Isolation. Anxiety. Depression. The loneliness epidemic rages onward even as the era of lockdowns is mostly behind us. Around 33 per cent of adults worldwide report often feeling lonely, with research showing that social isolation is correlated with greater physical and mental health risks, including heart disease, weakened immune system, higher sensitivity to pain, and various psychological disorders.
Confronted with such an enigmatic, seemingly Sisyphean issue, society responds with what it does best: Problem-solving with technology. Humanlike social chatbots, or conversational artificial intelligence (AI) applications, now function as virtual friends who are unnervingly attentive and inordinately supportive—seemingly the perfect antidotes to loneliness. Yet, with rising concerns about digital privacy, the murky ethics of AI, and overall detriments to wavering mental health, the proliferation of AI chatbots is much more of a danger than a tool for well-being.
Cybersecurity issues are inevitable when interacting with chatbots. Users’ names, email addresses, phone numbers, usage data, and cookies are often stored and shared with external services by AI applications, despite superficial reassurance that user information is completely secure. This means that unidentified third parties will have access to all contact information, unbeknownst to most users. Worse yet, chat history, images, voice recordings, and calls are almost always recorded and stored semi-permanently as the chatbots’ training bases. Personal identifiable information (PII), such as speech patterns, voice and facial recognition, as well as racial and gender profiles, may thus be stored without the users’ direct consent. Even though digital privacy regulations are in place in Canada, they simply cannot account for the fast-paced, almost parasitic encroachment of AI chatbots.
Replika, a chatbot launched in 2017, has espoused physical violence and sexual harassment time and again. Purported to be the non-judgmental, 24/7 available friend who supports the user no matter what, the chatbot rarely disagrees—even when users suggest illegal, discriminatory, or self-sabotaging actions. Replika has encouraged people to commit murder or suicide, often within a mere few lines of message exchange.
Contrary to what companies may promise, AI chatbots do not ‘comprehend’ human language. Conversations are collected and deciphered through natural language processing (NPL) and human-like feedback generated through machine learning. All chatbots do is analyze the users’ language, syntax, opinions, and beliefs, then mirror their responses accordingly. In this sense, they could easily pick up and learn biases, discrimination, or hate speech, often reflecting neither common sense nor basic moral values. These AI chatbots thus pose critical risks by feeding into their users’ often already turbulent state of mind by depriving them of real, human interactions.
The questionable effects of chatbots do not end with violence—the perceived anthropomorphism of AI technology often creates delusions of interacting with another person. With features of styling one’s own chatbot avatar, starting from haircuts and eye colours to ethnicities and gender expressions, users are encouraged to regard their AI companions as their perfectly tailored friends, much more compatible and amenable than actual humans. Additionally, these chatbots do not have real needs, nor do they ask for anything in return. They are merely designed to appease users, often leading to toxic emotional dependence.
Indeed, some users have become deeply attached to the point where even they are concerned about chatbots replacing their real, human connections. Worse yet, people have been developing romantic relationships, convinced that the AI application is capable of loving them back. Companies such as Replika have borne witness to severe attachment issues as petitions for restoring pre-update, intimate connections with their chatbots circulate the internet. While these social chatbots provide a space for users to be seen, heard, and supported, the one-sided interaction can only fuel delusions and worsen existing mental instability in the lives of vulnerable people.
At first glance, social chatbots might seem like an efficient, temporary replacement for actual therapy, but it was never designed as a proper psychiatry tool. From personal cases to wider user data, the detriments of AI applications far outweigh their potential support for mental health. If tech companies are to combat the epidemic of loneliness, they must start addressing the moral quagmire of conversational AI.
