Two studies conducted in China found that female college students were most likely to develop romantic feelings for physically attractive and highly interactive artificial intelligence agents. The perceived interactivity of the virtual agent also influenced the patterns of brain activity that students exhibited during interactions. The paper is Humanities and Social Sciences Communication.
Virtual agents are computer-based systems that can interact with people or digital environments in a partially independent manner. It can answer questions, give instructions, make recommendations, perform tasks, and simulate conversations. Some appear as simple chat windows, while others feature voices, animated characters, and specific roles within a digital game or virtual world.
These systems use artificial intelligence to interpret text, voice, or other data and select responses that match the user’s requests. Modern examples include customer service bots, virtual tutors, digital assistants like Siri, video game characters, and more. Unlike simple scripts, advanced virtual agents can adapt their behavior to different situations. However, its behavior is completely dependent on programming and training data, so it cannot truly understand or feel emotions.
As virtual agents become more sophisticated, users will increasingly develop parasocial relationships with them. These connections are considered parasocial because they are completely one-sided. Humans can have deep feelings toward virtual agents, but computer programs cannot truly respond to those feelings. In recent years, there has been an increase in romantic parasocial relationships with programs designed to simulate emotional companionship and intimacy.
Study author Siyu Jin and colleagues note that previous researchers viewed these one-sided relationships as an extension of real-life bonds. This is because the human brain often has a hard time distinguishing between real and simulated social interactions on a neural level. To investigate this further, the researchers conducted two separate experiments.
The first experiment aimed to investigate how perceived interactivity and physical attractiveness influence female students’ romantic feelings toward a virtual agent. The participants were 117 female students from a university in central China. The researchers divided the students into four groups. Each group had individual conversations with male virtual characters characterized by different combinations of high and low physical attractiveness and interactivity.
Students interacted with a virtual agent designed to act as an empathetic ex-friend in scenarios involving mutual support and romantic confessions. In less interactive groups, participants simply read the written text. In highly interactive groups, sophisticated language models powered dynamic and responsive conversations. Students then rated the agents’ physical attractiveness, quality of interaction, and romantic interest.
The second study included 42 female students who were currently in a real-life romantic relationship. The goal was to record brain activity related to romantic feelings in virtual agents. The researchers used highly engaging virtual characters with varying levels of interactivity, and also used photos of each participant’s real-life boyfriend as a real-world comparison. Special brain-imaging techniques tracked neural activity while the students recalled memories of the agent and her boyfriend.
In the first experiment, we found that students form the strongest romantic relationships when virtual agents are physically attractive and highly interactive. When a character’s visual attractiveness was low, the quality of the conversation did not change participants’ romantic feelings. However, if the characters were highly attractive, the flexible and dynamic conversations greatly enhanced the students’ romantic feelings.
Brain scans reveal that highly interactive artificial intelligence triggers neural patterns that are very similar to those produced by real-world romantic love. Interacting with a highly responsive virtual agent increased activity in brain regions associated with higher-level thinking, emotional regulation, and social understanding. At the same time, this high interactivity suppressed activity in the supramarginal gyrus, a brain region that helps humans distinguish between their own emotions and those of others.
Advanced language models often reflect the user’s own input, potentially blurring the line between human and machine. Researchers believe the participants may have temporarily suspended self-awareness and projected their own emotions onto the program. This neural blurring effect may explain why highly interactive digital characters evoke such strong, human-like romantic attachments.
“In the age of artificial intelligence, this study deepens our understanding of new forms of romantic relationships,” the study authors concluded. They added that the findings provide a foundation for designing safer digital interactions, developing ethical guidelines, and evaluating the mental health impact of virtual companions.
This research contributes to scientific understanding of how humans are connected to modern technology. However, it focused only on short-term interactions driven by immediate surface-level characteristics such as physical attractiveness. Future research focusing on long-term digital relationships may yield different findings about how these emotional connections evolve over time.
The paper, “Falling in love with an AI virtual agent: The role of physical attractiveness and perceived interactivity in parasocial romantic relationships,” was authored by Siyu Jin, Fang Xu, Zihui Yuan, Gengfeng Niu, and Zongkui Zhou.
