AI-powered therapy to set minds at rest

It made headline news when the leader of the UK’s biggest relationship support charity revealed the organisation was considering the possibility of deploying chatbots using artificial intelligence (AI) for live chat counselling services. The chief executive of Relate, Aidan Jones, told an interviewer that he was looking at the potential of “non-human intervention” in 2019 to assist his 1,500 online counsellors, as they were becoming stretched by demands made on the service.

AI can “learn as it interacts with different clients”, Jones says. “We know that some clients prefer the relative anonymity of Live Chat services, as it makes them feel more comfortable opening up about their problems, so this is an area where AI could have potential. The use of AI in the therapeutic process is being looked at internationally – and we’re watching with interest.”

As Jones notes, developments in the potential of digital assistive technology to help treat a range of mental health issues are growing, and AI is proving a foundation technology for bots designed for ‘human-like’ interaction with individuals. Their proponents say that these ‘virtual therapists’ can help with disorders from low-level depression to social anxiety. They can also play a part in destigmatising mental health monitoring in homes and workplaces.

NHS reports indicate that only around one in eight adults with a mental health problem receive treatment; a shortage of mental health professionals in some areas probably contributes to this relatively low figure. Medication is reported as the most common type of treatment for a mental health problem, but given the rising scale of awareness, and the popularity of lifelogging, some experts believe digitalised mental healthcare could bring lasting benefits to a wide range of people who might otherwise go untreated.

“Traditionally, mental health treatment has only been available to those who are ‘diagnosed’ and hasn’t been measured in any standardised way to ensure quality care and efficacy,” says Alex Boisvert, CTO at Ginger, a company founded on the idea that data from people’s interactions with mobile phones provides insights into their emotional health. Ginger has collected billions of data points to inform a behavioural health system that supports its teams of coaches, therapists and psychiatrists.

It’s fair to observe that across a range of applications, the term AI – and its subsidiary technology, machine learning (ML) – can be applied without specific elucidation of the part it plays in core functionality behind some ‘AI-powered’ chatbot platforms. It’s also fair to note that as AI has been applied to different application areas, the term has been used as a conceptual approach as much as a technological descriptor. At the same time, the AI component of any solution is likely to be its exponent’s most valuable intellectual property, an asset it is reluctant to discuss in detail.

Broadly, AI-enabled mental health applications can be deployed in three ways: as decision-support tools for mental health practitioners; to customise the digital patient interface with human therapist intervention for one-to-one counselling; and to steer patient interaction (via remote devices) with semi-autonomous ‘virtual therapists’, again digitally.

Many of the applications that have emerged set out to provide first-line assistance to human therapists as growing awareness of mental wellbeing places greater demands on health resources.

AI also holds the promise of making healthcare support available to people who have difficulty with initial human interaction when it forms part of their treatment, but who would feel more comfortable talking to a non-human therapist.

None of the AI-based systems now prototyped or deployed are totally devoid of human involvement. The human therapists are never far away; indeed, it’s they who are the ghosts in the machine learning, who correct AI/ML’s tendency to issue inadvertently biased guidance based (for example) on age, ethnicity or gender.

‘This evolution of psychological treatment should play a part in supporting employers to help employees with their mental wellbeing’

The use of advanced digital technology for mental health support has, however, been relatively slow in coming, says Valentin Tablan, senior vice president for AI at Ieso Digital Health, a provider of evidence-based online cognitive behavioural therapy (CBT) for people experiencing common mental health issues. “There has been a technological revolution in physical health over the last 50 years which can be seen, for example, in CAT and MRI scanners, but this has not been matched when it comes to mental health,” Tablan points out. “One possible reason for that, in the field of talking therapy in particular, is that most clinically relevant information is contained in patient and therapist language, for which standardised analysis techniques are not as well established as for blood chemistry, for instance.”

One technique suitable for the analysis of language, Tablan says, is natural language processing (NLP), a sub-area of AI concerned with the interactions between computers and human (natural) languages, in particular how to program computers to process and analyse large sets of natural language data. “There is a significant body of academic and industrial research in NLP, and these techniques have been deployed in applications, including mainstream apps like Apple Siri, Amazon Alexa, or Google Assistant,” adds Tablan. “Like many areas of AI, NLP has also benefitted from the recent development of ‘deep learning’, which has led to significantly more accurate language understanding models.”

Question-answering digital therapists have, in fact, been practising since 2014, with the arrival of Ellie, an embodied avatar who is the computer-generated interface of an ongoing project from the University of Southern California’s Institute for Creative Technologies. The system was developed with the US military, which wanted support tools to help it with service people stricken by post-traumatic stress disorder. For many affected by this, the most difficult stage in recovery is to start to talk with another person about their experiences.

The system is built on two software components, Multisense and SimSensei. Using video monitoring, Multisense tracks and analyses, in real time, a patient’s multimodal signals: facial expressions, body posture, acoustic features, linguistic patterns and higher-level behaviour descriptors (e.g. attention and fidgeting). Multisense infers from these signals and behaviours indications of psychological distress that directly inform SimSensei, the AI-integrated software that humanises Ellie. This can sense audio-visual signals captured by Multisense and engage in face-to-face interaction: Ellie instantaneously reacts to the perceived user state and intent, through her own speech and animation. So when Ellie appears to the patient on a screen, she responds to the perceived user state and intent with her own computer-generated speech and gestures.

The system’s early identification of a patient’s distress level can generate the appropriate information to help a clinician diagnose a potential disorder. User-state sensing can also be used to create patient profiles that help to assess behaviour change over time. However, Ellie is programmed only to encourage people to talk about their feelings, not to offer advice.

AI-founded virtual therapy platforms aim to minimise direct one-to-one interaction with therapists, so that expertise can be pooled into creating collective support, and that the system backends, where possible, draw on the newest digitally available professional guidance and research.

Some systems (such as Vancouver-based company AI-Therapy’s Overcome Social Anxiety) have machine-learning functionality that applies progressive elements learned with one patient to another, where human therapists deem it appropriate.

Ieso Digital Health is another provider that owns a large dataset of therapy-related data, which it’s using to develop a machine-accessible understanding of therapy, offering a data-supported model of therapeutic practice. “This model underlies the NLP-based tools under development to support Ieso therapists in their clinical work,” explains Tablan. “The application of NLP and deep learning to mental health data has great promise of closing the technology gap between physical and mental health. Like most applications of AI, and especially deep learning, this requires large amounts of high-quality data.”

Ieso data scientists have used AI to analyse more than 90,000 hours of therapy sessions and learn which treatment protocols have been effective. This data will be used to advise therapists about the best way to treat a patient, based on the information provided ahead of the session, Tablan explains.

A development to the service that is about to be launched makes real-time, AI-powered support available to therapists during therapy sessions, aiding their clinical decisions. This support tool provides therapists with extra information about each patient, the company says, and helps them to make more accurate diagnoses and considered decisions about course of treatment. Therapists deliver CBT to patients in real-time through written conversation online using a virtual therapy room. Therapy is accessible from a patient’s personal computer or smartphone.

“Therapy delivered by Ieso therapists is delivered by online written communication. This means that with each therapy session, a transcript exists [that records] what was said,” says Tablan.

“When Ieso began, we knew that having these transcripts was a very helpful thing, both for the therapist and for the patient,” he explains. “What we have since realised is that these transcripts also provide us with knowledge about what works and what does not work in therapy.”

The ability to recognise and correlate patterns of experience that could inform treatment for new patients is an attribute shared by Tess, the AI-enabled mental health chatbot from US-based company X2AI.

“One way this is done is through Q&A,” says CEO and founder Michiel Rauws. “If someone asks the chatbot a question that it is unable to answer, the system alerts the X2AI psychology team that an answer is required. The team then can decide on whether the question is relevant and worth answering – in that Tess is not meant to answer questions like ‘is the sky blue?’, but should be able to answer ‘can I talk to a person?’ Once an answer is provided, all versions of Tess benefit from the change.”

As the 2020s unfold, it’s likely that mental health monitoring will become an increasingly accepted aspect of employment. Employers will recognise that, as well as looking after the physical safety of their workforce, they have a duty of care towards their mental wellbeing. Moreover, mentally healthy employees are more productive and contribute to a better workplace culture.

Studies suggest, however, that a high proportion of adult mental health issues are undiagnosed, and many that are diagnosed go without treatment – in part due to challenges with accessing services by taking time away from work, or because of patient concerns that being identified as having mental health problems will have an adverse effect on professional relationships.

“As the stigma around mental health continues to lift, we will see employers take these conditions more seriously [in regard to the mental wellbeing of their work teams],” says Ieso’s Tablan. “Methods that embrace AI can play a big part in increasing access to care. This evolution of psychological treatment should also play a part in supporting employers to help their employees with their mental wellbeing.”

Use-case studies of employer-sponsored partnerships with digital mental health services are entering the public domain. For example, Saint Elizabeth Health Care (SE Health Care), a Canadian social enterprise specialising in home care for older people, has an ongoing project with X2AI’s Tess (see box). Digital media service provider BuzzFeed launched Ginger as a pilot to its staff: due to demand, now employees in the US, Canada, Australia and Germany have access, with 26 per cent of the workforce signed up.

Beauty store Sephora launched Ginger to 9,000 US employees in August 2018, with 7 per cent signed up to date. Statistics on the effectiveness of these services are patchy, but Ginger says that more than 70 per cent of its members assessed experienced a partial or full response to care over 8-12 weeks of use. Over eight months (November 2017 to June 2018), 46 per cent of Ginger users identified as ‘at risk’ reported a 50 per cent decrease in their symptoms, reports the company.

“Our experience is that employers are primarily looking to increase access to care and reduce wait times for employees who are in-network in their health plan,” explains Alex Boisvert at Ginger. “That is why virtual and remote mental health services are critical to [extending] access to care.”

Social anxiety is a common disorder and recognised clinical condition that can debilitate sufferers at all points in their daily lives – and that includes the place of work. The contemporary ethos of team-working and collaboration can present acute challenges for people with social anxiety. AI-Therapy’s CBT-based Overcome Social Anxiety online treatment programme builds on years of research and real-world clinical experience, the company says. Its research is ongoing, with multiple clinical trials currently under way.

Users are guided through successive lessons designed to help them ‘target’ their anxiety. These include pre-recorded sound files from the company’s directors, psychologists Dr Fjola Helgadottir and Dr Ross G Menzies, that explain the key concepts behind CBT and the other treatment strategies. Some stages of the treatment include videos of people talking about their experiences with social anxiety.

Overcome Social Anxiety creates a customised experience based on the symptoms users report at the start. “Our social anxiety treatment programme is algorithm-based. Through a series of questionnaires, the program builds a model of the user and their mental profile,” Helgadottir explains. “It then automatically tailors a CBT treatment programme specifically for that user.” The system also includes exercises designed to target key aspects of a user’s anxiety profile and delivers automated feedback in written and audio formats.

Overcome Social Anxiety was introduced over six years ago. “As users complete the programme, we gather information about clinical outcomes, as well as feedback about their experiences using the programme,” says Helgadottir. “This data is used to make improvements.” Another advantage of online provision is that programmes can be updated as best practices emerge.

AI-Therapy currently sells its solution directly to end-users, but the company plans to offer its product through employers as well. “In an office setting, social anxiety is known to have a significant impact on employee wellbeing and job performance,” says Helgadottir.

X2AI’s Tess acts as the front end for its transformative digital behaviour change programme, accessible via a patient’s computer or smartphone, and built on a proprietary platform that operates as an electronic health record to manage users, content, metrics, custom features and more.

Rauws explains how the system came about. “Ten years ago, I was struggling with depression,” he says. “I found a psychologist who was able to help me by using talk therapy. Later, I noticed that when speaking with friends and colleagues, I was repeating the conversations I previously had with my psychologist. That’s when I realised: if I can help people by repeating these conversations, then we could teach a machine to do the same. That’s how Tess was born.”

X2AI’s psychology team includes psychologists and mental health experts. Their range of experience and diverse perspectives is important to reducing bias during the initial content creation process. Rauws says: “X2AI also contracts external psychologists to update and add content based on each professional’s experience with specific concerns and intervention.”

To develop Tess’s leadership coaching conversations, for example, the company collaborated with an expert coach who has served as a senior consultant for the American Psychiatric Association with a focus on workplace mental health. Clinicians and researchers from Stanford University supported content development for a Tess version that helps young women with eating disorders.

Tess can also help with relationship counselling, Rauws says, by focusing on the experience and emotions of individuals ‘conversing’ with Tess. In return, Tess may offer tips to improve communication or simply a chance to vent about an interpersonal conflict that has occurred.

“Tess is being delivered by many employee assistance plan organisations around the world as an additional source of on-demand emotional support for employees [inside and outside of their workplace],” Rauws explains. “Tess is now also being delivered as voice-enabled support through Google Home and Amazon Alexa to older adults to help cope with social isolation and loneliness.”

According to AI expert Dr Andrew Ng, chairman of the firm behind CBT-based mood-management programme Woebot, mental health “may be the killer app for chatbots ... Even with its current limitations, [AI’s] transformation of mental health care will help millions of people, sometimes through life and death decisions.”