“I Don't Trust it, but I Use it”: Navigating Trust, Privacy, and Identity in Disabled People’s Use of Generative AI

Paper Title

“I Don't Trust it, but I Use it”: Navigating Trust, Privacy, and Identity in Disabled People’s Use of Generative AI

Publication Info

• Topic area: Accessibility and intersectional identity in Generative AI use by disabled people.
• Keywords: Generative AI, accessibility, disability, trust, privacy, intersectionality, identity, ethical AI, accessibility tax.

Background and Problem

• Problem / challenge: Existing research on Generative AI (GenAI) and accessibility often takes a narrow, single-axis approach, overlooking how trust, privacy, and intersecting identities shape disabled people’s experiences. There is limited understanding of how disabled people navigate GenAI in their daily lives, particularly across diverse disability groups.
• Significance: GenAI offers significant accessibility opportunities for disabled people but also introduces risks such as bias, privacy concerns, and ethical dilemmas. Understanding these dynamics is crucial to designing inclusive and trustworthy GenAI tools.
• Motivation and related work: Prior studies have explored GenAI’s accessibility potential and its biases along single identity dimensions (e.g., race, gender, disability). However, these studies often lack generalizability due to small samples, limited disability representation, and insufficient focus on intersectional identities. This paper aims to address these gaps.

Solution

• Proposed approach: The study investigates how disabled people navigate GenAI in their daily lives through seven cross-disability focus groups (N=20), focusing on accessibility, trust, and identity-based challenges and benefits.
• Novelty:
  1. Examines GenAI use across diverse disabilities and intersecting identities.
  2. Explores the concept of “accessibility tax” in GenAI use.
  3. Highlights identity-based benefits and harms in GenAI interactions.
  4. Provides design implications for trustworthy and inclusive GenAI tools.
• Procedure and key techniques:
  • Conducted seven semi-structured focus groups with 20 participants representing diverse disabilities, ages, and intersectional identities.
  • Used reflexive thematic analysis and descriptive statistics to analyze focus group data.
  • Explored themes such as trust, privacy, identity preservation, and accessibility taxes.

Results

• Concrete findings:
  • GenAI supports autonomy, efficiency, and communication but introduces accessibility taxes (e.g., financial costs, cognitive labor, privacy risks).
  • Trust in GenAI is conditional, often shaped by necessity rather than confidence in reliability or ethics.
  • GenAI tools both validate and misrepresent intersecting identities, with issues like racial and linguistic bias and normative defaults.
• Advantage over baselines: Provides a nuanced understanding of how disabled people navigate GenAI, addressing gaps in prior single-axis studies by incorporating intersectional perspectives and diverse disability experiences.
• Experiments / evaluation:
  • Focus groups included participants with disabilities such as blindness, neurodiversity, chronic illness, and hearing impairments.
  • Participants used tools like ChatGPT, BeMyAI, and DALL-E for tasks ranging from accessibility assistance to creative and professional support.
  • Data analysis revealed themes of trust, identity negotiation, and accessibility taxes.
• Limitations and future work:
  • Limited to U.S.-centric perspectives; global experiences remain unexplored.
  • Some disability, racial, and gender groups were underrepresented.
  • Resistance to GenAI was understudied due to a small number of non-users.
  • Future work should examine global contexts, additional identity dimensions, and reasons for GenAI resistance.

Summary

This study investigates how disabled people navigate Generative AI (GenAI) in their daily lives, focusing on accessibility, trust, and identity-based challenges. Through seven focus groups (N=20), the authors reveal that GenAI supports autonomy and efficiency but introduces accessibility taxes and ethical dilemmas. Trust in GenAI is often conditional, shaped by necessity rather than confidence in reliability or ethics. The study highlights identity-based benefits and harms, including validation of intersecting identities and misrepresentation through bias. These findings provide actionable insights for designing inclusive, trustworthy GenAI tools and call for further research on accessibility and intersectionality in AI.