Human-Centered AI: Towards Accessible and Customizable Language Models
The main objective of the HumanAI project is to develop accessible, customizable, and understandable multimodal generative AI tools for individuals with cognitive impairments. The aim is to promote their active participation in content creation, ensuring that generative AI is inclusive and usable by all.
The project is structured into two complementary subprojects:
- HumanAI-LANG (led by HULAT, UC3M): Focused on adapting large language models (LLMs), especially generative models, to the specific needs and preferences of users with cognitive impairments.
- HumanAI-UI (led by NIL): Centered on designing inclusive user interfaces for language-based AI systems, with special attention to cognitive accessibility and user experience.
You can visit the HumanAI project website for more information.
Context and Challenges
The rapid advancement of large language models (LLMs), powered by sophisticated neural networks, has led to the rise of human-like conversational systems, such as chatbots and voice assistants. Major tech companies—including OpenAI, Google, and Meta—are developing tools capable of generating text, images, and even code in response to user prompts. These systems are being widely adopted by both the general public and professionals for a variety of tasks.
However, this progress has not been equitable. People with cognitive impairments continue to face significant barriers when using generative AI. Interfaces are often unintuitive, with complex menus, ambiguous icons, and limited customization. The language used is frequently abstract and difficult to understand. This creates challenges in formulating prompts, interpreting outputs, navigating systems, and obtaining meaningful results. As a result, the full potential of generative AI remains out of reach for many.
Beyond accessibility concerns, generative AI also raises critical issues around trustworthiness (e.g., hallucinations and misinformation), biases, ethical considerations (such as copyright and data privacy), and environmental sustainability. The resource demands for training and running large models are substantial, contributing to energy consumption and carbon emissions. Addressing these concerns is essential for the responsible deployment of generative AI technologies.
Our Approach
Recent advancements in Natural Language Processing (NLP), Natural Language Generation (NLG), text simplification, and user-centered design have opened up new possibilities for inclusive AI development. HumanAI builds on this progress to create generative AI tools that are not only powerful but also cognitively accessible, environmentally responsible, and culturally relevant, particularly for Spanish-speaking populations.
The project will:
- Develop new datasets in collaboration with users, ensuring that the resulting AI models are aligned with their needs and preferences.
- Adapt LLMs to generate plain language and easy-to-read content, following cognitive accessibility guidelines.
- Explore visual modalities, such as symbols, icons, and graphic elements, to enhance understanding and communication.
- Design user interfaces with high customization, usability, and support for diverse cognitive profiles.
- Create methods for accessible prompt generation in Spanish, helping users interact effectively with AI systems.
- Optimize the use of existing models to reduce environmental impact, prioritizing ecological efficiency throughout the project.
Inclusive by Design, Beneficial for All
While HumanAI is focused on individuals with cognitive impairments, its impact will extend much further. Designing with cognitive accessibility in mind results in technology that is more intuitive, flexible, and helpful for everyone. Cognitive challenges can affect anyone, whether due to stress, distraction, unfamiliar languages, or temporary conditions. By embracing inclusive design principles, we can make generative AI more usable, ethical, and sustainable across all contexts.
Join Us
If you’re interested in learning more about our research, exploring our tools, or collaborating to build a more inclusive AI future, we invite you to visit the HumanAI project website.
Let’s shape the future of accessible and ethical generative AI—together.