On November 5th, it took place at the Faculty of Computer Science of the UCM an exploratory study to investigate the barriers faced by individuals with Intellectual Disabilities (ID) when interacting with Generative Artificial Intelligence (GAI) tools, using ChatGPT as a reference.
At HumanAI, we are strong advocates for User-Centered Design (UCD). To build truly inclusive technologies, it is essential to incorporate end-users from the initial phases. This study allows us to understand the barriers in situ, thereby enabling us to establish the requirements for our future Accessible GAI.
The 16 participants were students from the ACCEDE program at UCM (https://ucm.es/programa-accede), a university-specific certificate that trains its students (individuals with ID) as “Auxiliary Technicians in Inclusive Environment Evaluation”. The program is part of the UNIDIVERSIDAD programs funded by the ONCE Foundation and the European Social Fund and seeks to promote their socio-labor inclusion. It is crucial to highlight that this activity was intentionally integrated into their academic curriculum, ensuring that the research not only serves us but also offers a tangible benefit and a relevant learning experience for them.
The session was structured in two phases:
- Formative Phase: An introductory session on AI/GAI concepts and basic strategies for responsible use and effective question formulation.
- Practical Phase: Students performed three specific academic tasks (data research, conceptual classification, and formal content generation) to cover distinct interaction purposes.
Pending an exhaustive analysis, our first impressions are very clear and point to the need for action:
- The cognitive difficulties of the target group affect both question formulation and the understanding and verification of answers.
- The necessity for adapted interfaces is evident.
- A significant barrier: the students’ inability or lack of initiative to ask ChatGPT to adapt the information to their specific needs (e.g., simpler language).
The results of this study will be vital for guiding future research towards an inclusive design of GAI.
