Orca: Browsing at Scale Through User-Driven and AI-Facilitated Orchestration Across Malleable Webpages

Paper Title

Orca: Browsing at Scale Through User-Driven and AI-Facilitated Orchestration Across Malleable Webpages

Publication Info

• Topic area: Enhancing web browsing for large-scale information tasks with AI and user-driven orchestration.
• Keywords: web browsing, information foraging, AI facilitation, malleable interfaces, spatial canvas, multi-agent systems, user-driven orchestration, sensemaking, automation, parallel workflows.

Background and Problem

• Problem / challenge: Traditional tabbed browsers are inadequate for managing and synthesizing large volumes of information across multiple webpages. Fully automated AI systems reduce user agency and hinder contextual understanding.
• Significance: Addressing these limitations is critical for improving workflows in information foraging, sensemaking, and decision-making tasks that span multiple webpages.
• Motivation and related work: Prior work on AI-enabled browsers and malleable interfaces has explored query assistance, visualizing relationships among webpages, and automation. However, these approaches either focus on isolated tasks or lack user control, leaving gaps in supporting scalable, user-driven browsing.

Solution

• Proposed approach: Orca, a prototype browser that treats webpages as "malleable materials" within a "malleable space," enabling user-driven and AI-facilitated orchestration across webpages.
• Novelty:
  1. Introduces a spatial canvas interface for parallel viewing, organizing, and synthesizing webpages.
  2. Integrates AI to facilitate exploration, extraction, and automation while preserving user control.
  3. Supports dynamic workflows with features like batch operations, contextual expansion, and real-time multi-agent systems.
  4. Enables iterative sensemaking with dynamic summaries and customizable layouts.
• Procedure and key techniques:
  • Reconceptualizes webpages as flexible elements that can be arranged, grouped, and transformed.
  • Employs AI for tasks like batch opening links, extracting key content, and synthesizing summaries.
  • Implements a Web Canvas for spatial organization and a global command bar for feedforward prompting.
  • Leverages parallel web automation agents for scalable operations across multiple pages.

Results

• Concrete findings:
  • Participants reported increased exploration and faster navigation, with an average of 37 webpages managed during the study.
  • Page Extraction and batch operations were highlighted as the most helpful features.
  • Users felt more in control of information sources and trusted the results more compared to AI-driven search engines.
• Advantage over baselines:
  • Orca reduces cognitive and manual effort compared to traditional tabbed browsing and AI-driven tools by enabling parallel workflows and preserving user agency.
  • Encourages broader exploration and deeper sensemaking by reducing the overhead of managing context across pages.
• Experiments / evaluation:
  • Conducted a lab study with 8 participants (aged 21–27) using Orca for 60 minutes.
  • Tasks included guided walkthroughs, freeform exploration, and questionnaires.
  • Participants used features like spatial layout, batch operations, and dynamic summaries to complete diverse information tasks.
• Limitations and future work:
  • Short study duration limited participants' ability to fully adapt to the new interface.
  • Challenges with monitoring off-canvas agents and managing visual clutter in dense layouts.
  • Future work includes improving viewport management, exploring alternative layouts, and enhancing agent monitoring and control mechanisms.

Summary

Orca reimagines web browsing by introducing a spatial canvas and AI-facilitated features to support user-driven, scalable information tasks. It enables parallel viewing, organizing, and synthesizing of webpages, reducing the cognitive and manual effort of traditional tabbed browsing. Preliminary evaluation shows that Orca stimulates broader exploration, enhances user control, and fosters trust in results. Future research will focus on refining interface organization, improving agent monitoring, and exploring long-term usability in real-world settings.