Innovative AI Chatbot Supports Social Skills Development for Individuals with Autism

A new AI-based social coaching tool is making strides in aiding individuals on the autism spectrum to enhance their social communication skills. The tool, named Noora, is a sophisticated chatbot powered by large language models (LLMs) that interacts with users through text or speech, offering personalized guidance in various social scenarios.

The challenge of social interaction is a common barrier for many with autism spectrum disorder (ASD), particularly in everyday situations like responding to colleagues' conversations or expressing empathy. Difficulties in such interactions can have a negative impact on social relationships and can exacerbate co-occurring conditions such as depression. While traditional face-to-face social skills training involving professionals has shown to be effective, these interventions tend to be costly and limited in accessibility.

Researchers at Stanford University, including clinical professor Lynn Koegel and computer science professor Monica Lam, are exploring how AI can address this gap. Their recent study published in the Journal of Autism and Developmental Disorders examines the application of AI in improving social behaviors among people with ASD.

Koegel emphasized the importance of making interventions more accessible, stating, "Our aim was to translate face-to-face social skills training into a computer-based format because many individuals lack access to in-person resources, which can be expensive." The team developed Noora to simulate real-life social exchanges, allowing users to practice skills like asking questions, giving compliments, and responding empathetically.

A key feature of Noora is its empathy module. The chatbot prompts users with statements, such as expressing tiredness or difficulty concentrating, and asks them to classify the sentiment and respond empathetically. If the response is appropriate, Noora validates it; if not, it provides constructive feedback. The prompts are carefully crafted using both human responses and LLM-generated content, ensuring appropriateness and relevance.

The team conducted a randomized trial with 30 participants over four weeks. Half of them used Noora, completing ten interaction trials daily, while the others received no intervention. The goal was to assess if Noora could foster improvements in empathetic responses that would transfer to real-world human interactions.

Results showed that 71% of users exhibited increased empathetic responses after the program's completion. Moreover, participants' ability to display empathy in live Zoom interviews improved significantly, with the intervention group showing an average increase of 38% compared to the control group. This indicates that a short period of engaging with the AI chatbot can lead to meaningful enhancements in social interactions.

Following this success, the researchers plan to expand Noora's functionalities and make it available for beta testing in both clinical and public settings. Besides social skills, Koegel is exploring AI applications in other areas of ASD treatments, including motivational strategies for young children developing communication skills.

This innovative approach demonstrates how AI technology has the potential to deliver scalable, effective social skills training for individuals with autism, breaking down barriers of cost and accessibility while providing personalized support.