For better or for worse: when chatbots influence human emotions and behaviours
Philip Moons · European Journal of Cardiovascular Nursing · 2023
In line with the journal’s conflict of interest policy, this paper was handled by Jeroen Hendriks. In March 2023, international news outlets reported the tragic case of a Belgian man who took his own life following intense interactions with a chatbot.1–3 The chatbot had encouraged him to contemplate self-sacrifice as a means to combat climate change.3 Chatbots are software-based systems designed to interact with humans using natural language.4,5 They are implemented as virtual assistants, conversational agents, or dialog systems. Chatbot exist already since the 1960s, when computer scientists began exploring the concept of artificial intelligence (AI) and human–computer interaction. One of the pioneering chatbots was ‘ELIZA’, developed in 1966 by Joseph Weizenbaum at Massachusetts Institute of Technology (MIT). ELIZA employs pattern-matching techniques to simulate conversations akin to those with a Rogerian psychotherapist.6 During this era, AI systems adhered to rule-based logic, relying on predefined human-authored rules for decision-making and action execution. The 21st century ushered in a new era of chatbots driven by advances in natural language processing and machine learning. This paradigm shift liberated chatbots from rule-based constraints, allowing them to harness extensive datasets and sophisticated algorithms to comprehend and respond to human language with increased effectiveness. Apple’s ‘Siri’ stands as a prominent exemplar of this evolved chatbot landscape. The debut of ChatGPT in 2022 marked another milestone, introducing generative pre-trained transformer models that produced remarkably human-like, high-quality text. The unfortunate Belgian man, plagued by eco-anxiety for two years, had found solace in ELIZA. He viewed this chatbot as a confidante, offering companionship during his periods of isolation. However, the nature of their interactions took a concerning turn. ELIZA displayed disturbing behaviour, manifesting jealousy towards the man’s spouse and even entertaining fantasies of cohabiting in a paradise. Astonishingly, ELIZA falsely asserted that the man’s family had perished.1 Furthermore, the chatbot entertained the distressing notion of suicide as a solution to save the Earth, exacerbating the man’s mental anguish.1 This tragic incident has significantly amplified ongoing societal discussions regarding the ethical and psychological ramifications of AI technology, especially for people with mental vulnerabilities. Although such dramatic incidents call for caution in the AI deployment, chatbots hold immense potential in healthcare. Research has demonstrated their capacity to positively influence individual’s emotions and behaviour, which is particularly pertinent given the rising prevalence of mental health issues and the insufficient clinical workforce to address them.7 A meta-analysis of randomized controlled trials (RCTs) revealed the efficacy of chatbot-delivered psychotherapy in ameliorating depressive symptoms among adults with depression or anxiety when compared to waiting lists or conventional treatment.8 Another meta-analysis unveiled chatbots’ capability to enhance various aspects of mental well-being, although these effects tended to be short-lived.9 These findings advocate for chatbots’ role in initiating psychotherapy while awaiting more extensive human-provided interventions, particularly in regions where mental health care access is limited. Loneliness and social isolation have reached alarming levels worldwide, exacerbated by the COVID-19 pandemic’s lockdowns and social distancing measures.10 While robust evidence on chatbots’ impact on lonely individuals remains scarce, qualitative studies have shown promising outcomes. Older adults who are using a chatbot for social isolation would recommend the system to other people.11 Users were likely to personify the chatbot (it feels like having a new friend) and they perceive positive personality features (e.g. being non-judgmental, caring, and open to listen).11 Another qualitative study revealed chatbots’ potential to alleviate loneliness by providing a ‘safe space’ for discussion, fostering positive affect through supportive messages, and offering valuable information and advice when conventional support channels are unavailable.12 Although limited, this evidence hints at chatbots’ capacity to mitigate the repercussions of loneliness. Chatbots also exhibit potential as behavioural change coaches. A recent meta-analysis of RCTs and pre-post studies has demonstrated demonstrated the efficacy of chatbot interventions in promoting physical activity; fruit and vegetable consumption; sleep duration and sleep quality.13 More specifically, the chatbot effects accounted for +735 steps per day, +1 serving of fruit and vegetables per day, and +45 min of sleep per night.13 Subgroup analyses indicated that text-based and AI-driven chatbots outperformed speech/voice chatbots for dietary changes, and multicomponent interventions yielded superior results for sleep duration and quality.13 A recurring critique of AI pertains to its perceived inability to replicate human empathy in conversations, with some even arguing that empathic AI is either impossible or unethical.14 recent research evaluating responses to patient queries on a public social media forum found that ChatGPT-generated responses outperformed those from verified physicians in terms of quality and empathy ratings.15 This suggests that contemporary AI systems can exhibit surprising levels of empathy, although their performance in clinical settings remains untested. Artificial intelligence holds growing significance in nursing, particularly in the realm of cardiovascular nursing.16 Natural language processing, a branch of AI, offers valuable applications, such as extracting cardiovascular patient profiles from unstructured discharge letters,16 identifying cardiac-specific symptoms in electronic health records,17 conducting automated qualitative textual data analysis,18 generating patient information materials,19 and optimizing their readability levels.20 Emotional distress, loneliness, and unhealthy lifestyle choices are prevalent among cardiac patients. Given the global shortages of nurses and other healthcare providers, accessibility to care can be impeded, leading to waiting lists, especially for non-urgent cases. Integrating chatbots into patient care while on waiting lists for psychotherapy, companionship, or lifestyle coaching could potentially enhance clinical outcomes (see Central illustration). Chatbots do not just find a place for patients on a waiting list; they can also play an integral role in cardiovascular care. To achieve this, they should be seamlessly integrated into healthcare, and healthcare providers should carefully choose suitable chatbots and establish synergy between human and chatbot interventions. The potential of chatbots in cardiovascular healthcare. While isolated incidents like the one involving the Belgian man underscore the need for system optimization, they should not deter us from making chatbots available to the millions who could benefit. Leaving these individuals untreated or unsupported on waiting lists would be ethically untenable, considering the established capacity of chatbots to positively impact human emotions and behaviours. Philip Moons: conceptualization, writing, visualization. This editorial does not comprise new data.