‘It gave me hope’: How a single AI session put Kelly on a path to eating disorder recovery

Thirty-three-year-old Kelly has struggled with an eating disorder for most of her life. But it’s only in the past few years, with the help of AI, that the tide started to turn.

As a child, Kelly, who is using a pseudonym for privacy reasons, developed Avoidant Restrictive Food Intake Disorder (ARFID), meaning she significantly limited the volume or kinds of food she consumed.

A former ballet dancer, she says the culture around exercise and disordered eating also contributed to the development of anorexia.

Credit: Marija Ercegovac

Kelly was part of the trial for a world-first artificial intelligence-powered chatbot, called Eating Disorder Electronic Single-Session Intervention, or ED ESSI for short. The chatbot’s lived-experience videos and the opportunity to talk to a neutral party to arrive at a diagnosis were unlike anything she had experienced in the healthcare system before.

“Any time I have tried to seek help, I was really invalidated,” she says. Certain behaviours, like purging without bingeing, meant she slipped through the cracks.

“It [ESSI] gave me hope,” says Kelly, who is now in treatment.

Professor Gemma Sharp, founding director of The Consortium for Research in Eating Disorders, led the development of the chatbot, and says the main impetus for its design is the long waiting times, exacerbated after COVID, patients faced before accessing clinical treatment.

The ED ESSI chatbot, which has been trademarked, was developed at Monash University and is now owned and managed by Sharp.

“We know that, if you leave people waiting for treatment, they are less likely to take up the opportunity of treatment when it’s presented to them, and also that their symptoms can deteriorate,” she says.

There is no official data in Australia as to the length of, or the number of people on, waiting lists but a 2020 report from the Butterfly foundation identified waiting times as a key issue for respondents.

Helplines, such as the Butterfly foundation’s free national helpline, have long helped fill this gap for eating disorder patients. But Sharp says the prospect of calling someone can be daunting for some patients.

“Some people really love helplines. But there are other people who would never choose to use a helpline. It’s about giving them an option for engaging with treatment that we know is evidence-based and effective,” she says.

ED ESSI provided one 30-minute session to 60 patients around Australia on waiting lists for eating disorder treatment. The chatbot is not designed to replace treatment for patients, but to provide interim support while they wait for care. The information discussed by the patient with the chatbot is then passed on to the patient’s clinician.

While one half-hour session may not seem like much, Sharp says it can still have a powerful effect.

“Single Session Interventions are considered a newer innovation, particularly in mental health care, showing that even a small dose of therapy can be very beneficial and start to change some of those unhelpful thoughts and behaviours,” she says.

A screenshot from the Eating Disorder Electronic Single-Session Intervention chatbot developed by Professor Gemma Sharp. Credit: Courtesy of Professor Gemma Sharp

Sharp’s team recorded participants’ experiences at four points – before using the chatbot, after 72 hours, after one month and after three months.

The results of the randomised controlled trial were published late last month in the Journal of Medical Internet Research. Almost 100 per cent of patients who used the chatbot went on to start treatment, and had significantly improved outcomes like reduced eating disorder symptoms, depression and anxiety at one and three months.

The trial’s 12-month results, which are yet to be published, followed a similar trajectory according to Sharp.

For Kelly, her experience with ED ESSI almost two years ago, helped her feel seen for the first time, and set her on the path to recovery.

“It was me almost like talking to myself and figuring stuff out by myself, but with some tools, and there wasn’t any judgment,” she says.

The chatbot also helped lead her to an autism diagnosis, which she says is connected to some of her sensory food issues and ARFID diagnosis. When she was finally able to see her clinician, they had all the information they needed to treat her.

“To be finally validated, I think that was the biggest thing for me ... it was being able to see my own words reflected back to me, and then to go into a session and actually be heard and seen for who I am,” she says.

Professor Gemma Sharp, founding director of The Consortium for Research in Eating Disorders, led the development of the chatbot,Credit: Ben Searcy

While the study was relatively small, it joins a growing body of international research on the potential uses of artificial intelligence in supporting eating disorder diagnosis, treatment and recovery.

Dr Sarah Barakat, a postdoctoral research associate at the InsideOut Institute for Eating Disorders, who was not involved in the study, says Sharp’s findings are in line with her own research into digital interventions.

“There is a real role for these technologies, and they can be really helpful, particularly in a space like eating disorders where there’s quite a lot of shame and ambivalence around seeking help,” she says.

While she emphasises that chatbots like ED ESSI needed to be integrated with human-led care, such technologies can help eliminate certain biases clinicians may hold in the pre-screening stage.

“It is a problem in the healthcare system that clinicians don’t receive enough training around how to screen and treat eating disorders.

“And that means that particularly for people who struggle with binge-eating disorders, who might be in normal or larger bodies, thinking to screen those people for an eating disorder is quite uncommon.”

Then there are ethical concerns that can arise from using AI in healthcare.

In 2023, a chatbot promoted by the United States’ National Eating Disorders Association sparked concern about the use of AI when it offered dieting advice to a patient.

Sharp says these ethical concerns are why her team decided to go with a rule-based chatbot, which meant it could only offer recommendations it had been trained on, rather than generate its own responses.

Sharp and her team are currently testing the use of generative AI, and are rolling out the chatbot across Australia and internationally by year’s end.

Where to learn more

• ED ESSI is being rolled out in clinics across Australia. If you are interested in using it, or for more information, email [e-mail].
• The InsideOut eClinic offers free, evidence-based online self-help programs for individuals aged 16 and over who are experiencing symptoms of an eating disorder.

Butterfly National Helpline, 1800 33 4673.

Make the most of your health, relationships, fitness and nutrition with our Live Well newsletter. Get it in your inbox every Monday.