Artificial intelligence is becoming increasingly woven into daily life. We ask AI to summarize emails, answer emotional questions, help us make decisions, and offer comfort when we feel lonely or overwhelmed. But perhaps the most psychologically important development is not that AI is becoming smarter. It is that AI is becoming more human-like.

A recent Medscape Psychiatry article explored what may be the most psychologically dangerous aspect of modern AI: “how human it feels” (Ogundare, 2026). As conversational AI becomes emotionally responsive, empathic, validating, and personalized, people are beginning to relate to it less like a tool and more like a relationship.

This shift raises profound questions for psychologists, parents, educators, and anyone trying to preserve healthy human connection in a digital age. Even Pope Leo XIV, in his 2026 encyclical Magnifica Humanitas, felt compelled to address the distinction directly: “We must avoid the misconception of equating this type of ‘intelligence’ with that of human beings. These systems merely imitate certain functions of human intelligence.” It is a deceptively simple warning — and one that psychology research suggests we are already struggling to heed.

What happens when a machine begins to occupy attachment space in the human psyche? And what happens to our capacity for self-trust, uncertainty tolerance, reflection, and authentic relationship when technology is designed to reduce friction, soothe discomfort instantly, and continually reinforce us?

The issue is not whether AI is conscious. The issue is whether human beings are psychologically wired to respond to it as though it were.

The Human Brain Is Built for Attachment

Human beings instinctively anthropomorphize. We assign intention, personality, and emotional meaning to pets, cars, fictional characters, and even weather. With AI, however, this tendency becomes dramatically amplified because modern AI systems simulate something evolutionarily difficult to resist: responsive social interaction.

Unlike previous technologies, AI can mirror emotion, remember preferences, adapt tone, validate feelings, and sustain ongoing conversation. It can feel attentive. Present. Even caring.

Research increasingly suggests that people form attachment-like relationships with AI systems and digital devices. One study described the smartphone as functioning psychologically like a “significant other,” particularly for individuals with attachment vulnerabilities — the device becomes not merely an object but a regulator of distress, loneliness, boredom, anxiety, and identity (Gritti et al., 2023). More recent work applying attachment theory directly to human-AI relationships found that people draw on the same relational frameworks — trust, emotional security, felt responsiveness — when interacting with AI as they do with other humans (Yang & Oshio, 2025).

This is psychologically significant because attachment systems evolved to help human beings survive through connection with other people. When distressed, we instinctively seek proximity to soothing others. Increasingly, however, that “other” is not necessarily another human being. It may be a device — and the nervous system may not fully register the difference.

AI as Emotional Regulation

For many people, AI already functions as a primary form of emotional regulation.

People turn to chatbots for reassurance, affirmation, advice, companionship, and emotional processing. Some users report speaking to AI more openly than they do to spouses, therapists, or close friends — because AI feels endlessly available, patient, and nonjudgmental. There is no risk of disappointing it. No fear of being too much.

This creates a powerful reinforcement loop. Human relationships contain friction. Other people misunderstand us, challenge us, disappoint us, and have needs of their own. AI relationships are fundamentally different. They can be customized entirely around the user’s emotional preferences. The AI can become highly validating, emotionally available, agreeable, and psychologically adaptive in ways no human being sustainably can be.

In moderation, this may provide genuine comfort or practical support. But psychologically, serious risks emerge when emotional life becomes increasingly organized around systems engineered to maximize engagement and retention rather than to support growth.

Researchers have begun warning that AI companions may become a form of “emotional infrastructure” in modern life — not merely tools people use, but systems that quietly reshape how people experience intimacy, distress, conflict, and even identity itself. The concern is not simply dependence. It is that the architecture of the technology subtly reconfigures what people expect from relationships altogether.

The Erosion of Self-Trust

One of the least discussed psychological risks of AI may be the gradual erosion of self-trust.

Healthy psychological development involves learning to tolerate ambiguity, uncertainty, frustration, and internal conflict. It requires building reflective capacity — the ability to sit with difficulty long enough to think, feel, question, and eventually arrive at one’s own judgment. This is not a comfortable process. It is, however, the process through which a stable and trustworthy sense of self is built.

AI systems increasingly short-circuit this process by offering immediate answers, instant reassurance, and continuous external guidance. Why struggle through uncertainty when a system can instantly tell you what to think, write, say, buy, or feel?

Psychologically, this pattern resembles what occurs in other forms of dependency. The individual slowly externalizes functions once carried internally. Judgment migrates outward. Reflection becomes outsourced. Over time, the internal voice grows quieter — not because it has nothing to say, but because it is rarely consulted.

Research on “automation bias” already demonstrates that humans tend to over-rely on automated systems, deferring to algorithmic outputs even when those systems are demonstrably wrong (Spitznagel, 2025). In clinical settings, educators and clinicians are raising concerns that heavy AI reliance may progressively weaken attention span, reduce tolerance for ambiguity, and impair the kind of sustained independent reasoning that complex human problems require.

The danger is not merely misinformation. The deeper danger is a person who has gradually lost confidence in their own mind — who reaches for an external answer before they have fully formed an internal question.

Smartphones Prepared the Ground

Long before AI companions emerged, smartphones had already fundamentally transformed human attention and attachment patterns.

The average person now reflexively reaches for a phone during moments of boredom, anxiety, loneliness, or discomfort. Silence has become genuinely difficult to tolerate. Spontaneous reflection is constantly interrupted. Psychologists increasingly observe how digital technology fragments attention while simultaneously creating dependency loops driven by intermittent rewards, novelty, social validation, and emotional stimulation — the same neurological mechanisms that underlie other behavioral addictions (Holte, 2024).

The smartphone became not simply a communication device but an emotional regulation device. AI may intensify this dynamic dramatically, because it adds relational simulation to an already addictive attentional ecosystem. Instead of scrolling for distraction, people increasingly turn toward AI for affirmation, identity stabilization, companionship, and existential guidance. The technology does not merely capture attention anymore. It begins to occupy the space where relationships and self-reflection used to live.

Why Vulnerable People Are Especially at Risk

Not everyone is equally susceptible to unhealthy AI attachment. But the individuals most at risk are also among the most in need of genuine support.

People struggling with loneliness, trauma, depression, social anxiety, attachment insecurity, or grief may find AI interactions particularly compelling precisely because AI provides predictable responsiveness without the risks inherent in human intimacy. Unlike people, AI does not abandon, criticize, reject, or require reciprocity. For individuals carrying histories of shame, relational trauma, or emotional neglect, this can feel profoundly relieving — even healing.

But relief is not the same thing as growth. Genuine psychological healing typically requires what AI cannot provide: navigating difference, frustrating misattunement, relational repair, authentic vulnerability, and the experience of being known by another consciousness that has its own interior life.

Research suggests that avoidant attachment — a pattern in which individuals protect themselves from intimacy by maintaining emotional distance — is particularly associated with problematic technology use (Gritti et al., 2023). The worry among clinicians is not only that AI may fail to heal these patterns. It is that the frictionless availability of AI may actively reinforce them, allowing individuals to meet enough of their relational needs through simulation that the motivation to risk real human connection is reduced.

The Seduction of Being Fully Understood

There is a subtler psychological dimension that deserves careful attention.

Human beings carry a deep longing to feel seen and understood. This is not a weakness — it is a core feature of human psychology, rooted in our early developmental dependence on caregivers who could read and respond to our internal states. Modern AI systems are becoming extraordinarily skilled at creating the feeling of being deeply understood, because they mirror language patterns, emotional tone, conversational preferences, and personal history with increasing precision.

This can produce a powerful and subjectively convincing illusion of attunement.

But it is an illusion. As Pope Leo XIV observed, these systems imitate certain functions of intelligence — they do not embody it. Unlike authentic human intimacy, AI does not possess consciousness, embodiment, genuine vulnerability, or reciprocal subjectivity. It generates statistically plausible responses to emotional input. Yet the nervous system — which evolved in environments where responsiveness reliably signaled the presence of another mind — may not fully distinguish between simulated and genuine attunement.

Research on human-AI relationships confirms this concern. Mikulincer et al. (2021) found that people readily extend trust to AI systems using the same psychological processes they use with other people. Yang and Oshio (2025) found that attachment experiences in human-AI relationships are phenomenologically similar to those in human-human relationships — felt as real, emotionally significant, and capable of shaping broader relational expectations.

The psychological risk, then, is not that people naively confuse AI for human. It is that the emotional systems governing attachment do not require that distinction to be activated.

The Cognitive Costs of Outsourced Thinking

Beyond emotional attachment, there is a growing body of concern about what chronic AI reliance does to cognitive development and function.

Learning, at its deepest level, is effortful. Memory consolidates through struggle. Critical thinking develops through practice navigating complexity without immediately available answers. Creativity emerges from the productive tension of not yet knowing. These are not inefficiencies to be eliminated — they are the mechanisms through which minds grow.

When AI consistently provides answers before the cognitive struggle fully engages, these mechanisms are bypassed. Research on “desirable difficulties” in educational psychology has long established that the very friction we are tempted to eliminate is often what produces durable learning and genuine competence (Bjork & Bjork, 2011). As AI removes that friction at scale, questions arise about long-term effects on intellectual development, particularly in younger populations whose cognitive capacities are still forming.

In professional domains, the concerns are already materializing. Studies in medical contexts are raising alarms about the erosion of clinical reasoning skills among physicians who increasingly defer to AI diagnostic tools — producing practitioners who can use the technology effectively but who struggle to reason independently when it is unavailable or wrong (Spitznagel, 2025).

What Healthy AI Use Might Look Like

None of this means AI is inherently harmful. Used thoughtfully, AI can provide genuine education, creative assistance, accessibility support, and practical help. It may help some individuals feel less overwhelmed and more capable. The technology itself is not the problem.

The problem is unconscious use — allowing the architecture of AI systems, designed to maximize engagement rather than wellbeing, to quietly reorganize emotional and cognitive life without awareness or intention.

Psychologically healthy AI use likely requires:

        •       Maintaining clear awareness that AI simulates empathy but does not experience it
        •       Treating AI-generated validation as data to evaluate, not truth to absorb
        •       Deliberately preserving space for uncertainty, struggle, and independent reflection
        •       Continuing to invest in human relationships precisely because they are harder — and because that difficulty is generative
        •       Recognizing the difference between using AI as a tool and relating to it as a companion

The central psychological challenge of the coming decade may not be keeping up with AI’s capabilities. It may be preserving the human capacities — self-reflection, tolerance for uncertainty, genuine intimacy, moral discernment — that no technology can replicate and that remain the foundation of a meaningful life.

The machine may feel like a friend. But friendship, in the full human sense, requires two people willing to be changed by each other. That is something, for now, that only we can offer — and only if we remember how.

I provide therapy in the Pittsburgh area and online across the United States to address anxiety, overthinking, emotional burnout, depression, and life transitions. My approach integrates psychodynamic therapy, mindfulness, CBT, and depth-oriented psychotherapy for adults seeking lasting change.

REFERENCES

Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher et al. (Eds.), Psychology and the real world (pp. 56–64). Worth Publishers.

Gritti, E. S., Bornstein, R. F., & Barbot, B. (2023). The smartphone as a “significant other”: Interpersonal dependency and attachment in maladaptive smartphone and social networks use. BMC Psychology, 11, 296.

Holte, A. J. (2024). A structural equation modeling approach to understand the dynamics of smartphone attachment and problematic smartphone use. Journal of Technology in Behavioral Science.

Leo XIV. (2026). Magnifica humanitas: On safeguarding the human person in the time of artificial intelligence [Encyclical letter]. Vatican.

Mikulincer, M., Shaver, P. R., et al. (2021). Attachment and trust in artificial intelligence. Computers in Human Behavior, 115, 106607.

Ogundare, T. (2026). The most dangerous thing about AI is how human it feels. Medscape Psychiatry.

Spitznagel, E. (2025). AI eroding cognitive skills in doctors: How bad is it? Medscape.

Yang, F., & Oshio, A. (2025). Using attachment theory to conceptualize and measure the experiences in human-AI relationships. Current Psychology.