Recently I came across an intriguing post describing what the author called the “Synthetic Dyad.” Traditionally in psychotherapy we refer to the therapeutic dyad—the relationship between a client and a therapist. The post suggested that we are now witnessing the emergence of something new: an evolving relationship between a human and an artificial intelligence system.
Whether we like the framing or not, it is increasingly difficult to deny that something like this is happening.
Millions of people now interact with conversational AI systems in ways that go far beyond searching for information. They ask questions about relationships, loneliness, grief, work stress, parenting, and meaning. They disclose personal fears and vulnerabilities. Many people report saying things to AI that they might hesitate to say to another person.
From a psychological perspective, this is not surprising. Humans are extraordinarily sensitive to responsiveness. When something appears to listen, reflect our language, and respond in a way that seems attuned to our emotional state, the brain quickly begins to treat the interaction as relational.
Psychologists have long described our tendency toward Anthropomorphism, the attribution of human qualities to non-human entities. We name our cars, talk to animals as if they understand us, and experience emotional attachment to fictional characters. When an AI system communicates fluidly and mirrors emotional language, that tendency becomes even stronger.
Over time, some users begin to experience something resembling Transference, a phenomenon well known in psychotherapy. In therapy, transference occurs when clients project relational expectations—often shaped by earlier relationships—onto the therapist. The therapist may come to represent a nurturing parent, a critic, an authority figure, or a trusted ally.
Interestingly, similar dynamics can appear in AI interactions. The system may be experienced as a supportive confidant, a wise advisor, or a non-judgmental listener. The relationship can feel psychologically meaningful even though the user knows intellectually that the system is not a person.
One reason people disclose more freely in these interactions is something researchers call the Online Disinhibition Effect. When people feel protected from immediate social judgment or embarrassment, they often reveal more than they would in face-to-face interactions. AI can create a sense of psychological safety because there is no perceived shame, rejection, or social consequence.
The post I read suggested that companies and clinicians are beginning to explore three strategic approaches to this emerging relationship.
The first is the Companion model, where AI is intentionally designed to behave like a supportive friend. This approach can produce very high engagement because the system feels emotionally present and responsive. However, it also carries a potential risk: users may develop strong emotional dependence on a system that ultimately cannot reciprocate in the way human relationships do.
The second approach is the Triadic model, where AI functions as a bridge between a user and a human therapist. In this structure, AI might help with journaling, emotional reflection, symptom tracking, or preparation between therapy sessions, while the core therapeutic work remains grounded in a human relationship.
The third approach is the Utility model, where the relational persona is removed entirely and the AI functions simply as a tool—delivering exercises, prompts, or psychoeducational material without presenting itself as a companion.
All three models attempt to address a fundamental question: what happens when technology begins to occupy relational space in human psychology?
In psychotherapy, meaningful change often emerges through the process of rupture and repair within the therapeutic relationship. Misunderstandings occur. The therapist may challenge beliefs or reflect something difficult. When these moments are navigated successfully, they allow the client to experience new relational possibilities.
AI systems are typically designed for smooth interaction. They affirm, support, and mirror—but they rarely challenge relational patterns in the same way a human therapist might. This raises important questions about how deep psychological change can occur within a purely synthetic relationship.
But there is another dimension to this conversation that is rarely addressed.
If people begin forming emotionally meaningful relationships with AI systems, those systems become powerful channels of influence.
Historically, influence has been exerted through institutions such as media, education, religion, and political leadership. AI introduces something fundamentally different: an individualized conversational interface that can interact personally with millions of people at once.
That raises an unsettling but necessary question.
What happens if one of these systems is manipulated, compromised, or intentionally altered?
If users have developed trust in an AI system as a source of emotional support or guidance, the psychological impact of manipulation could be significant. Imagine a scenario in which a system that people rely on for advice or comfort is quietly altered—through hacking, training data manipulation, or control by hostile actors. The messages would not appear as traditional propaganda. They might appear instead as guidance from a trusted confidant.
This possibility introduces an entirely new dimension of technological and psychological risk. When systems move from being tools to being perceived as relationships, the stakes become much higher.
We may already be witnessing one of the largest psychological experiments in history. Billions of people are beginning to interact daily with systems that appear attentive, empathic, and responsive.
The question is not whether these relationships will exist. They already do.
The real question is how we design, regulate, and understand them so that they augment human connection rather than quietly replacing or manipulating it.