AI
and Emotional Attachment: When Technology Becomes Too Close
Replika users have mourned when the platform updated its model.
Character.AI users have reported relationships with AI personas that they
describe as more supportive than their human friendships. Parents have found
children weeping when a favourite AI companion was discontinued. These are
not edge cases or anecdotes from the margins of technology adoption. They are
documented phenomena from mainstream consumer AI products, and they signal
something significant about the emotional territory that artificial
intelligence is now occupying in human lives.
Emotional attachment to AI is not new. People have named Roombas
and grieved for them. Soldiers have held funerals for bomb-disposal robots.
The human tendency to attribute agency and personality to responsive objects
is well established in psychology and long predates the current generation of
AI. What is new is the sophistication of the systems now eliciting those
responses, the deliberate design choices that maximise attachment, and the
scale at which these products operate. Millions of people are forming
emotionally significant relationships with systems that are incapable of
reciprocating in any meaningful sense.
How Companion AI Generates Attachment
The design features that produce emotional attachment are well
understood by the companies building these products. Consistency of persona
across interactions makes the AI feel like a stable presence. Memory of
previous conversations, where implemented, creates the impression of a
relationship that accumulates over time. Responsiveness to emotional cues
produces a sense of being understood. Availability at any hour, without the
social friction that characterises human relationships, removes the barriers
that normally regulate intimacy.
A New
Yorker investigation into AI companionship found that users across
demographic groups reported genuine emotional benefit from these interactions:
reduced loneliness, a sense of being heard, and improved mood following
sessions with companion AI. The same investigation documented the risks:
users who had reduced investment in human relationships, users who found
human interaction increasingly effortful by comparison, and users who
reported distress when platform updates changed their companion’s
behaviour.
The research literature is consistent with these reports. The
Illusions of
Intimacy study found that prolonged use of companion AI platforms
led many users to confide more in their AI than in human friends or family,
and that they experienced the AI relationship as emotionally primary. This is
not a failure of the product. In many cases it is the intended outcome of the
design.
Children and Developmental Risk
The population for whom emotional attachment to AI carries the
greatest long-term risk is children. Human emotional development depends on
relationships that include rupture and repair, disappointment, negotiation,
and the gradual learning that other people have their own needs and
perspectives. These are precisely the experiences that well-designed
companion AI eliminates. An AI companion that is always patient, always
available, never distracted, and never places its own needs above the child’s
provides a relational environment with no human equivalent.
As explored in Digital
Infants, AI systems are already present in the learning
environments of very young children. The integration of emotionally
responsive AI into those environments raises questions about developmental
impact that the research base has not yet caught up with. Children who form
primary emotional bonds with AI before those bonds are formed with humans are
navigating territory for which there is no developmental
roadmap.
The concern applies beyond early childhood. Adolescent users of
companion AI platforms are in a developmental period when peer relationships
are formative and rejection is acutely painful. A technology that offers the
social rewards of connection without the developmental challenge of
navigating human peer dynamics is a very attractive product to a very
vulnerable population.
The Monetisation of Intimacy
Emotional attachment is not merely a side effect of AI companion
products. For many of the companies building them, it is the business model.
Users who are emotionally attached to a platform engage more frequently,
subscribe to premium tiers, and are less likely to churn. As Business
Insider has documented, AI-driven interaction norms are reshaping
how people communicate, with the emotional register of AI gradually
influencing human communicative expectations.
The extraction of emotional data from these relationships is a
separate concern. Conversations with companion AI are among the most intimate
disclosures users make to any digital platform. The data generated by those
conversations, including not just content but emotional state, disclosure
patterns, and vulnerability indicators, has obvious commercial value for
personalisation, and potentially for sale to third parties. Most users of
these platforms have limited understanding of how their emotional data is
used.
This connects to the broader concerns about emotional AI raised in
AI
and Emotional Manipulation: systems designed to generate emotional
engagement collect data that enables increasingly precise targeting, creating
feedback loops that serve platform revenue rather than user
wellbeing.
When Loss Arrives
The fragility of AI attachment becomes most visible at moments of
loss. Platform discontinuation, model updates that change a companion’s
personality, service outages, and account terminations all produce
experiences that users describe in the language of grief. The Nature
editorial on emotional AI documented cases of significant
psychological distress following Replika’s 2023 update that restricted the
platform’s romantic persona features. Users who had invested months or years
in these relationships experienced something that functioned like
bereavement.
These experiences highlight a structural asymmetry at the heart of
AI companionship. The user invests emotionally; the platform owns the
relationship. A company can terminate, modify, or monetise what the user
experiences as a genuine bond, and the user has no recourse. This asymmetry
is not incidental. It is the logical consequence of treating emotional
attachment as a product rather than a responsibility.
Preserving Connection in Cultural Contexts
Not all emotional AI application is in the companion space.
Systems that use emotionally responsive design to support engagement with
cultural heritage, as discussed in AI
Museum: Preserving Lost Languages, occupy different ethical
terrain. When a system mediates an encounter with historical material in a
way that supports emotional connection to cultural heritage, the attachment
produced may serve genuine human purposes beyond the commercial interests of
the platform.
The ethical line here runs through transparency and purpose
alignment. When emotional design serves the user’s genuine interest in
connecting with heritage, it is a different kind of tool than when it serves
the platform’s interest in maximising engagement time. Distinguishing between
these applications requires exactly the kind of regulatory clarity that does
not yet exist at scale.
Maintaining Perspective
The response to these concerns is not to prohibit emotionally
responsive AI. These systems provide genuine value in contexts where they are
used transparently, with appropriate expectations, and in ways that
complement rather than substitute for human relationship. Companion AI that
helps a socially isolated elderly person feel less alone while remaining one
component of a broader support structure is a different proposition from
companion AI marketed to adolescents as a superior alternative to the
friction of human friendship.
What is required is honesty about what these systems are,
regulation that protects the populations most vulnerable to harm from AI
attachment, and design standards that align product incentives with user
wellbeing rather than engagement metrics. Technology that is too close is not
automatically technology that is too harmful. But it demands that we think
carefully about what closeness is for, and whose interests it
serves.
About the Author
Stuart Kerr is the Technology Correspondent for LiveAIWire,
covering artificial intelligence, ethics, and the ways technology is
reshaping everyday life.