Tightly coupled requirements
Human-data interaction systems must integrate responsive interaction, multimodal data handling, guidance, visualization, and trustworthy AI feedback.
Abstract
The rapid advancement of AI is transforming human-centered systems, with major implications for human-AI interaction, human-data interaction, and visual analytics. Data analysis increasingly involves large-scale, heterogeneous, and multimodal data that is often unstructured, while foundation models such as LLMs and VLMs introduce new uncertainty into analytical workflows.
The paper examines how these changes expose persistent challenges in latency, scalability, interaction, exploration, reliability, and interpretability. It argues for end-to-end, human-centered systems that co-design data management, AI components, interfaces, and visualization, while keeping humans meaningfully involved through feedback, verification, and trust calibration.
Key insights
Latency and scalability are barriers
Multimodal data, complex interactions, embeddings, streams, and deep learning models all intensify scalability and responsiveness constraints.
End-to-end design is essential
System architecture, data operations, AI models, interfaces, and visualization need to be designed together, not optimized independently.
AI requires human oversight
AI can support guidance, semantic understanding, and usability, but brittleness, errors, unpredictability, and bias make human-in-the-loop oversight indispensable.
Visualization becomes generative
Visualization is shifting from static depiction to progressive, narrative, aesthetics-aware, and adaptive guidance for sensemaking and trust.
Cross-disciplinary work is crucial
The research agenda spans systems, databases, AI, information visualization, HCI, computer graphics, and cognitive science.
Challenges and Opportunities
Latency issues
Interactive systems need real-time response to preserve analytical flow.
Human-centered data management
Query processing, indexing, and storage can be optimized for interactive analytics.
Scalability
Large datasets and high-velocity analysis require scalable data operations.
Guided exploration
AI suggestions can help users discover what can be asked and analyzed.
AI uncertainty
Errors and bias require verification, provenance, and human oversight.
AR/VR interfaces
Virtual and augmented environments can support in-the-world analytics.
Multimodal complexity
Text, images, audio, video, and rich documents require new abstractions.
Narrative visualization
Data storytelling can guide attention and improve interpretation.
BibTeX
@misc{fekete2026humanDataInteractionAI,
title = {Human-Data Interaction, Exploration, and Visualization in the AI Era: Challenges and Opportunities},
author = {Fekete, Jean-Daniel and Hu, Yifan and Moritz, Dominik and Nandi, Arnab and Basu Roy, Senjuti and Wu, Eugene and Bikakis, Nikos and Papastefanatos, George and Chrysanthis, Panos K. and Li, Guoliang and Yu, Lingyun},
year = {2026},
eprint = {2603.05542},
archivePrefix = {arXiv}
}