Mixed-Ability Team Study Reveals 5 AI Adoption Pitfalls

A mixed-ability research team at the 2026 ACM SIGACCESS conference (ASSETS '26) is shining a light on the hidden costs and benefits of generative AI in academia. In a new paper, authors Shalini Madan, Sreelakshmi Surabiyil Bindu, Veronica Pimenova, Ellie Seehorn, and Venkatesh Potluri conducted qualitative interviews with all five members of their own team to understand how generative AI tools (like ChatGPT, Copilot, or similar LLMs) are being adopted in a lab that includes researchers with a range of abilities. Their findings reveal a nuanced landscape: while AI boosts productivity for some, it also introduces a 'disability tax' — extra effort needed to make AI tools accessible and avoid unintended harm. For instance, voice-based interfaces may clash with speech impairments, and automated text generation can strip away markers of disability identity, leading to 'homogenizing identity.' The team also flagged risks around private information disclosure when using cloud-based AI services.

The paper builds on these five themes to offer a set of practical recommendations for balanced AI adoption. Key suggestions include: explicitly managing 'crip time' (allowing extra time for accessibility-related tasks), encouraging self-experimentation with different AI tools, and establishing clear norms for risk disclosure when sharing sensitive data with AI. The authors emphasize that AI should augment — not replace — the agency of disabled team members. As generative AI becomes a staple in research labs worldwide, this work provides a crucial framework for inclusive adoption. For tech leaders and researchers, the takeaway is clear: one-size-fits-all AI workflows can inadvertently alienate mixed-ability teams, but thoughtful design and team-wide agreements can turn generative AI into an empowering tool rather than a burden.