HiLSVA: Design and Evaluation of a Human-in-the-Loop Agentic System for Scientific Visualization

2026-06-25 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, quick

Summary

HiLSVA is a human-in-the-loop agentic system designed for scientific visualization (SciVis) that prioritizes human analytical control over full autonomy. It features a plan-first multi-agent architecture, explicit human oversight, stepwise provenance tracking, and learn-at-test-time adaptation from user feedback. The system enables fluid handoff between humans and agents through natural language and direct manipulation, with sandboxed execution ensuring safe, reproducible workflows. Evaluation through case studies and a controlled user study with twelve participants demonstrated that mixed-initiative interaction improves task completion, user control, and workflow transparency across varying expertise levels. Findings also revealed a tradeoff between execution efficiency and human oversight, underscoring the importance of human-centered design in agentic SciVis.

Key takeaway

For AI Engineers designing agentic scientific visualization systems, prioritize human oversight and mixed-initiative interaction. Your systems should augment, not replace, human analytical reasoning, even if it introduces a tradeoff with execution efficiency. Integrating features like stepwise provenance tracking, learn-at-test-time adaptation, and fluid handoff via natural language and direct manipulation will significantly improve user control and workflow transparency.

Key insights

Human-in-the-loop agentic systems enhance scientific visualization by prioritizing human control and collaboration over full autonomy.

Principles

• Mixed-initiative interaction improves SciVis task completion and transparency.
• Human-centered design is crucial for agentic SciVis systems.
• A tradeoff exists between execution efficiency and human oversight.

Method

HiLSVA integrates a plan-first multi-agent architecture with explicit human oversight, stepwise provenance tracking, and learn-at-test-time adaptation. It supports fluid handoff via natural language and direct manipulation, with sandboxed execution.

In practice

• Implement plan-first multi-agent architectures for SciVis.
• Integrate direct manipulation with natural language control.
• Provide sandboxed execution for reproducibility.

Topics

• Agentic Systems
• Scientific Visualization
• Human-in-the-Loop AI
• Mixed-Initiative Interaction
• LLM Agents
• Human-Centered Design

Best for: AI Scientist, AI Engineer, Research Scientist

Related on AIssential

• SciVisAgentSkills: Design and Evaluation of Agent Skills for Scientific Data Analysis and Visualization — Agent skills augment general coding agents with domain-specific procedural knowledge for scientific data analysis and visualization.
• Building Human-In-The-Loop Agentic Workflows — Human-in-the-loop design is critical for LLM agents to mitigate errors and ensure quality in subjective tasks.
• Anthropic Lead: HTML Increasingly Better Than Markdown at Keeping Humans Engaged in Agentic Loops — HTML's visual and interactive capabilities enhance human engagement and productivity in complex AI agent workflows.
• Agentic Authoring of Interactive Multiview Visualizations in Genomics — Agentic LLM iteration significantly enhances genomics visualization quality, but complex agent architectures offer no further benefit.
• Plausible but Wrong: A case study on Agentic Failures in Astrophysical Workflows — Agentic AI systems in scientific workflows often fail silently by producing plausible but incorrect results.
• AI agents are ‘aeroplanes for the mind’: five ways to ensure that scientists are responsible pilots — AI agents can accelerate scientific discovery, but require human oversight, specialization, and engineered trust to be effective.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.