$1m to develop AI-driven emotional recognition training to reduce ‘social blindness’

MBIE has granted $1.1 million in funding to University of Canterbury researchers to develop a hyper-realistic virtual therapy avatar to help high-functioning people with Autism Spectrum Disorder (ASD) to better recognise emotions and reduce ‘social blindness’.

  • Distinguished Professor Geoff Chase and Lecturer Dr Lui Holder-Pearson

    University of Canterbury Engineering researchers Distinguished Professor Geoff Chase and Lecturer Dr Lui Holder-Pearson have been awarded $1.1 million in funding for their project, ‘AI-driven Two-Way, Feedback Controlled Emotional Recognition Training for Individuals with Autism Spectrum Disorder’.

Sustainable Development Goals 10 - Reduced Inequalities

University of Canterbury Engineering researchers Distinguished Professor Geoff Chase and Lecturer Dr Lui Holder-Pearson have been awarded $1.1 million in funding through a partnership between MBIE and Soul Machines, an AI company in Auckland.

Their project, titled ‘AI-driven Two-Way, Feedback Controlled Emotional Recognition Training for Individuals with Autism Spectrum Disorder’ has been granted $1,105,412.00 (excl. GST). The Canterbury engineering academics will work with European research partners including UC Adjunct Professor Knut Möller, based at Furtwangen University in Villingen-Schwenningen, Germany.

Autism is a neurodevelopmental condition that affects approximately 93,000 New Zealanders. According to the CDC, Autism Spectrum Disorder (ASD) is growing at a rate of 5% to 10% per year due largely to increased diagnosis of high-functioning ASD. It is estimated ASD affects about 1-in-44 children with boys four times more likely to be diagnosed with autism than girls (according to US data). It can play a part in socially and economically debilitating cognitive problems, including significant depression and anxiety co-morbidities.

“Sometimes called ‘social blindness’, the inability to accurately recognise emotions in other people is common and the only therapy is intensive one-to-one or small-group training,” says Dr Holder-Pearson. “This approach is costly, in short supply, and thus often infrequent.”

However, strong growth in high-functioning ASD diagnosis, particularly in boys, threatens to create a “lost generation” unable to achieve their full potential. The University of Canterbury researchers recognised the need to significantly increase access to, and the positive outcomes of, emotional recognition training therapy for high-functioning ASD individuals.

“Our proposed solution is a virtualised, two-way, feedback-controlled emotional recognition training therapy combining AI, clinical therapy, and real-time subject physiological/emotion recognition measurements to virtualise 1-to-1 training,” Professor Chase says.

It combines three key elements:

  • Hyper-realistic and responsive avatars from Soul Machines Digital DNA Studio able to show detailed emotions
  • Computer vision to read subject emotional state, reaction rates in integrated tasks, stress levels (via heartrate, etc.), focus, and attention, incorporating critical subject feedback
  • Programmed standard, accepted therapeutic methods (behind the avatar) to respond to measured subject behaviour/actions

These technologies enable a virtualised true two-way therapeutic session, where current emotion recognition software has no subject feedback (is only one-way).

“Critically, the AI avatars are not just a display, and subject feedback enables a true form of virtualised one-to-one therapy,” Professor Chase says.

According to the UC engineering academics, the software-based solution driven by accepted clinical therapeutic methods dramatically increases access and scalability while lowering other costs. The overall solution creates a highly extensible platform for other therapies.

/Public Release. This material from the originating organization/author(s) may be of a point-in-time nature, edited for clarity, style and length. The views and opinions expressed are those of the author(s).View in full here.