AI for Autistic User Independence, Wellbeing, and Self-Regulation

Investigating how Large Language Models and AI systems can be designed to support the unique needs of autistic individuals.

Background

Currently, the US workforce integration of neurodivergent talent remains a societal challenge. This population, specifically individuals with autism spectrum conditions and ADHD, often faces neurodevelopmental constraints related to executive function (EF) and self-regulation (SR). While intellectually capable, these individuals may experience deficits in working memory, set-shifting, and inhibitory control.

This vulnerability peaks during the “Services Cliff” (ages 15–26), where the cessation of mandated transition planning often leads to high postsecondary disconnection. My work explores how AI can bridge this gap by providing autonomous, real-time support in technical workplace environments.

Research Aim & Methods

To address the neurodivergent unemployment crisis, this project focuses on mapping workplace ambiguity and points of friction through a multi-stakeholder qualitative approach:

• Stakeholder Interviews: Engaging neurodivergent employees and supervisors to identify where sociotechnical dynamics break down.
• Speculative Design: Utilizing Generative AI tools during sessions to co-ideate future technology supports with participants.
• Cognitive Scaffolding: Analyzing how LLMs help users offload executive dysfunction and regulate emotions.

Initial Findings

In my study on autistic users’ engagement with ChatGPT, we applied the Technology Affordance framework to analyze LLMs as cognitive scaffolding. While these tools provide significant support, they also introduce risks:

1. Reinforcing Delusional Thinking: Bots may inadvertently validate non-factual beliefs.
2. Automated Masking: By outsourcing communication to conform to neurotypical norms, users risk erasing their authentic identities.