It's easy to think of artificial intelligence as a productivity tool: something that writes emails, summarizes reports and helps us find information faster. But in 2025 the technology has started to slip into a far more intimate space — the mental. As more people use AI in everyday life, it's no longer just what we do that changes, but also how we feel, interpret and manage life.
A recent Norwegian survey reported by Shifter points to a marked increase in use: 68 percent say they have used AI in the past year, a clear rise from 55 percent just a few months earlier. Many use it often — "one in four ... does it daily," according to the article. This means that artificial intelligence is already becoming a constant background voice in the daily lives of a significant share of the population.
When a technology becomes "ubiquitous," it also creates psychological ripple effects. We'll see AI as a low-threshold source of support, a kind of digital aide for stress, anxiety and loneliness. At the same time we will get new types of vulnerability: bad advice, dependency, and pressure to optimize ourselves. The question is not whether AI will affect mental health, but which direction that effect will take — and who pays the price.
This article looks ahead: based on today's adoption rate and the ethical tensions already visible, what can we expect over the next five years? And what needs to be in place for AI to become a psychological safety net — not a new stressor?
The Shifter piece paints a two-part picture: two out of three find AI useful, and four out of ten believe they "to a large extent can trust the information" they receive. That's an interesting starting point for mental health because it tells us two things:
Over the next five years we will likely see AI move from being an "app" to becoming an infrastructure for digital health: integrated into phones, browsers, wearables and healthcare services. This won't necessarily come through a single big revolution, but via small, frictionless upgrades: a new feature here, a better voice assistant there, and increasingly personalized experiences based on our data.
The most likely trend is that AI will increasingly be used for emotional regulation: everything from "venting" frustration to getting structure for rumination. This is already happening today, but will accelerate when:
The consequence can be positive: more people gain a tool that helps them put words to feelings, create action plans and experience support in the moment. But the psychological cost may be that we move intimate conversations away from people — and thereby lose parts of the social "muscle" that sustains mental health.
It's tempting to be skeptical of "AI therapy." Skepticism is healthy, but we should also be honest about why the technology feels attractive: it is accessible, cheap, and nonjudgmental. For many this can be a first step toward help.
In traditional mental healthcare there are often wait times, financial barriers or shame-related thresholds. AI can help lower those thresholds by providing:
If implemented responsibly, AI can become a "first filter" that picks up signs of overload before they become serious. In a five-year perspective it's realistic that some workplaces, insurance plans and digital health services will offer AI-based support programs as standard.
When "talking to an AI" becomes as normal as searching on Google, it also becomes more socially acceptable to work on mental health in everyday life. Normalization of mental tools can have a large effect, especially for groups that underuse healthcare today.
But there's also a hidden risk here: if AI becomes the default solution, society may gradually accept that people receive "cheap support" instead of real access to psychologists and treatment capacity. In other words: destigmatization can paradoxically combine with a political temptation to deprioritize human treatment.
When a technology becomes emotionally relevant, the consequences of errors grow. For mental health, "almost right" is often more dangerous than "completely wrong," because it can lock us into beliefs that feel plausible.
Over the next five years there will be more cases where AI advice either:
When four out of ten in the survey say they largely trust the information they get, it says something about the risk: Trust can outpace quality assurance. People in vulnerable situations may interpret a well-written answer as professionally correct, even if it was generated without clinical judgment.
This doesn't mean AI always gives poor help, but that error margins must be handled as a health problem — not as a "technical glitch."
The most underestimated risk is not that AI "gets things wrong"; it's that AI can become too good at providing immediate relief. If a conversation is always available, always empathetic, always tailored to you — it can create a habit that resembles addiction:
In a five-year perspective we may see a new phenomenon: digital self-medication with conversational AI. This will particularly affect young people, the lonely, and those with social anxiety — groups already vulnerable to replacing human contact with digital substitutes.
If an AI service is free, attention and your data are often the product. The ethical challenge becomes acute when the system learns:
Then "psychological support" can slide into an attention economy — a model where the goal is maximal usage time, not maximal health. In a scenario without clear regulation we could see AI that subtly:
This is a future where ethics become as important as technology.
Mental health data is some of the most sensitive we have. Conversations about trauma, sexuality, substance use, shame or suicidal thoughts are information that can harm people deeply if leaked, sold or used for profiling.
The same Shifter piece also raises a concern: energy use and climate footprint. That may seem like a sidetrack, but it points to an important mechanism: when people become aware of costs (energy, climate, privacy), pressure for regulation increases.
Over the next five years we'll likely see:
The most critical point will be consent. It's one thing to consent to a model learning from your text to improve a service; it's another to consent to your vulnerable moments being used for commercial optimization.
There is a fundamental distinction between:
In practice the boundaries will become blurred, because users don't necessarily think in categories. They simply ask: "What should I do?"
To address this uncertainty the health sector will likely develop new roles:
At the same time we may see a "hybrid model" become common: AI performs mapping and follow-up between sessions, while psychologists handle assessment, relationship and treatment. This can increase capacity without reducing human accountability.
But for this to be defensible, we need clear requirements for:
The fastest adoption often happens among the young, and the Shifter article notes that AI has become part of everyday life "especially for younger people." School and student life are therefore logical arenas where mental health effects will be visible.
On the positive side, AI can help pupils and students with:
On the negative side, AI can worsen comparison and perfectionism. When "everyone" has access to an assistant that can improve text, ideas and presentations, the threshold for what is considered "good enough" may rise. The result can be:
In a five-year perspective we will likely see schools develop a form of digital health literacy: not only source criticism, but also how to use AI without losing self-esteem, autonomy and rest.
Work and mental health are closely linked, and AI will be embedded in the tools we already use: email, Teams/Slack, documents, CRM. This will make "micro-coaching" common: phrasing suggestions, prioritization and decisions in real time.
AI can reduce stress by:
But it can also increase stress by:
If AI is used to analyze communication, tone, absence patterns or "risk of burnout," the intention may be good, but the consequence can be that employees feel mapped in their inner lives. In a five-year perspective the best employers will be those who create clear, trust-based frameworks: AI for support, not AI for control.
The Shifter article highlights that 48 percent are worried about AI's climate footprint. This is relevant for mental health in an indirect way: climate anxiety and technological guilt can increase when people feel their digital habits contribute to a bigger problem.
When people experience that technology use collides with their values, it can create cognitive dissonance and stress. In the years ahead, energy-efficient AI, local processing (on-device), and more transparent environmental accounting will therefore not only be climate arguments, but also help users feel confident that they are not "making the world worse" to get support.
This can also create a new social norm: it becomes "responsible" to use AI in ways that are energy-efficient, privacy-friendly and purpose-driven.
AI ethics in mental health can quickly become abstract. But in practice the next five years will be decided by a few concrete principles.
AI must not be marketed as a therapist if it is not one. Users must understand the limitations, especially in crises, self-harm and severe depression.
It should be standard that users can:
When AI is used in digital health, it must be clear who is responsible when something goes wrong: the provider, the service owner, the employer or the health institution. Without a chain of responsibility the risk is shifted onto the user — often the most vulnerable.
AI becomes a safe, regulated part of digital health. It functions as support between human contact points, with clear crisis routines, documented effect and strong privacy. The result is earlier help, less stigma and reduced pressure on health services.
Free "therapy-AI" dominates, built for engagement and data collection. People become more dependent, more monitored and more influenced. Professional treatment becomes more expensive and less accessible. Mental health becomes a market, not a right.
Resource-rich groups use quality-assured, private solutions and get the best outcomes. Others use free, unregulated services with greater risk. The mental health gap widens.
The most realistic outcome is scenario C — unless we make active choices now.
If we take population adoption seriously — and the rapid increase Shifter reports — we should act before habits are set.
Artificial intelligence is becoming more than a tool. When 68 percent have already used AI in the past year, and many do so daily, we are in the middle of a transition where the technology can become a permanent part of our mental self-care. That can bring huge gains: low-threshold support, faster coping and earlier help.
But precisely because AI can feel safe, accessible and empathetic, it can also become a source of new problems: poor advice in vulnerable moments, emotional dependence, privacy risks and market-driven manipulation. The next five years will therefore not mainly be about how "smart" AI becomes, but about how responsibly it is integrated into our lives.
If we succeed, AI can become a robust complement to human support — a tool that makes us more capable and less alone. If we fail, it can become a new arena for pressure, surveillance and loneliness in disguise.
The choice is not made in 2030. It is made in the products we accept, the regulations we demand, and the habits we form — right now.