The use of AI for mental illnesses

Mental illnesses are widespread: according to the German Association for Psychiatry, Psychotherapy and Psychosomatics, they affect 17.8 million people in Germany. Will artificial intelligence also revolutionise the treatment of mental illnesses?

Many people with mental illnesses need therapeutic help. However, finding a therapist can be a challenge – especially for regions outside of urban centres, where the German Federal Chamber of Psychotherapists has identified a serious shortage of psychotherapists.

Can artificial intelligence help here and step into the breach? Indeed, a great deal of research is currently being done into the possible uses of AI in the diagnosis and treatment of mental illnesses. Various therapeutic chatbots are already in use. But let's take this in order.

AI is good at analysing large and complex amounts of data. Data can also be collected on mental illnesses using questionnaires, self-reports, written therapy sessions, video and audio recordings. But smartphones and smartwatches can also record patterns of interaction and body data. All of this data contains information about the state of mental health. 

AI detects risks and supports diagnoses

It makes sense to train AI models with such data and then use them to diagnose mental illnesses. However, it is not only fully developed clinical pictures that are identified. In principle, AI could detect the very beginnings or even the mere risk of an illness. Some examples:

• The Commitment research project at the Mannheim-based Hector Research Institute for Artificial Intelligence in Psychiatry is using AI to try to uncover links between mental illnesses and physical factors such as diabetes, obesity and high blood pressure.
  
• A project of the Institute for Applied Informatics in Leipzig is working on an application that analyses data from smartphones and smartwatches. From information such as movement patterns, heart rate or the way someone operates their smartphone, the AI can detect mental changes and give the user advice, such as getting more sleep or doing more exercise.
  
• The AI chatbot ‘Limbic Access’ in the UK communicates with people about their problems, assesses the symptoms, gives advice and, in some circumstances, recommends therapy. It can also search for and request a suitable therapy slot directly. Overall, however, the system of psychotherapeutic care in England is difficult to compare with the situation in Germany, where the allocation of therapy places is organised differently. You can find more information here: https://www.tagesschau.de/wissen/technologie/ki-psychotherapie-102.html
  

Intelligent chatbots – digital therapists?

‘Limbic Access’ also represents another strength of AI: it is very good at interacting with people. AI-controlled chatbots are probably the best-known applications of artificial intelligence. A special feature of mental illnesses is that their symptoms usually also show in what and how someone speaks; therapies also largely involve conversations. Shouldn't AI then find a large field of application here? Could AI ultimately replace the human therapist?

Chatbots in the form of medical apps have actually been used for some time in prevention and treatment. Originally, such apps had little to do with artificial intelligence. However, various applications are gradually being retrofitted with AI or redeveloped. Some of the many models are highlighted below:

• The chatbot ‘Woebot’, developed at Stanford University, is a pioneer of the genre. It recognises symptoms through conversations with users, helps identify illnesses and suggests exercises and reflection. The app checks in every day to ask about the user's state of mind and to work with them on practical techniques tailored to their needs. Woebot is based on cognitive behavioural therapy.
  
• The German online therapy platform MindDoc also offers an app that asks users questions about their well-being several times a day, assesses their emotional health and occasionally provides personalised feedback with evaluations and suggestions for exercises. The app is designed to help users monitor their own well-being over time and is also used to accompany online therapy.
  
• The Mannheim-based Central Institute of Mental Health has developed an app for young people that is designed to prevent mental illness. The app allows users to describe their current mood, provides personalised tips and suggests exercises.
  

What's next?

The story of AI in the treatment of mental illness has only just begun – and further fields of application are on the horizon. For example, AI models can also support therapists in planning and evaluating therapies. In a project at the University of Basel, for example, artificial intelligence is analysing video sessions and calculating the probability of a patient dropping out of therapy.

A spectacular advance could be the combination of AI with virtual or augmented reality. For example, AI could create a customised, interactive virtual training environment for patients in which they could practice coping with everyday situations or overcoming fears.

The limits

So, are capacity problems in psychotherapy on the verge of being solved? Caution is advised – when it comes to mental illness, expectations of AI should not be set too high. It has limitations that have a serious impact on psychotherapy. AI does not really understand what people say, nor is it capable of genuine empathy. It basically proceeds statistically: ‘In the learning data, people with these language peculiarities had depression in 77% of cases.’ Or: ‘If a person does not respond to therapeutic suggestion A, there is a high probability from their medical history, after comparison with other medical histories, that they will respond to suggestion B’.

However, mental illnesses manifest themselves differently in each person. They are dynamic, change their appearance, and experience both progress and setbacks. Mere statistics do not do them justice. The empathetic understanding of therapists must therefore classify and put the AI findings into perspective.

Using chatbots without the support of a professionally qualified person is a stopgap. The AI bot cannot replace the therapist – but it can help people who can't find a therapist or who are hesitant to seek treatment.

Incidentally, in a British study, ChatGPT-4 was tasked with assessing whether the author was suicidal based on short texts. At the same time, experienced psychotherapists were also consulted. The AI came to the same conclusions as the humans. However, since an AI does not reveal how it arrived at its assessment, human observers cannot judge how reliable the qualities of the model actually are. There is much to suggest that, in the long term, AI in the treatment of mental illnesses will not go beyond the role of assistant, merely supporting human decisions.

Text: Thorsten Kleinschmidt