Artificial intelligence (AI) in medicine has been characterized as a major disruptor with the potential to change the way health care is delivered. Interest in AI in Canada began early and is continuing, with a number of advanced AI research and development labs now established. This report is based on information from a limited literature search (English documents published between January 1, 2014 and September 5, 2019) and targeted stakeholder consultations. Stakeholder interviews — which included individuals involved in AI and mental health via research, industry, clinical practice, and lived experience — sought to supplement information from the literature and provide perspectives for decision-makers.

This Environmental Scan explored the types of AI emerging or currently in use in mental health. The scan also aimed to identify AI research and development initiatives, professional groups and organizations using or developing AI for mental health care (across Canada and internationally), and policy considerations for AI as they relate to mental health.

Types of AI for the prevention, diagnosis, or treatment of mental health issues

The most common types of AI applications for mental health are conversational agents and computerized adaptive testing. Conversational agents mimic natural human-to-human conversations and “talk” with participants and clinicians. They can be text-based or include an embodied agent (i.e., an avatar using body language, movement, and facial expressions). A familiar example is the customer service chatbot (such as in LiveChat Facebook Messenger). Computerized adaptive testing allows a computer to give exam-takers more targeted questions based on answers to previous questions.

AI use in mental health care

While several research and development projects are in progress, AI has not been widely integrated into clinical practice — although some AI programs are being used or explored for use in mental health services. In diagnosis and prevention, AI is mainly used to determine the presence of mental illness or assess risk (e.g., by analyzing words in messages or detecting changes in behaviour), and to direct users to appropriate support. In treatment, AI usually takes the form of a conversational agent that delivers psychological treatment (e.g., cognitive behavioural therapy).

AI research and development initiatives

The environmental scan includes a list of Canadian and international organizations working in the field of AI mental health research. Current research and development initiatives focus mainly on prevention, diagnosis, and prognosis. These initiatives include robot companions to address loneliness in elderly people, as well as the use of data obtained through brain scans, voice recordings, or wearable sensors to help with early or improved diagnoses of mental illness and predict responses to treatment (prognosis). There is also research on AI initiatives to support mental health care in military and veteran populations (e.g., early diagnosis of posttraumatic stress disorder and detection of suicide risk). In addition, the analysis of social media posts has been explored as a way to support the early detection and diagnosis of mental health illness in people experiencing major depression, anxiety, and postpartum depression. There also appears to be increasing interest in wearable devices that can record real-time data and notify individuals or their counsellors of progress, and in other mobile applications and e-health applications for client use.

Policy considerations

In addition to planning, incorporating AI into the care pathway must consider several factors. These include determining:

• how applicable lab-based findings are to a clinical practice setting
• the effectiveness of AI for mental health treatment
• the feasibility of adding AI to the workload of clinicians
• whether health-care providers focus only on patient data while overlooking an individual’s lived experience.

Future research could focus on the suitability of AI for mental health interventions (i.e., whether they should be used as a screening tool, a replacement for human therapists, or an adjunct intervention) and on including individuals with lived experience (e.g., asking them why they use AI and for what purpose).