Mental health needs are growing faster than health systems can respond.  Artificial Intelligence (AI) is emerging as one way to widen access – if we design and govern these tools responsibly and keep human care at the centre.

Over the past decade, I have worked in hospitals and health systems in four countries. In Bangladesh, where I trained as a physiotherapist, patients often came with back pain or fatigue – only for it to become clear that the root cause was untreated anxiety or long-term loneliness. In Thailand, I saw migrant communities whose mental health needs rarely even appeared in official data. Japan was a different lesson: social isolation had become so common it had its own vocabulary, yet stigma silenced discussion even within health services. And now in Australia, where I have started my PhD research, the pattern repeats. People are struggling, and mental health services remain out of reach for many.

That disconnect between need and access is what drives my curiosity!

“Can AI play a meaningful role in mental health support, especially for those current systems are failing to reach? Here I offer my perspective on what made me move working in this research, shaped by my own personal experiences and influences.”

A challenge every country faces!

The global burden of mental health related conditions is significant. According to the World Health Organization (WHO), around 1.1 billion people were living with a mental disorder in 2021, and most received no effective care. Access is not equally distributed. People in rural areas, culturally diverse communities, and people facing language or cost barriers often miss out on care. Older adults are also affected: WHO estimates around 14.1% of adults aged 70+ live with a mental disorder, and these conditions are often under-recognised.

Where AI can support right now

Conversational AI, in the form of chatbots and voice assistants, offers around-the-clock support. They do not require an appointment or a waitlist.  They can guide someone through a breathing exercise at 2 a.m., ask gentle check‑in questions, or offer information about when to seek professional help. Before 2022, most chatbots used decision trees. Today’s large language models have changed that. These newer systems hold more natural conversations and can adapt their responses in ways that feel warm to many users. They are not a replacement for a real person or professional treatment, but they open possibilities that simply were not there a few years ago.

I am especially interested in what this could mean for groups that tend to miss out on mental health support: older adults living on their own, young people who will not set foot in a clinic, migrant workers who need support in a language their local system does not offer.

For a Burmese worker in a Thai border town, a chatbot that speaks their language and is available on a smart phone could be the first mental health resource they have ever had access to. For someone in any of these situations, a chatbot might be the first step toward thinking about their own wellbeing before they are ready to talk to a professional.

Designing for inclusivity, not just usability

Inclusive design is an all-age strategy. If a tool only works for digitally confident users, it can widen existing inequities.

Older adults are an important test case for quality design because they surface issues that affect everyone: trust, usability, privacy, and clarity about what the tool can and cannot do. In Australia, the 2025 Australian Digital Inclusion Index reports high levels of digital exclusion among people aged 75+.

Access therefore cannot be assumed.  Designing inclusive systems means building simple, intuitive interfaces; providing larger text options and voice input; supporting multiple lang; and having clear handovers to human care when risk elevates.

Leadership and policy choices will determine impact

The main barrier now is not technical capability; it is leadership capability and policies we make. Health services need implementation models and well thought policies and frameworks that are safe, transparent and accountable from day one. WHO has already set out governance guidance for large AI models in health. In Australia, AHPRA makes clear that professional responsibilities remain with practitioners, and the Royal Australian and New Zealand College of Psychiatrists reinforces safe, ethical clinical responsibility.

None of this should be taken as uncritical enthusiasm. AI systems get things wrong. Large language models can generate inaccurate or misleading information, and the WHO has warned that strong governance and safety measures are needed before these tools go into health settings. For vulnerable populations, the risks are serious. Privacy is a real concern. So is accessibility. And there is a real danger that if the technology is poorly designed, it ends up reinforcing isolation rather than reducing it.

The interface matters as much as the algorithm behind it. Voice input, large text, simple navigation, multilingual support: these are not optional extras for users who already face barriers to access. If these tools are only usable by people who are already comfortable with technology, they will widen the gap rather than narrow it.

Health leaders considering these tools should be asking hard questions. Who validated this system, and with which populations? What happens when the chatbot encounters a user in crisis? Where does the data go, and who controls it? Without clear answers, deployment is premature.

This is why the conversation must involve more than just technologists. Health leaders, policymakers, clinicians, and the communities these tools are meant to serve all need to be part of how they are developed and evaluated. Technology alone will not fix the mental health gap. But built with the right input and grounded in evidence, it could help.