ICAT Playdate — Empathy-Driven Interactive Human-AI System for Countering Negative-Self Talk for Autistic Individuals

Generative language models (LMs), such as the recent Generative Pre-trained Transformers (GPT) offer tremendous potential to revolutionize how we support the mental well-being and communication of neurodiverse individuals, including those on the autism spectrum. Autistic people often struggle with communication and social interactions that can lead to higher rates of depression and anxiety. In fact, co-occurring mental health disorders are common across autistic populations. Such individuals tend to engage in negative self-talk (NST) that can exacerbate these mental health issues. Traditional therapy may not be effective for those with limited verbal communication skills. Thus, there is a need for a tool that can help autistic individuals to self-encourage and support their mental wellbeing. To address this, we propose to develop EmotionAIze, an interactive positive affirmation tool powered by GPT-4 that generates contextualized, tone-adaptive empathy-based counter responses to NST expressed as multimodal inputs (text, audio, images, videos, tactile interactions with physical objects) from the user. This short talk will introduce the motivation behind EmotionAIze, and discuss its implications for future research and practical applications that addresses problems of LLM accessibility for neurodiverse individuals.