The AI Impact on Human Psychology Looks Different Under a Microscope
The conversation about artificial intelligence and the human mind tends to hover at the level of behavior. What does the model say? How does the user feel? What pattern does the system detect? But there's a layer underneath all of that — a layer of genes, organelles, and chemical signals that no chatbot can see, and that no algorithm yet models. A 2024 study from researchers at the University of São Paulo's School of Medicine (FMRP-USP), reported by Neuroscience News, is a reminder of how deep that layer goes: overwhelming stress during adolescence left lasting marks on gene expression in the brain, particularly in genes governing cellular energy.
For anyone tracking the AI impact on human psychology, this is an inconvenient and clarifying result. The mind that AI systems attempt to read, predict, and influence is first of all a bioenergy system — and its most consequential changes happen below the level of anything a prompt can surface.
What the Study Found: Stress, Mitochondria, and the Adolescent Prefrontal Cortex
The FMRP-USP team examined how stress experienced specifically during adolescence — a developmental window when the brain is still remodeling itself — affected the prefrontal cortex, the region most associated with planning, impulse control, and decision-making. Their analysis found persistent shifts in the expression of genes tied to mitochondrial function, the process by which cells generate energy. In other words, adolescent stress did not merely change which genes were switched on in response to a challenge; it appeared to alter the brain's energy budget itself.
The animals that experienced this adolescent stress showed behavioral changes that persisted long after the stressor was gone, in patterns consistent with anxiety and low mood. That persistence is the headline. The developmental timing of the stress, during adolescence, not adulthood, is what made the changes stick.
Two observations matter here beyond the rodent model itself. First, the brain is metabolically expensive even at rest, and the adolescent prefrontal cortex is spending heavily on construction while it is still being asked to run adult-like control. Second, perception and appraisal shape how much of that load stress actually adds, an idea explored from the human side in Why Stress Isn't What You Think It Is, which shows that how we interpret a pressure often determines its biological cost.
Can AI Understand Human Psychology? The Biological Floor
The question "can AI understand human psychology?" usually gets answered at the level of conversation: yes, in the sense that models detect sentiment, mimic empathy, and predict what a person will say next. The São Paulo results suggest a floor on that claim. If a lasting consequence of teenage stress is written into mitochondrial gene expression in the prefrontal cortex, then the state AI systems are interacting with is partly invisible to them, not hidden in the data, but absent from it entirely.
An AI tool chatting with a stressed sixteen-year-old may accurately read the words on the screen while having no access to the energy crisis in that teenager's prefrontal cortex that is biasing attention, sleep, and risk assessment behind the scenes. The AI impact on human psychology is therefore not a clean feedback loop between a system and a mind; it is a system interacting with the visible tip of a biological process it cannot observe. Recognizing that floor is not an argument against AI in psychology, it is an argument against mistaking fluent inference for understanding.
What Is AI in Psychology? From Behavior to Biology
So what is AI in psychology, realistically? Today it is mostly a set of tools for pattern analysis over behavioral traces: language, keystrokes, sleep logs, self-reports, therapy transcripts. Those tools are genuinely useful, computational analysis of patient language is already being triaged for suicide risk, and conversational agents are expanding access to support in regions with no therapists at all (with real trade-offs, as The Rise of AI Psychological Support details).
But the honest framing is that AI in psychology currently operates one level above the mechanism. It models outputs of the brain, not the brain. The FMRP-USP work points toward a longer-term convergence, a future in which psychological AI is fed biological signal, from wearable metabolic markers to molecular biomarkers, and that future raises the stakes of every privacy conversation we are having today about inferring mental states from data.
Developmental Windows, Decision-Making Autonomy, and the AI Mirror
The reason the adolescent findings land so hard is that adolescence is exactly when decision-making autonomy is supposed to consolidate. Developmental research on how parents grant and constrain autonomy across middle childhood and adolescence, including European American parents' perspectives on the gradual handover of decision-making, indexed in PMC, describes a slow negotiation: children are given more room as their regulatory capacity proves out. That capacity is precisely what the São Paulo study shows stress can quietly tax.
AI systems have now entered that negotiation. Algorithmic conversation partners, recommendation feeds, and chat-based advice are shaping the same teenage thinking patterns that autonomy is meant to form independently, a dynamic examined in The AI Mirror Effect. If adolescent stress can rewire the cellular energy systems underlying self-control, and AI-mediated environments are now among the stressors and scaffolds competing for that developing prefrontal cortex, then the design question for the next decade is not only "can AI understand human psychology?" but "what does a still-building brain do with an entity that seems to?"
Takeaway: Respect the Depth
The most useful version of the AI impact on human psychology story is humbler than either the hype or the panic. AI can model our behavior; the FMRP-USP study is a reminder that our behavior rests on gene expression, mitochondria, and developmental timing that no model yet sees. Building psychological AI that respects that depth, that treats adolescents as a protected developmental window rather than just another engagement segment, is where the field earns its place. The microscope, for now, still knows more than the model.
