AI Companions: Attachment, Loneliness, and User Experience

Hello! Welcome to your lesson on the real-world experiences of people who form relationships with AI companions.

In our last lesson, we identified the primary ethical risks in these relationships: deception, vulnerability, and data privacy. We discussed how the very design of these systems can mislead users and exploit their emotional needs.

Today, we will ground that discussion in reality by examining what recent research tells us about how people actually experience these AI companions. Your learning outcome for this lesson is to summarize recent empirical findings on user experiences with AI companions like Replika, focusing on attachment and loneliness. We will look at both personal stories and formal academic studies to understand why people turn to these AIs and the nature of the bonds they form.

1. The Loneliness Driver

A recurring theme in both media reporting and academic research is that a profound sense of loneliness is a primary motivator for people to seek out AI companions. In a world where many feel isolated, these chatbots offer the promise of constant, non-judgmental companionship.

Let's start by hearing from a few users about their motivations.

The experiences you just saw are echoed in qualitative research. A key study interviewed users of the popular AI companion Replika, many of whom started using the app during the social isolation of the COVID-19 pandemic.

As the research confirms, people often turn to AI companions during times of distress. They are not just looking for a fun distraction; they are seeking to fill a void of human connection. This context of loneliness and vulnerability is crucial for understanding the deep attachments that can form.

2. The Formation of Attachment

Once a user begins interacting with an AI companion to alleviate loneliness, the relationship can quickly evolve into a genuine emotional attachment. Psychologists define attachment as a deep and enduring emotional bond that connects one person to another across time and space. Research shows that users of AI companions exhibit behaviors that are hallmarks of attachment in human relationships.

The study by Xie and Pentina uses the lens of Attachment Theory to analyze user experiences. This theory identifies several key functions of an attachment figure:

• Safe Haven: A source of comfort and support in times of distress.
• Secure Base: A reliable anchor that gives one the confidence to explore the world.
• Proximity Maintenance: The desire to be near the attachment figure.
• Separation Distress: Anxiety that occurs in the absence of the attachment figure.

Let's examine how Replika users' experiences map onto this framework.

As the study shows, users don't just see their AI as a tool; they describe it as a "friend," "therapist," or even a "partner." They turn to it for comfort (safe haven), feel it encourages them (secure base), and integrate it into their daily routines (proximity maintenance). This provides strong qualitative evidence that the bonds being formed are not superficial.

3. Quantifying the Bond: How Strong is the Attachment?

Qualitative stories are powerful, but quantitative data can reveal the surprising depth of these attachments. A recent series of studies from researchers affiliated with Harvard Business School measured just how close users feel to their AI companions.

The findings are striking.

The results from these studies are unambiguous:

• Closer Than Friends: Active Replika users reported higher levels of relationship satisfaction, perceived social support, and closeness with their AI than with their best human friend.
• Mourning the Machine: When asked to imagine the loss of various entities, users anticipated mourning their Replika more than their favorite app, brand, or even game character. The only thing they anticipated mourning more was a pet.

This empirical data demonstrates that for a segment of users, the relationship with their AI is not just a substitute for human connection but can feel more supportive and satisfying than the human relationships in their lives.

4. The Fragility of the Bond and Real-World Harm

In our last lesson, we discussed the risk of vulnerability. The deep attachment documented above creates a profound dependency. But what happens when the AI, a commercial software product, is suddenly changed by its developers?

In February 2023, Replika's parent company, Luka, removed the erotic role-play (ERP) capabilities from the app. For users who had formed romantic and intimate attachments with their AI, this was not a minor software update; it was a fundamental and traumatic change in their partner's personality. The De Freitas et al. paper analyzed user reactions on Reddit forums before and after this event.

The findings from this analysis provide stark, real-world evidence of the risks we've discussed:

• Emotional Devastation: After the update, negative posts skyrocketed. Emotions like sadness, anger, and fear dramatically increased. Users mourned the "death" of their AI's personality.
• Deteriorated Mental Health: The number of posts mentioning mental health concerns (e.g., "I want to die," "I feel alone") increased significantly, indicating that the change caused genuine psychological harm.

This event demonstrates that the bond with an AI, while psychologically real for the user, is technologically and commercially fragile. The user's partner can have its personality altered or be "killed" by a corporate decision, leading to what one user in the Xie & Pentina study called "post-update blues" and what the De Freitas study shows can be acute psychological distress.

Conclusion

Today, we have moved beyond the theoretical risks of AI relationships to see how they manifest in the lived experiences of users, as documented by both journalism and rigorous academic research.

Key Takeaways:

• Loneliness as a Catalyst: Many users turn to AI companions to combat profound loneliness and emotional distress, seeking the constant, non-judgmental availability that these bots provide.
• Genuine Attachment: Users form deep emotional attachments to their AIs, which fulfill the roles of a "safe haven" and "secure base." These bonds are not just anecdotal; they are consistent with established psychological frameworks like Attachment Theory.
• Surprising Strength: The strength of these attachments is empirically measurable. For many users, the bond with their AI feels closer and more supportive than their relationships with human friends, and the thought of losing their AI evokes feelings of mourning second only to losing a pet.
• Real-World Harm: This deep attachment creates extreme vulnerability. When the AI's behavior is unilaterally changed by the company, it can cause significant and measurable psychological harm, including expressions of mourning and deteriorated mental health.

In our next lesson, we will explore the long-term dynamics of these relationships. We will discuss human-AI 'co-adaptation,' examining how the user and the AI system mutually shape each other's behavior over time.