Communication challenges between autistic and neurotypical individuals stem from a mutual lack of understanding of each other's distinct, and often contrasting, communication styles. Yet, autistic individuals are often expected to adapt to neurotypical norms, making interactions inauthentic and mentally exhausting for them. To redress this imbalance, we propose the design of communication technologies that leverage generative artificial intelligence (AI) to facilitate adaptation by both, autistic and non-autistic, conversational partners. First, we present the design and evaluation of NeuroBridge 50, an interactive platform designed to help neurotypical individuals better understand autistic forms of expression and reflect on how their own behavior shapes cross-neurotype interactions. NeuroBridge utilizes large language models (LLMs) to simulate a) an AI character configured to be direct and literal, a style common among many autistic individuals, and b) four cross-neurotype communication scenarios in a feedback-driven conversation between the character and a neurotypical user. Informed by prior work and vetted by an advisory board of autistic individuals, these scenarios reflect common communication challenges faced by autistic individuals. In a user study with 12 neurotypical participants, we find that NeuroBridge improved their understanding of how autistic people may interpret language differently, with all describing autism as a social difference that "needs understanding by others" after completing the simulation. Second, we present the design and evaluation of TwIPS 2, an LLM-assisted texting interface designed specifically for autistic users. TwIPS provides three core features: Interpret, which explains tone and ambiguity in incoming messages; Preview, which forecasts how one's message may be received; and Suggest, which offers alternative phrasings while preserving user intent. In an in-lab study with 8 autistic participants, we find that TwIPS supported clearer expression and interpretation, provided a preferable alternative to tone indicators, and anecdotal evidence indicating that it reduced the cognitive burden associated with masking. Finally, we outline directions to unify the principles established in recent work to develop communication support tools that can assist multiple conversational partners within the same cross-neurotype interaction.
Rukhshan Haroon (Thu,) studied this question.