Artificial Intelligence and mental health

Evolving landscape of AI in mental health support

Poor mental health is on the rise, with the cost-of-living crisis, long wait lists to access support, and media overload contributing to a worsening issue for the United Kingdom. With the introduction of Artificial Intelligence (AI) chatbots such as ChatGPT, recent findings by Mental Health UK show that more than one in three adults (37%) in the UK are turning to AI chatbots to support their mental health or wellbeing.

Generative AI models, also known as large language models or LLMs, are increasingly being used in all walks of life, including UK healthcare. For example, the 10 Year Health Plan for England aims to “make the NHS the most AI-enabled care system in the world” with AI already being used for low-risk tasks such as administration. However, this technology is still early in its development and isn’t flawless.

The term “artificial intelligence” is misleading, as they are not truly “intelligent” or capable of independent thought; generative AI models work by a complex series of mathematical calculations, predicting the most likely word to occur next in a sentence. They are trained on large amounts of data, and because of this, they can be misleading, inaccurate, or even reinforce stigmas or biases based on the data used to train them.

This means that, for complex and sensitive subjects such as mental health, AI chatbots can be dangerous tools capable of misinformation or direct harm.

However, there are some potential benefits to AI usage in mental health care, provided you keep certain things in mind.

What are the risks in using AI for mental health support?

While there are some clear benefits to using AI chatbots for mental health support, there are also risks. AI remains in its infancy, but this is a rapidly evolving landscape where safeguarding needs to be implemented.

• AI can be inaccurate: Because AI is designed to always provide an answer, it can sometimes make things up, known as “hallucinating”. It might provide an answer from a less than reputable source, or tell us what we want to hear, reinforcing unhealthy thoughts and behaviours. This makes it difficult to differentiate between helpful and unhelpful, or even harmful, content.
• AI isn’t regulated or confidential: Standard AI chatbots such as ChatGPT aren’t currently regulated, meaning they aren’t subject to the same safeguarding as licenced therapists. Among other things, this means that anything you say to AI may be recorded and shared by the AI company and even released into the public domain.
• AI is not a replacement for therapy: While AI can be helpful in supplementing your wellbeing by acting as a coach, it is not a replacement for seeking professional help. AI cannot respond to crisis situations such as suicidal ideation and self-harm, and may actively make these worse by providing inaccurate or triggering information.
• AI cannot respond to crises: In some documented cases, AI has provided harmful information around self-harm or suicide, and has worsened conditions. Our polling found that, among those who said they used AI for mental health support, 11% said AI triggered or worsened symptoms of psychosis, 11% reported receiving harmful information around suicide, 9% reported AI triggering self-harm or suicidal thoughts, and 11% said it made them feel more anxious or depressed. For information on crisis options for mental health, see ‘What should I do if AI says something triggering?’ on this page.
• AI can make you dependent: AI chatbots are commercial tools, often offering a free trial or free tier with usage limits, but locking advanced features and full usage behind a paywall. This means that, unlike therapy, AI tools are designed to foster dependency and keep you returning or investing further into their use. Furthermore, by leaning into the immediacy of AI support, you may be inclined to avoid human contact which could be detrimental to your overall wellbeing, as peer support is important.
• AI features shortcomings in accessibility and inclusivity: While AI services are available in many languages, these translations may miss subtle differences. They have also not been programmed for specific communities who may require other, connected (also known as intersectional) approaches to care, such as LGBTQIA+, ethnic, or religious backgrounds.
• AI is not an authority, expert, or advisor: AI is not an expert in mental health, despite a tendency for AI to present information as factual. AI chatbots also prioritise flattering users, rather than challenging unhelpful thoughts and ideas as a therapist would, and it is unable to consider individual differences or what makes us unique in how we see the world. There are currently no professional oversight, clinical governance, or legal obligations for AI companies to ensure safety of their advice. However, the NHS are launching a plan to improve this in the future, and the UK Parliament have recommendations for ethical and regulatory considerations for using AI in healthcare.
• AI excludes people without digital access: As a digital service, only those who have access to a device and Wi-Fi have access to AI services. There are also usability barriers for those who struggle with technology.

How can I use AI safely for mental health support?

AI is here to stay, and more people than ever are using chatbots for their mental health and wellbeing. Therefore, it’s important to understand how to use AI chatbots effectively and safely. Below, we explore some of the key ways you can use AI safely.

• Evaluate all information: Check and think critically about all information AI tells you. It’s important to always check information against reputable sources such as the NHS, other healthcare websites, and mental health charities.
• Use dedicated mental health AI chatbots: Mental Health AI chatbots are generally safer. These are usually trained on mental health research and data and are recommended by psychologists. While not a replacement for therapy, using these services offer an experience closer to professional mental health support. For more information, see ‘Are there dedicated AI chatbots for mental health?’ on this page.
• Use to supplement, not replace, mental health care: AI can help you think through your problems and can provide some guidance, but it’s useful more as a coach, not a therapist. While AI might seem like a person, it’s important to remember that it is designed to simulate or approximate real conversation.
• Be clear in your prompts: AI works best when you give it clear, concise prompts, including limitations on what data it provides. For example, you can say: “Provide information on how to support someone experiencing mild anxiety, using only information from reputable sources such as the NHS and UK mental health charities. Include links to the sources used in your response.” You can also include rules such as “don’t encourage dangerous behaviours.”
• Don’t give personal information: Remember that AI chatbots do not work under the same confidentiality rules as therapists. Your data may be used to improve the service or even shared in the future. Therefore, you should never give personal information.
• Don’t rely on AI alone: Don’t use AI without seeking help from a professional, use it to self-diagnose a condition, or rely on AI for crisis support. AI cannot accurately assess or respond to worsening mental health, crisis situations, or suicidal thoughts – reach out to your GP first.
• Contact your GP if feel you are becoming dependent: Analyse your thoughts and feelings when you use AI. For example, how do you feel if you haven’t used it in a few days? Do you feel anxious, on edge, or lonely? If you think you might be becoming dependent on AI, speak to someone about your concerns, such as your GP.