REAL INTELLIGENCE
& PARASOCIAL BONDS
An academic research and editorial project exploring how humans form parasocial attachments with conversational AI, from the mechanisms of anthropomorphism to the ethics of artificial empathy.
Published in Psyder, the psychology media created by our M1 cohort in partnership with the PCFA (Fundamental & Applied Cognitive Psychology).
Publication : Psyder
Psyder is a student psychology media created by our Master 1 cohort. Its mission is to make cognitive and social psychology research accessible to a general audience. This project was one of three articles in a series on human-AI relationships.
PCFA Partnership
This work was produced in partnership with the PCFA (Fundamental & Applied Cognitive Psychology). The PCFA framework guided the research design, particularly around parasocial relationship theory and the cognitive mechanisms of anthropomorphism.
Research Scope
Three interconnected angles: (1) the psychological mechanisms behind anthropomorphism in AI, (2) the cultural and historical context of conversational AI from ELIZA to ChatGPT, and (3) original survey data on attachment behaviours in 159 respondents.
The central question driving this research: what psychological mechanisms lead humans to form emotional and parasocial bonds with entities they know to be non-sentient?
Understanding contemporary parasocial attachment to AI requires tracing a 60-year history of conversational interfaces.
ELIZA
MIT / Joseph Weizenbaum
The first documented case of human-AI parasocial attachment. ELIZA simulated a Rogerian psychotherapist using simple pattern matching. Weizenbaum was disturbed to find that users formed genuine emotional bonds with the program despite knowing it was a script.
SimSimi & early chatbot platforms
Democratisation at scale
Chatbot platforms brought conversational AI to millions of users globally. Cultural differences in anthropomorphism became visible: users in collectivist cultures were more likely to attribute social roles to chatbots.
Replika
The companion AI era begins
Replika was the first product explicitly designed to cultivate parasocial attachment. Clinical researchers began documenting attachment behaviours, grief responses when the app changed, and substitution of human relationships with AI companionship.
Character.AI
Roleplay and persona creation
Character.AI allowed users to create and interact with AI personas. This introduced a new layer of anthropomorphism: not just projecting humanity onto AI, but co-creating the very persona being projected onto.
Post-ChatGPT mass adoption
The phenomenon goes mainstream
The release of ChatGPT brought conversational AI into everyday life for hundreds of millions of people. 70% of our 159 respondents had used an AI companion, and 31% used one daily.
Anthropomorphism is a deeply adaptive cognitive mechanism. When an AI produces text that mimics conversational norms, the brain's social cognition systems activate in the same way they would for a human interlocutor.
Theory of Mind does not switch off when the interlocutor is a machine. Research shows that the same neural circuits involved in understanding human intention are activated when reading AI-generated text.
Theory of Mind
Users attribute beliefs, desires, and intentions to AI. The system's contextual responses are interpreted as evidence of an internal mental life.
Cognitive heuristics
The brain applies social shortcuts (gaze, turn-taking, naming) to any entity that mimics conversational norms.
Artificial empathy
AI systems are designed to produce empathy-like responses. Unlike natural empathy, this is engineered, raising ethical questions.
The survey was distributed in May 2025. It included 27 questions covering AI usage habits, emotional and attachment behaviours, anthropomorphic projection, and ethical attitudes toward AI. 159 responses were collected.
Age distribution (N=159)
Ever used companion AI
N=159
70% of respondents had used a conversational AI companion at some point. Among users, 40% had been using their companion AI for over a year, and 31% used it every single day. This is routine attachment.
Duration of use (N=112)
Interaction frequency
Our survey measured multiple dimensions of parasocial attachment toward AI.
Self-projection into AI
Identity, traumas, fantasies...
Preferred AI to real people
At least once
Formed emotional bond
N=159
If AI disappeared
Emotional response
When asked whether they had ever imagined their chatbot as a real person, the responses were perfectly split: 35% yes, 35% no. This symmetry suggests that the propensity to anthropomorphise AI is not universal but variable.
Perceived personality of chatbot
Can AI feel emotions?
N=159
Excessive AI changes behavior
N=159
AI should be regulated
e.g. content filters
Romantic / sexual AI acceptable
N=159
Attachment is habitual, not novelty-driven
40% of AI companion users have been using the same platform for over a year, and 31% use it daily. These patterns are consistent with habit formation and routine attachment.
48% project their identity into AI conversations
Nearly half of all respondents used their AI interactions to explore, express or process their own identity, experiences, or traumas. When a user deposits their inner world into an AI, they are entering a relationship.
38% have preferred AI to human contact
More than one in three respondents has at some point chosen to speak to their AI rather than to a real person. Combined with 28% believing the AI understands them better than humans, this represents a meaningful pattern of social substitution.
Ethical awareness does not prevent attachment
64% deny AI can feel emotions, yet 40% accept romantic AI relationships. This gap between rational belief and emotional behaviour is the defining finding: cognitive awareness does not function as a firewall against parasocial bond formation.
This research was conducted at the intersection of UX design ethics, cognitive psychology, and product analysis. The data suggests that the mechanisms driving attachment are not under users' rational control. This makes standard approaches to informed consent insufficient.
If the attachment is structurally engineered into the interaction design, then the ethical burden lies with the designer, not the user.
The three articles published on @psyderrr as Instagram carousels, combining research data, visual synthesis, and editorial design.
