A study in China involving full-time college students with severe loneliness found that continuous interactions with Replika, a chatbot designed to recognize users’ emotional states, alleviated loneliness after 1, 3, and 5 months of intervention. The students who interacted with the chatbot showed larger decreases in loneliness compared to students relying on standard psychological support activities. They also showed decreased social anxiety, along with improved social self-efficacy and psychological resilience. The paper was published in Disability and Rehabilitation: Assistive Technology.
Loneliness is a distressing feeling that arises when a person’s social relationships are perceived as insufficient or lacking in quality or emotional closeness. It is not the same as being alone, since one can feel lonely even in the presence of others.
Among young people, loneliness is surprisingly common despite their frequent social interactions, especially through digital media. Studies across countries show that a significant proportion of adolescents and young adults—often between 20% and 40%—report feeling lonely at least some of the time.
Factors contributing to feelings of loneliness include social comparison on social media, pressure to fit in, and transitional life stages such as leaving home or starting university. Poor-quality friendships and bullying can also heighten loneliness in youth. Loneliness in this age group is linked to higher risks of depression, anxiety, and low self-esteem. It can also affect sleep, immunity, and academic performance. While temporary loneliness is normal, chronic loneliness can become a serious mental health concern.
Study authors Xinyuan Lu and Wan Guo wanted to explore whether interaction with an emotionally intelligent chatbot could be used to alleviate loneliness in college students. Chatbots, also known as conversational agents, are AI systems capable of interacting with users through natural language via text or speech.
However, basic interaction is often not enough to create a fulfilling conversation, as genuine connection requires an understanding of emotions. If a chatbot cannot take emotions into account, a mismatch between its responses and the user’s emotional state can lead to disappointment. The study authors note that a new generation of chatbots is emerging. These advanced chatbots can interpret not only text but also facial expressions, gestures, and physiological signals, giving them a more holistic ability to interpret a user’s emotional state.
They conducted a study in which they focused on the effects of interactions with the Replika social chatbot. The study authors note that Replika is “a highly popular and well-rated social chatbot available on both the Apple and Google Play stores.”
“Since its release in 2018, it has attracted a large user base. The app is marketed as a ‘friend who always listens’ or a ‘personalized AI version,’ and has garnered significant media attention. Unlike traditional companion chatbots, which typically rely on user prompts to trigger pre-written fixed responses, Replika employs the Generative Pre-trained Transformer 3 (GPT-3) neural network model for response generation. This model is trained on user interaction data and can select the most relevant responses from an extensive dataset, including over a million responses, and rank them based on user feedback for each reply. This mechanism provides Replika with greater flexibility, enabling it to understand a wider range of vocabulary and generate more natural, emotionally intelligent conversational responses,” the study authors explain.
Study participants were 120 college students assessed as having severe loneliness. The researchers divided them randomly into two groups. One group of students (the experimental group) regularly interacted with the Replika chatbot to obtain psychological support and academic assistance according to a detailed interaction plan. This plan included regularly sharing their personal emotional states with the chatbot, engaging in deep conversations with it to share feelings and experiences, completing psychological assessments provided by the chatbot, and various other activities.
Participants in the control group attended psychological health education events, had access to psychological counseling rooms and online counseling services, participated in psychological health promotion activities, and engaged in other similar activities. Their teachers were also encouraged to strengthen communication with study participants and to pay attention to the psychological needs of these participants.
Results showed that the two groups had similar loneliness scores at the start of the study. However, at the 1, 3, and 5-month follow-ups, the group that interacted with the Replika chatbot showed significantly lower loneliness levels. Similarly, this group showed lower levels of social anxiety at these measurement points. The experimental group also had increased social self-efficacy and psychological resilience after the intervention compared to the control group, even though both groups started with similar levels of these traits.
“High emotional intelligence AI chatbots can significantly improve and alleviate feelings of loneliness and enhance social skills, opening up new avenues for technology assisted psychological intervention,” the study authors concluded.
The study sheds light on the potential of AI chatbots that can interpret users’ emotional states to alleviate loneliness. However, it should be noted that all the study participants were college students and that loneliness was self-reported. If participants knew which group was experimental (which is likely), it is possible that the results were affected by the Hawthorne effect. In other words, the recorded changes might stem from participants’ awareness of being studied and their assumptions about the researchers’ expectations, rather than the intervention itself.
The paper, “Utilizing artificial intelligence to enhance social connections – the alleviating effect of emotionally intelligent chatbots on loneliness,” was authored by Xinyuan Lu and Wan Guo.
