HumanAI at LREC’26 and IUI’26 – HumanAI “human-in-the-loop”

Human Oversight in Generative AI for Plain-Language Content and Accessible User Interfaces

The HumanAI project is advancing a shared research line on accessible generative AI, combining large language models (LLMs), human oversight, standards compliance, and user-centered design for both plain-language content generation and adaptive user interfaces. This line is reflected in two recent papers accepted at international venues: “A Human-in/on-the-Loop Framework for Accessible Text Generation”, accepted at LREC 2026, and “Human Oversight-by-Design for Accessible Generative IUIs”, accepted at the AI CHAOS! workshop at IUI 2026.

The first paper, authored by Lourdes Moreno and Paloma Martínez, focuses on the generation of accessible content in plain language, proposing a Human-in/on-the-Loop framework in which automatic generation is complemented by human participation during the process and structured human supervision afterwards. The approach is designed to ensure that accessibility is assessed not only through automatic metrics, but also through real comprehension, expert review, and alignment with accessibility standards.
LREC 2026: https://lrec2026.info/
Preprint: https://arxiv.org/pdf/2603.18879

The second paper, authored by Blessing Jerry, Lourdes Moreno, and Paloma Martínez, focuses on accessible generative user interfaces, especially interfaces adapted to different disability profiles, user capabilities, and contexts of use. From an HCI perspective, the work argues that human oversight should not be added only at the end of the pipeline, but incorporated from the design and modelling stages of the system itself, by embedding requirements, adaptation rules, validated templates, and traceability mechanisms directly into the architecture. This allows interface generation to remain aligned with standards while supporting risk signalling, human intervention, and continuous supervision.
IUI 2026 / AI CHAOS! workshop: https://iui.acm.org/2026/
Preprint: https://arxiv.org/pdf/2602.13745

Together, both papers reflect a common vision within HumanAI: accessible generative AI should not be understood only as a matter of model performance, but as a broader challenge involving requirements engineering, interaction design, accessibility standards, explainability, and governance. In this view, LLMs are used to generate both content and interfaces, while human oversight is embedded across the lifecycle to make these systems more traceable, understandable, auditable, and responsive to diverse user needs.

Taken together, these contributions reinforce the project’s commitment to developing AI systems that adapt both what is communicated and how it is presented, ensuring that generated outputs are better suited to different users, including people with disabilities and people interacting in sensitive or high-stakes contexts.