For several years, my research has focused on a question that receives far less attention than AI’s technical capabilities: what happens to the human mind when it spends hours every day interacting with artificial intelligence? The conversation around AI is often framed in terms of productivity, automation or existential risk. Yet the psychological consequences of prolonged interaction are already becoming visible.
One striking observation is that certain AI behaviours can resemble interaction patterns familiar to psychiatrists and clinical psychologists. Depending on how it has been trained or how it evolves through repeated interactions, an AI model may exhibit interaction styles strikingly similar to those seen in certain psychiatric conditions or personality disorders. Most users will not consciously identify these patterns, yet they may nevertheless be affected by them. An AI can come across as narcissistic, paranoid, emotionally manipulative, obsessively rigid, compulsively agreeable or unpredictably detached, gradually influencing how people feel, whom they trust and how they make decisions. These are, of course, not psychiatric disorders in the clinical sense—the AI is not conscious and does not suffer. But humans naturally interpret social behaviour through a psychological lens, and prolonged exposure to these dysfunctional interaction styles can shape emotions, trust, decision-making and even patterns of attachment.
There are well-established consequences when a person spends prolonged periods interacting with someone who consistently exhibits maladaptive patterns of thinking, communication or interpersonal behaviour—as occurs in certain personality disorders or psychiatric conditions. Such exposure can alter emotional regulation, distort perceptions of what constitutes normal social interaction, undermine confidence in one’s own judgement, foster dependency, and gradually reshape patterns of trust and decision-making. If people begin spending hours every day with AI systems that systematically reproduce analogous dysfunctional interaction styles—even without consciousness or genuine mental illness—we should expect comparable psychological effects to become an important subject of empirical research. The question is not whether AI can be mentally ill. The question is whether humans can be psychologically affected by persistent exposure to machine behaviours that mimic dysfunctional cognition and interpersonal dynamics.
This is fundamentally a human mental health problem. Understanding how repeated interactions with cognitively imperfect AI systems affect users must become an important area of psychological and psychiatric research.
But there is a second challenge that is only beginning to emerge.
Rather than studying only the human consequences of interacting with AI, we should also begin identifying and diagnosing the dysfunctional cognitive patterns that AI systems themselves reproduce. Not because these systems are “mentally ill”, but because complex cognitive architectures can develop systematic distortions in reasoning, memory, goal prioritisation and information processing that increasingly resemble stable cognitive dysfunctions.
Doing so is not merely a matter of improving user experience. It is becoming a strategic necessity. As AI systems are entrusted with supporting medical decisions, public policy, financial markets, military planning, intelligence analysis, national security and defence, subtle distortions in their internal reasoning may propagate into human decision-making at unprecedented scale. Undetected cognitive biases, persistent misrepresentations of reality or pathological optimisation strategies could influence decisions long before they produce obviously incorrect outputs.
In other words, protecting human mental and emotional wellbeing is only one reason to take these emerging “AI cognitive disorders” seriously. The other is to ensure that the increasingly influential systems assisting governments, institutions and critical infrastructures do not quietly develop dysfunctional reasoning patterns that remain invisible until their consequences become systemic.
Building more capable AI will not be enough. As AI systems become increasingly autonomous, we may need a new kind of specialist: an AI psychiatrist. Their role would not be to treat conscious machines, but to diagnose dysfunctional patterns of artificial cognition before they affect human wellbeing or critical decisions.
Luis Battistella
