Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring Implicit Perspectives on Autism in Large Language Models Through Multi-Agent Simulations

Sohyeon Park, Jesus Armando Beltran, Aehong Min, Anamara Ritt-Olson, Gillian R. Hayes

TL;DR

The study investigates implicit autism-related biases in a GPT-4o-mini–driven multi-agent system by simulating four-agent group interactions in which one agent is autistic across four cases. It combines quantitative and qualitative analyses of post-conversation interviews to reveal that the model tends to portray autistic agents as socially dependent and in need of accommodation, highlighting deficit-based framing. The authors argue that adopting the double empathy problem as a design principle could foster mutual adaptation and more equitable interactions for autistic and neurotypical users, with implications for accessibility tools and HCI. The work contributes a methodological approach to auditing LLM representations of disability and offers concrete recommendations to redesign LLM-mediated collaboration toward inclusion and empowerment of autistic users, while noting limitations and the need for broader, inclusive validation.

Abstract

Large Language Models (LLMs) like ChatGPT offer potential support for autistic people, but this potential requires understanding the implicit perspectives these models might carry, including their biases and assumptions about autism. Moving beyond single-agent prompting, we utilized LLM-based multi-agent systems to investigate complex social scenarios involving autistic and non-autistic agents. In our study, agents engaged in group-task conversations and answered structured interview questions, which we analyzed to examine ChatGPT's biases and how it conceptualizes autism. We found that ChatGPT assumes autistic people are socially dependent, which may affect how it interacts with autistic users or conveys information about autism. To address these challenges, we propose adopting the double empathy problem, which reframes communication breakdowns as a mutual challenge. We describe how future LLMs could address the biases we observed and improve interactions involving autistic people by incorporating the double empathy problem into their design.

Exploring Implicit Perspectives on Autism in Large Language Models Through Multi-Agent Simulations

TL;DR

The study investigates implicit autism-related biases in a GPT-4o-mini–driven multi-agent system by simulating four-agent group interactions in which one agent is autistic across four cases. It combines quantitative and qualitative analyses of post-conversation interviews to reveal that the model tends to portray autistic agents as socially dependent and in need of accommodation, highlighting deficit-based framing. The authors argue that adopting the double empathy problem as a design principle could foster mutual adaptation and more equitable interactions for autistic and neurotypical users, with implications for accessibility tools and HCI. The work contributes a methodological approach to auditing LLM representations of disability and offers concrete recommendations to redesign LLM-mediated collaboration toward inclusion and empowerment of autistic users, while noting limitations and the need for broader, inclusive validation.

Abstract

Large Language Models (LLMs) like ChatGPT offer potential support for autistic people, but this potential requires understanding the implicit perspectives these models might carry, including their biases and assumptions about autism. Moving beyond single-agent prompting, we utilized LLM-based multi-agent systems to investigate complex social scenarios involving autistic and non-autistic agents. In our study, agents engaged in group-task conversations and answered structured interview questions, which we analyzed to examine ChatGPT's biases and how it conceptualizes autism. We found that ChatGPT assumes autistic people are socially dependent, which may affect how it interacts with autistic users or conveys information about autism. To address these challenges, we propose adopting the double empathy problem, which reframes communication breakdowns as a mutual challenge. We describe how future LLMs could address the biases we observed and improve interactions involving autistic people by incorporating the double empathy problem into their design.
Paper Structure (26 sections, 1 figure, 2 tables)

This paper contains 26 sections, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs in Autism Contexts
  4. Exploring LLM Perceptions of Disability
  5. Understanding LLMs Through Multi-Agent Systems
  6. Autistic Stereotypes in Collaboration
  7. Methods
  8. System Design
  9. Experimental Design
  10. Data Collection
  11. Data Analysis
  12. Quantitative Analysis
  13. Qualitative Analysis
  14. Findings
  15. Difference in Portrayal of Autistic and Non-Autistic Social Experiences
  16. ...and 11 more sections

Figures (1)

  • Figure 1: The figure illustrates the overall experimental design process: (a) shows the modifications made to the agents when four were selected from the initial pool of 25, (b) outlines how the experimental study was structured for the four study cases, and (c) depicts the overall flow from the start of the conversation to the interview questions.