TLDR: A new study from Stanford HAI reveals that Noora, an AI-powered chatbot, significantly improves empathetic responses and social communication skills in individuals with autism spectrum disorder (ASD). The research, led by Stanford professors Lynn Koegel and Monica Lam, demonstrates that the chatbot’s benefits generalize to real-life interactions, offering a scalable and accessible alternative to traditional, often costly, face-to-face interventions.
Stanford University’s Institute for Human-Centered AI (HAI) has announced groundbreaking research demonstrating the effectiveness of an AI-powered chatbot, Noora, in teaching empathy and social communication skills to individuals with autism spectrum disorder (ASD). This innovative approach addresses the significant challenges many people with ASD face in social interactions, which can negatively impact their personal and professional lives and exacerbate co-occurring conditions like depression.
Traditional face-to-face interventions, while proven effective, are often expensive and not widely accessible. Dr. Lynn Koegel, a clinical professor of psychiatry and behavioral sciences at Stanford, and Professor Monica Lam from Stanford’s Computer Science Department, spearheaded this research, published in the Journal of Autism and Developmental Disorders. Their goal was to translate the benefits of clinical social skills training into a more accessible, computer-based format.
“Our research has shown that face-to-face work does help with social conversation … so we wanted to see if we could translate that into computer use,” stated Koegel. She emphasized the critical need for accessibility, noting, “Accessibility is really important because a lot of people don’t have access to a face-to-face provider and the providers can be really expensive.”
Noora, a chatbot built with a large language model (LLM), provides individualized guidance through one-on-one, written or spoken interactions. It helps users practice various social communication scenarios, including asking questions, giving compliments, and responding empathetically. The recent study specifically focused on Noora’s empathy module.
The chatbot’s empathy module operates by presenting users with a leading statement, such as “I’m feeling really tired lately and it’s been so hard to concentrate,” and then prompts them to assess the statement’s emotional tone (positive, neutral, or negative). Following this, Noora grades the assessment and asks the user to formulate an empathetic response. Based on the user’s reply, Noora provides either a gentle correction or validates a correct response. The research team meticulously crafted prompts and responses, using a pool of 330 verified statements to ensure appropriate and safe interactions, preventing the chatbot from generating inappropriate questions on the fly.
To evaluate Noora’s efficacy, Koegel, Lam, and their colleagues conducted a randomized trial involving 30 participants. Half of the participants used Noora for four weeks, completing 10 trials per day, five days a week, for a total of 200 trials. The other half served as a control group, receiving no intervention.
The primary objective was to determine if Noora could not only improve users’ empathetic responses within the program but also if these improvements would generalize to real-life human-to-human communication. “There’s a lot of AI research out there that shows that the ASD students improve using a program, but doesn’t show that it generalizes to real life,” Koegel explained. “So that was our main goal.”
The results were highly encouraging. Comparing responses from the beginning to the end of the experiment, 71 percent of participants who used Noora showed an improvement in their number of empathetic responses. Furthermore, to assess generalization, participants engaged in Zoom calls with a team member before and after the experiment, which included leading empathic statements. The experimental group demonstrated a significant average increase of 38 percent in empathetic responses, while the control group’s scores remained consistent.
Buoyed by this success, Koegel and Lam plan to test other modules of Noora and are working towards opening the chatbot for beta testing for public and clinical use. Dr. Koegel is also integrating AI into other areas of her autism research, including motivational treatment for children with ASD in the early stages of communication. “I’d like to take a lot of the work that I’ve done over the decades that’s face-to-face and see how much we can translate to AI,” Koegel said, adding, “Since kids really like the computer, we want to see if instead of just spending time on their computer or iPhone we can create a learning experience.”
Also Read:
- Generative AI Fuels $120 Million Revenue Boom for Companion Apps by 2025
- Mount Sinai Study Reveals AI Chatbots Prone to Spreading Medical Misinformation, Highlights Simple Safeguard
Koegel’s research team is actively testing other Noora modules with individuals aged 11-35 and is developing modules for first words for children, inviting additional participants to join their studies.
