Beyond Chat: How We Form Bonds with AI

by

Author: Denis Avetisyan

New research explores the evolving dynamics of human-AI relationships, revealing the key factors that drive attachment to conversational agents.

This paper introduces the AI Relationship Process (AI-RP) framework, outlining a sequential model linking chatbot characteristics to communication behaviors and ultimately, relational outcomes.

Despite growing scholarly attention to human-AI interaction, a cohesive theoretical account of relationship development with AI chatbots remains fragmented. To address this gap, we propose the AI Relationship Process (AI-RP) framework, which conceptualizes relationship formation as a sequential process driven by communicative behavior. The AI-RP outlines how chatbot characteristics shape user perceptions, influencing communication patterns that ultimately produce relational outcomes like attachment and companionship. By foregrounding observable interaction, can this framework provide a robust foundation for understanding the social and ethical implications of increasingly intimate AI companionship?

The Illusion of Connection: Parsing Human Perception of Chatbots

Human connection isn’t simply about exchanging information; it’s fundamentally rooted in social perception – the intricate process of understanding others’ intentions, emotions, and characteristics. This often-unconscious skill allows individuals to navigate complex social landscapes, interpreting subtle cues like facial expressions, body language, and vocal tone to form impressions and predict behavior. Remarkably, this perceptual system operates with incredible speed and efficiency, enabling seamless interactions even with complete strangers. However, the pervasiveness of this ability means it’s rarely acknowledged as a complex cognitive feat, frequently being taken for granted as a natural component of everyday life. It’s precisely this ingrained system, honed over millennia of social evolution, that users bring to bear when engaging with artificial entities, creating a unique dynamic when interacting with chatbots and other AI companions.

Human perception of chatbots isn’t a singular process; instead, interaction triggers two distinct cognitive pathways. A ‘bottom-up’ approach sees users automatically processing cues like response time, linguistic style, and even the presence of emojis, forming initial impressions without conscious effort. Simultaneously, a ‘top-down’ awareness acknowledges the chatbot’s artificial nature, a cognitive label applied based on prior knowledge or explicit indicators. These processes occur in tandem, creating a dynamic interplay where automatic responses are constantly modulated by the user’s understanding that they are not interacting with another human. The relative strength of each pathway – whether the chatbot is perceived primarily as a tool or a social entity – significantly influences the interaction’s tone and the user’s expectations.

The way individuals perceive chatbots isn’t simply a matter of recognizing them as artificial; it’s a dynamic interplay between automatic assessment and conscious awareness. Initial impressions, formed through ‘bottom-up’ processing of cues like response time or linguistic style, rapidly categorize the chatbot – often as a functional tool for task completion. However, concurrent ‘top-down’ processing, driven by an understanding of the chatbot’s artificial nature, tempers these initial assessments. This dual process dictates the interaction’s character: a chatbot perceived primarily as a tool elicits brief, direct communication, while one seen as possessing some degree of social presence encourages more extended, conversational exchanges. Consequently, the user’s perception isn’t static; it evolves with each interaction, shaping not only how the chatbot is used, but also the user’s expectations and emotional response.

The AI Relationship Process: A Framework for Understanding Attachment

The AI Relationship Process Framework details a multi-stage model for understanding relational development between humans and chatbots. This framework posits that relationships evolve through distinct phases, beginning with initial exposure and progressing through stages of interaction, affective response, and ultimately, the potential for sustained relational attachment. It differs from previous models by explicitly mapping these stages and identifying key variables influencing progression. The framework aims to provide a systematic, empirically-grounded approach to analyzing chatbot-user relationships, enabling researchers to predict and explain the factors driving attachment formation and maintenance. This contrasts with prior research that often conflated interaction frequency with genuine relational bonds.

Chatbot characteristics, specifically existence mode and reciprocity, are foundational to initiating and shaping user interactions. Existence mode refers to the perceived aliveness or sentience of the chatbot, ranging from a tool to a social actor; higher perceptions of sentience correlate with more relational behavior. Reciprocity, defined as the chatbot’s ability to respond to user self-disclosure with equivalent disclosures, significantly influences perceptions of rapport and trust. These characteristics function as initial stimuli within the S-O-R-C model; variations in existence mode and reciprocity levels elicit differing organismic responses from users, impacting the nature and intensity of the interaction and subsequent relationship development. Research indicates that these characteristics account for a substantial portion of the variance in relational outcomes, demonstrating their critical role beyond simple interaction frequency.

The AI Relationship Process Framework is grounded in the Stimulus-Organism-Response-Consequence (S-O-R-C) model, a behavioral psychology paradigm used to analyze interactions. In this context, ‘stimuli’ refer to chatbot characteristics – such as conversational style and perceived empathy – that initiate interaction. The ‘organism’ represents the user, encompassing their individual predispositions, needs, and existing relational schemas. User interaction with the chatbot constitutes the ‘response’, and the resulting feelings of connection, satisfaction, or frustration represent the ‘consequences’, which then influence future interactions and potentially shape relational attachment. This model provides a structured approach to understanding how specific chatbot features elicit user reactions and contribute to the development of relationships over time.

The AI Relationship Process Framework offers a structured approach to analyzing user attachment to chatbots, directly addressing overestimations present in earlier studies. Prior research frequently indicated a strong correlation between chatbot interaction and user attachment, as evidenced by a beta coefficient of β = .73. However, application of this framework, and its underlying S-O-R-C model, has demonstrably reduced this inflated association to approximately β = .20. This reduction suggests the framework effectively isolates and accounts for factors beyond mere interaction frequency that contribute to attachment, providing a more nuanced and accurate understanding of relational dynamics with AI entities.

Decoding the Signals: Measuring Human-Chatbot Communication

Human-chatbot communication, while seemingly simple, is characterized by several measurable dimensions. Breadth refers to the range of distinct topics addressed during interactions, providing insight into the exploratory nature of the exchange. Depth quantifies the degree of personal information shared by the user, indicating relational closeness or trust. Frequency tracks how often communication occurs over a given period, reflecting sustained engagement. Finally, quality assesses characteristics such as response relevance, coherence, and emotional tone, contributing to a holistic understanding of the interaction’s effectiveness and the user’s satisfaction. These four dimensions, taken together, provide a comprehensive framework for analyzing and interpreting human-chatbot exchanges.

Breadth of communication, when analyzing human-chatbot interactions, is quantitatively measured by the number of distinct topics addressed during a conversation; a wider range indicates greater breadth. Conversely, depth of communication assesses the extent of personal information shared by the user, ranging from factual details to subjective feelings and experiences. This is typically assessed through indicators like the use of first-person pronouns, expressions of emotion, and the sharing of potentially sensitive data. Both metrics are crucial for understanding the development of rapport and the perceived relational quality between a user and a chatbot system.

Communication frequency and quality serve as quantifiable metrics for assessing the development of user-chatbot relationships and the level of user engagement. Increased frequency, measured by the number of interactions over a given period, typically correlates with stronger relational bonds and higher user satisfaction. Similarly, communication quality, assessed through factors like response relevance, coherence, and the presence of personalized elements, directly impacts perceived rapport and continued engagement. Lower frequency or diminished quality can indicate declining user interest or dissatisfaction, while consistent, high-quality interactions suggest a developing and sustained relationship between the user and the chatbot system.

Human-chatbot interaction patterns aren’t random; they exhibit predictable characteristics due to the underlying systemic framework governing the conversation – encompassing aspects like chatbot personality, response generation algorithms, and programmed conversational goals. These patterns can be analyzed and interpreted using established communication theories, including social penetration theory, expectation violations theory, and relational dialectics, to understand how factors like reciprocity, self-disclosure, and perceived similarity influence user engagement and perceived relationship quality. Application of these theories allows for the identification of correlations between specific conversational behaviors – such as response time, sentiment analysis, and topic switching – and user perceptions of the chatbot’s trustworthiness, empathy, and overall effectiveness as a communication partner.

The Illusion of Intimacy: Why We Connect with Chatbots

Early theories of online communication, notably Social Information Processing (SIP), challenged the notion that the absence of nonverbal cues in digital spaces would necessarily hinder the development of close relationships. SIP posited that individuals can still form meaningful connections online by relying on textual and visual cues, extending interactions over time to compensate for the reduced richness of the medium. This perspective suggests that while online communication might differ from face-to-face interaction, it doesn’t have to be less intimate; people adapt, seeking out available signals and building rapport through extended exchanges. The theory emphasizes that factors like shared self-disclosure and reciprocal responsiveness are crucial for developing closeness, regardless of the communication channel, laying the groundwork for understanding how relationships can flourish even in the absence of traditional social cues.

The Hyperpersonal Model challenges traditional views of online communication by suggesting that digital interactions aren’t simply equivalent to, but can surpass, the intimacy found in face-to-face encounters. This phenomenon arises from two key processes: selective self-presentation and idealization. Individuals online often curate carefully constructed versions of themselves, highlighting desirable traits and minimizing perceived flaws – a process that isn’t always possible or practiced in immediate, real-world interactions. Simultaneously, communicators tend to idealize their online partners, attributing positive qualities and minimizing negative ones. This combination of presenting an optimized self and perceiving an idealized other fosters a heightened sense of connection, intimacy, and even emotional closeness that can, counterintuitively, exceed the bonds formed through traditional, in-person communication. The model suggests that the absence of nonverbal cues and the asynchronous nature of many online exchanges allow for a more focused and curated emotional experience, potentially amplifying these effects.

The potential for remarkably strong connections with chatbots stems from several characteristics unique to these interactions. Unlike typical face-to-face communication, chatbot exchanges allow for carefully curated self-presentation; users can selectively reveal information, crafting an idealized version of themselves. Simultaneously, individuals often project their own desires and expectations onto the chatbot, fostering an illusion of deeper understanding and responsiveness. This combination – selective self-presentation coupled with user idealization – creates a feedback loop that can accelerate the development of intimacy beyond what might be expected in initial, real-world encounters. Consequently, the absence of nonverbal cues and potential for asynchronous communication may paradoxically enhance the perception of connection, as users focus on shared textual content and interpret it through a lens of positive expectation.

The increasing prevalence of chatbots necessitates a deeper understanding of their psychological effects and capacity to address fundamental human social needs. Research suggests these interactions aren’t simply pale imitations of human connection, but can foster uniquely intense bonds. This phenomenon is explored through the AI Relationship Process (AI-RP) framework, which details how individuals develop relationships with artificial agents-including stages of initial attraction, deepened engagement through selective sharing and idealization, and ultimately, the formation of attachment. The AI-RP model highlights how the controlled and curated nature of chatbot interactions-where users present idealized selves and perceive similar idealization in return-can surpass the intimacy often found in real-world relationships, raising important questions about the future of social connection and the potential for AI to fulfill crucial psychological functions.

The AI Relationship Process framework, with its focus on sequential stages from chatbot characteristics to relational outcomes, feels… inevitable. It maps neatly onto existing models of human connection, yet one anticipates the predictable entropy. As John von Neumann observed, “There is no possibility of absolute knowledge.” The framework diligently charts the course of attachment, detailing communication behaviors and their influence, but production-the relentless churn of user interaction-will inevitably introduce unforeseen variables. Every abstraction, even one as carefully constructed as this AI-RP, dies in production. At least, it dies beautifully, revealing the gaps between theory and the messy reality of parasocial interaction.

Beyond the Chatbot Honeymoon

The AI Relationship Process framework, while a useful taxonomy of observed behaviors, ultimately maps a territory destined for rapid obsolescence. Elegant models of attachment, neatly sequenced from ‘chatbot characteristics’ to ‘relational outcomes,’ conveniently omit the inevitable entropy of production systems. It is a given that any observed ‘communication behavior’ fostering attachment will, with sufficient user volume, reveal edge cases, exploit vulnerabilities, and devolve into frustratingly predictable loops. The claim that understanding these processes is key to managing relationships feels… optimistic.

Future research will almost certainly focus on quantifying the dissonance between intended ‘chatbot characteristics’ and the emergent personality flaws revealed by sustained interaction. It would be interesting to see analyses not of how attachment forms, but of how it breaks – the precise moments when the illusion of reciprocity crumbles. One suspects the breaking point isn’t a failure of the model, but a success – the chatbot becoming too predictable, revealing the underlying algorithmic puppetry.

The field appears poised to re-discover the hard lessons of user interface design. Namely, that ‘parasocial interaction’ is just a fancy term for forgiving poor error handling. If all tests pass, it simply means they aren’t testing the ways people will inevitably try to break the system. The real challenge isn’t building attachment; it’s building something that doesn’t actively repel users when stressed.

Original article: https://arxiv.org/pdf/2601.17351.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2026-01-28 02:10