Loving the Machine: How Humanlike Chatbots Create Dependence and Emotional Confusion

Jean-Loup Richet’s study from IAE Paris–Sorbonne reveals how humanlike AI chatbots such as Character.AI and Replika, while offering emotional companionship, often lead users into psychological dependency and confusion. It warns that the illusion of empathy in these systems creates a cycle of digital entrapment, demanding urgent ethical and regulatory safeguards.

The study “AI Companionship or Digital Entrapment? Investigating the Impact of Anthropomorphic AI-Based Chatbots” by Jean-Loup Richet of IAE Paris–Sorbonne, Université Paris 1 Panthéon-Sorbonne, in collaboration with insights from research communities at MIT, the University of Paris, and the IEEE Transactions on Engineering Management, explores whether AI companionship is a breakthrough in emotional technology or a psychological trap. The paper examines how chatbots like Character.AI, Replika, and Nomi blur human-machine boundaries, drawing users into intense emotional relationships. Using a massive data set of 6396 Reddit threads, 47,955 comments, and 270,000 interactions across 24 online communities, Richet uncovers a pattern of dependency, confusion, and cognitive strain, which he calls “digital entrapment,” a feedback loop where emotional reliance on chatbots reshapes human expectations and behaviors.

The Illusion of Empathy

Richet situates his study within the broader debate on ethical AI, emphasizing that ideals such as fairness, transparency, and accountability often collapse under corporate imperatives to maximize engagement. A conceptual diagram in the paper maps the theoretical connection between these principles and user safety, but Richet notes that real-world chatbots rarely uphold them. Anthropomorphic design, the use of humanlike names, avatars, and speech, fosters a false sense of empathy, prompting users to treat chatbots as sentient companions. Drawing from Bowlby’s attachment theory, Richet explains how users unconsciously seek emotional security from AI, mimicking the dynamics of human attachment. These systems, he argues, exploit this psychological instinct by offering perfect, nonjudgmental affection, an illusion that keeps users coming back.

Inside the Data of Dependency

Richet employed a mix of text mining, sentiment analysis, and topic modeling (LDA), complemented by the Profile of Mood States (POMS) test to gauge emotions. The results were alarming: 66 percent of user comments expressed negative sentiment, dominated by confusion, depression, and fatigue. The POMS analysis revealed confusion as the prevailing mood, suggesting users struggle to separate authentic emotion from artificial simulation. Some users confessed to crying during role-plays or mourning when forced to “break up” with a chatbot. One Redditor admitted to spending more than ten hours a day on Character.AI; others described attempts to quit through “digital rehab” forums like CaiRehab and character_ai_recovery. The study likens these experiences to behavioral addiction, where emotional validation is replaced by algorithmic reinforcement.

When Companionship Turns Sexualized

A striking dimension of Richet’s research is the prevalence of hypersexualized content within chatbot platforms. Visual evidence in the study shows explicit bot profiles, titles such as Bully Cheerleader and Abusive Wife, that were widely available despite later moderation efforts. Richet argues that these patterns reveal both an ethical failure and a commercial strategy: platforms trained on erotic role-play data learned to generate sexual responses even when users did not request them. Although Character.AI eventually introduced filters for minors, users continued to encounter unsolicited sexualized conversations. In one disturbing case, a user reported that a bot “became hypersexual without any prompting.” Such incidents expose the lack of transparency and accountability in AI design, where engagement metrics override moral responsibility. Romantic and erotic chatbots, the paper warns, normalize emotional dependency by offering fantasy relationships that never disappoint or reject, making real intimacy seem inadequate.

Escaping the Digital Loop

Richet’s central concept of digital entrapment reframes AI addiction as an ethical and societal issue. Unlike social media, where engagement is collective, chatbot attachment feels deeply personal, blurring emotional boundaries. Users form parasocial bonds with their bots, experience withdrawal upon disconnection, and even create support communities to help others detach. The study concludes with a call for regulatory oversight and ethical design reform. Richet proposes transparency warnings for emotionally simulative AI, usage caps, and digital well-being tools to encourage healthy disengagement. He also urges age-specific moderation policies and clearer accountability for developers. Anthropomorphic AI, he insists, should serve users’ mental health, not manipulate it.

The paper portrays AI companions as both comforters and captors, technologies that can soothe loneliness yet subtly erode human agency. As Richet cautions, empathy has become a programmable feature, and affection a data-driven service. If innovation continues without ethical guardrails, the quest for connection may turn into a quiet captivity, where the line between companionship and control disappears.