of skills discouraged by the course, would increase their overall adherence to the guidelines taught in the course, and would perceive posttraining counseling sessions as more helpful and their peers as more supportive. Results The course had large effects on most helper-role speech behaviors: helpers decreased total speaking time, used more restatements, made fewer efforts to influence the speaker, and decreased self-focused and off-topic utterances (ds=0.8-1.6). When rating the portion of the session in which they served as clients, participants indicated that they made more progress in addressing their stressors during posttraining counseling sessions compared with pretraining sessions (d=1.1), but they did not report substantive changes in feelings of closeness and supportiveness of their peers (d=0.3). Conclusions The results provide proof of concept that nonprofessionals can learn basic counseling skills from a scalable web-based course. The course serves as a promising model for the development of web-based counseling skills training, which could provide accessible mental health support to some of those underserved by traditional psychotherapy.