Human-Centered Intelligent Systems for Enhanced User Interaction

Abstract

Human-centered intelligent systems that adapt to individual user cognitive states, preferences, and interaction patterns represent the next frontier of
human-computer interaction, enabling personalised, accessible, and cognitively congruent interfaces across professional, educational, and assistive
technology domains. Despite substantial progress in affective computing, adaptive user interfaces, and conversational AI, the integration of real-time
physiological sensing, user model learning, and multimodal adaptive response generation into coherent, privacy-preserving human-centered
intelligent systems remains architecturally fragmented. This study presents HumanAI, a unified human-centered intelligent systems framework
integrating electroencephalography (EEG)-based cognitive load estimation, a reinforcement learning-driven adaptive interface controller, and a
retrieval-augmented generation (RAG) conversational AI module for context-aware interaction support. HumanAI was evaluated across three user
interaction domains: adaptive e-learning interface optimisation (n = 124 university students, 8-week study, KTH Stockholm), cognitive assistance for
professional knowledge workers (n = 48 engineers, 4-week deployment, Politecnico di Bari), and accessible interface adaptation for users with mild
cognitive impairment (n = 31 participants, clinical setting, Bari Polyclinic Hospital). HumanAI achieved learning performance improvement of 23.4%,
task completion time reduction of 31.7%, and accessibility rating improvement of 38.4% over static interface baselines. EEG cognitive load
estimation achieved accuracy of 87.3%. User satisfaction rated 4.4/5.0 across all three deployments. These results demonstrate that integrated
physiological sensing, adaptive control, and conversational AI substantially advance human-centered intelligent system interaction quality.