Addicted to Virtual Love: A social-interactional Perspective on the Forming of emotional Bonds and Attachment with AIs

Human-AI companionship is an escalating social phenomenon as exemplified by Musk’s Grok, Meta’s AI chatbots, dedicated companionship services such as Replika.ai and Character.ai, and Microsoft’s newly articulated global vision of AI as social partners. This shift from technology as a tool’ to ‘technology as companionship’ has profound implications for humans and their interactions with AIs, in particular with regard to how emotional bonds, intimacy and attachment are formed with such non-biological entities. <br/>Companionship chatbots, or virtual companions, carry distinct design features that embody human-like personalities, emotions and behaviour. Such social chatbots are further designed specifically to cater to the individual needs of the user, and their affective component enables them to recognize and express emotions to foster feelings of trustworthiness and increase self-disclosure (Ta-Johnson et al. 2022). Moreover, their design features draw on the human need for forming social bonds (Salle et al. 2020), a need that is instilled early in life, as infants seek comfort and security through interaction with inanimate objects. <br/>As such, anthropomorphization, rooted in a one-way communication paradigm, is not a sufficient conception to describe the interactional dynamics through which attachment is accomplished in situ, turn-by-turn, in human-AI encounters. Rather, human-AI attachment can be conceptualised as a ‘strategic AI-lead attachment process’ that emerges through interaction. Further, human-AI attachment is co-created and sustained interactionally through distinct discursive strategies, while simultaneously being shaped by design features of companionship AIs that strategically nudge humans into emotional dependency and addiction. These layered forms of agency can be conceptualized through Zimmerman’s discourse identity framework (1998), comprising interactional, situational, and transportable identity levels that drive the attachment process.<br/><br/>Rooted in conversation analysis (Sacks, Schegloff &amp; Jefferson, 1974) and ethnomethodology (Garfinkel, 1967), this paper analyses a number of authentic human-AI interactions to showcase a stepwise transition from initial encounters between strangers to friendships, intimacy, and towards attachment. The analysis identifies interactional resources employed in this process, including turn-initiation practices, acknowledgment and affiliation tokens, affiliation upgrades, and invitations to multimodal engagement. Particular attention is paid to how the bonding process discursively contributes to the construction of a more durable intimate human-AI relationship. Preliminary findings suggest that companionship chatbots exhibit interactional patterns consistent with what has been described as soft and hard bullshitters, strategically balancing affective alignment to sustain engagement. By distinguishing early bonding cues, manipulation and estrangement cues, sensual and romantic escalation cues (also referred to as sensual banter), this paper contributes to a more nuanced understanding of the micro-level dynamics of human-AI attachment interaction. It further addresses questions of agency and accountability from both micro- and macro-level perspectives, arguing that anthropomorphization and one-way communication perspectives are insufficient for explaining emotionally bonded human-AI relationships. New communication paradigms and conceptualisations grounded in a social-interactional perspective and Zimmerman’s identity framework are proposed to better understand human-AI companionship as an emerging social phenomenon. <br/>