The Internet and social media have become fraught spaces for mental health, particularly for youth. In a 2019 project called ‘Selfie Harm’, British photographer Rankin photographed 15 teens and asked them to edit their photos to make them “social media-ready”. The teens dramatically changed the shape of their features, added makeup and/or smoothed out their complexions, making themselves almost unrecognisable. This provided a unique insight into the damaging impact of social media on young people’s self-image.

Then entered COVID-19. The World Health Organisation’s ‘Social Media and COVID-19 Report outlines how the pandemic and lockdowns were not only direct sources of stress, anxiety and isolation, but also increased reliance on social media as life and work activities moved online.

However online spaces can also fill gaps in and augment the provision of mental health services. Digital mental health tools are greatly enhancing current methods of detecting, diagnosing and treating mental health, as well as facilitating the creation of novel methods. They are able to increase the accessibility, affordability, accuracy and speed of the diagnosis and treatment of mental health disorders and increase patient autonomy and sense of responsibility for their own mental health care.

Detecting and diagnosing mental health disorders

Mental disorders are commonly diagnosed by asking the patient a large number of questions (eg. questionnaires based on the Diagnostic and Statistical Manuel of Mental Disorders, International Classification of Diseases and Symptom Checklist 90-Revised diagnostic guidelines) and interpreting the responses. Tutun et al’s 2022 research on AI-based Decision Support Systems for Predicting Mental Health Disorders outlines the issues stemming from how these guidelines are complex, involve an excessive number of questions and require a significant amount of time to complete. They also raise the possibility of human error and biases in the mental health professional who interprets the responses.

To address these challenges, various studies have taken Tutun et al’s approach of designing and deploying AI-based decision support systems. An algorithm identified the diagnostic questions which had to be included for an accurate diagnosis. Machine learning models were then trained to predict the existence and type of mental disorder by using participants’ answers to these questions and other historical data. Tutun et al’s system was able to automatically diagnose mental disorders using only 28 questions without any human input, to an accuracy level of 89%.

AI has also enabled the creation of novel assessment tools to detect mental disorders. There is a significant growth of programs which analyse social media posts to determine whether the user is suffering from a mental illness. For example:

• Dartmouth College’s Emotion-based Modeling of Mental Disorders on Social Media focused on the emotional – as opposed to textual – content of conversations on Reddit to identify emotional disorders, such as major depressive, anxiety and bipolar disorders. The paper outlines how the researchers trained the model to label the emotions expressed in user posts and identify the patterns of emotional transitions that are associated with emotional disorders. They tested the model on posts that were not used during the training and found that the model accurately predicts which users may or may not have these disorders.
• A Pakistani study on Anxiety-displaying activities recognition found an AI program was able to detect anxiety symptoms in participants, whose behaviours were monitored by motion sensors, with 92% accuracy. This may lead to the introduction of motion sensors in smartwatches and an accompanying app, which would greatly assist in the early detection and evaluation of anxiety and other mental illnesses for smartwatch users.
• Chinese research into the Application of AI on Psychological Interventions and Diagnoses has shown that deep learning, a type of AI-based machine learning, can be an efficient and accurate tool for analysing large volumes of clinical files and social media information, to identify factors which may increase the risk of mental illnesses, such as certain personality traits, traumatic or stressful events or relatives with a history of mental disorders. Deep learning can even be applied to neuro-imaging (fMRI scans) to help detect brain changes and neuropathological mechanisms associated with mental disorders.
• The University of Sydney’s Brain and Mind Centre has national medical research funding to apply modern data science methods to build explainable and integrated machine learning models that can be utilised by health services to make real-time, data-informed clinical decisions in youth mental health care.

Treating mental health disorders

AI is also capable of assisting in the treatment stage. One program is Mental Health Chatbot Tess, which learnt how to talk like a therapist by exchanging text messages with support workers, nurses and therapists about what was troubling them. Users then provided feedback on whether Tess’ response was helpful. Communicating with the user through text message exchanges, Tess doesn’t use pre-selected responses but adapts to the shifting information provided by the user to provide suitable clinically proven coping skills and strategies. Chatting with Tess has been shown to reduce, on average, depression by 29% and anxiety by 30%.

Figure 1: Tess the AI Chatbot

Separate from the application of AI, the COVID-19 pandemic led to the widespread adoption of online counselling and cognitive behaviour therapy, in the form of online sessions, videos or workbooks which are guided by a mental health professional. Anyone who personally experienced the pandemic can imagine the numerous, ongoing advantages of online therapy: convenience, lack of travel time, reduced costs and the ability for mental health professionals to work with a greater number of patients.

One treatment-focused online intervention sought to bring the diverse human stories behind the Ashmolean Museum’s art and artefacts to young people. This Oxford University randomised controlled trial was the first reliable experimental evidence lending support for online engagement for depression and anxiety in young people. Participants reported that engaging with the stories, art and culture enhanced their perspective, their sense of connection on a human level and provided opportunities for learning, escapism and creativity. This disrupted negative thought patterns and increased feelings of calm and proactivity, notwithstanding that this study was conducted whilst COVID-19 restrictions in the first UK lockdown increased.

Flow chart of participants in the Oxford University trial

Key: NA: Negative Affect, PA: Positive Affect, FERT: Facial Expression Recognition Task, PILT: Probabilistic Instrumental Learning Task; WoB: Ways of Being, the human stories-based online intervention; Ash: the Ashmolean’s typical museum website, as a comparison to WoB

As one participant said, “because it’s not in a therapy setting of someone telling you, this is how you should try and accept yourself and get better…I think seeing it in a more of an artistic setting with artists’ works. It took away the mental health focus and let me come to those conclusions by myself.”

The need for caution when employing digital mental health tools

Reading the Frequently Asked Questions about Chatbot Tess hints at the risks these tools may have: “Is Tess confidential and secure? Will anyone else see my conversations with Tess?” There is the risk that confidential patient data obtained through digital mental health applications will not be adequately secured and potentially breached or shared with third parties.

Further, it is important to not overestimate the effectiveness of digital tools, which in many cases cannot replace the role of a trained mental health professional. The Guardian article on Tess notes that human-robot interaction expert Julie Carpenter doubts AI will ever be able to truly understand the subjective experience of a human.

Finally, while digital mental health tools can provide more affordable and convenient ways for rural or lower-income communities to access mental health resources, these communities also face the greatest digital barriers such as poor network coverage or a lack of digital literacy. This catch 22 is especially pertinent to Australia, with the 2021 Australian Digital Inclusion Index recording 11.1% of Australians being highly digitally excluded, with regional areas and low income households being particularly left out.

Yet the beneficial potential of digital mental health tools remain

While online spaces bring challenges for the mental health and privacy of its users, it is a promising way to identify people with mental health disorders and provide effective treatment, especially for those who would otherwise have received limited or no treatment.

The research above shows digital mental health tools’ potential to:

• Increase accessibility and affordability of diagnosis and treatment of mental health disorders, especially for rural and low-income communities;
• Increase engagement with young people, many of whom are reluctant to seek help;
• Reduce time and costs associated with misdiagnosis, overdiagnosis and unnecessary treatment; and
• Increase autonomy and responsibility in patients for taking care of their own mental health development.

To make the most of these benefits:

• The developers of these tools, as well as the mental health professionals employing them, must ensure confidential patient data is adequately secured.
• This is not a task suited to ‘no humans in the loop’ AI. The AI and skilled professionals will need to work together, and probably with a level of human governance and review of the AI learning.
• Effort should be directed to reducing the digital divide in rural and lower income communities. These communities typically have the least access to mental health professionals, but stand to most greatly benefit from online tools’ affordability and accessibility.

Read more: Digital Mental Health for Young People: A Scoping Review of Ethical Promises and Challenges