Independent nonprofit AI research institute
We investigate how AI systems can become usable by, and accountable to, people with disabilities and low-resource communities. CompAccess means computational accessibility.
The CompAccess Institute is an independent nonprofit AI research institute focused on Accessible Intelligence. Our work examines the technical and social conditions required for AI systems to support people who are often absent from benchmark design, product testing, and deployment planning.
Accessibility in AI is not limited to interface compliance. It includes privacy for assistive use, language coverage for communities with limited digital resources, robust interaction for disabled users, and evaluation methods that measure whether systems are genuinely usable in daily life.
We study compact models that can support assistive computing without routing sensitive disability-related contexts to remote services. The work includes model compression, latency constraints, and privacy-preserving deployment.
We investigate modeling methods for languages and communication systems that are poorly represented in large public datasets. This includes sign-language representation, multilingual evaluation, and data governance for small communities.
We examine agents that operate software interfaces on behalf of users with different sensory, motor, and cognitive needs. Research questions include control, confirmation, error recovery, and the legibility of agent actions.
We develop evaluation practices that measure whether AI systems are usable, reliable, and accountable for disabled users and low-resource communities. Benchmark work focuses on tasks, contexts, and harms that current leaderboards often omit.
Accessibility exposes unresolved questions in representation, interaction, evaluation, and institutional accountability.
Assistive AI often handles intimate contexts, making local inference a civil and technical requirement.
Scale alone does not solve language exclusion when the training record is thin, uneven, or governed by others.
Usability requires more than task accuracy. It requires recovery, consent, context, and sustained use.
Benchmarks for agents should include users whose access needs change how software tasks are performed.
Accessible Intelligence refers to AI systems that are designed, evaluated, and governed so that disabled people and low-resource communities can use them effectively and contest their failures.
No. The CompAccess Institute is a research institution. It is not a software product, school-advisory service, relay service, or proxy service.
The institute prioritizes open-access publications, public notes, and reusable evaluation methods when release does not expose private community data or create foreseeable misuse.
The institute is structured as an independent nonprofit research organization. Funding inquiries and institutional disclosures can be directed to the research office.
Prospective collaborators can contact the institute with a concise description of the research question, community context, and expected public-interest contribution.
For collaborations, evaluation proposals, and institutional inquiries, contact the CompAccess research office.