Zur Seitenansicht

Titelaufnahme

Links
Download
Klassifikation
Abstract
  • The integration of Large Language Models (LLMs) into engineering workflows is transforming automation from solely deterministic execution into hybrid intelligence, where humans and AI agents jointly reason and act. Compound LLM systems expand automation possibilities, however there is also a risk of obscuring their decision-making processes, challenging engineers’ ability to validate results. Transparency therefore becomes a prerequisite for trust and accountability in designing such systems for engineering tasks. We explore how transparency can be both embedded in and empirically tested in such AI based expert systems. Drawing on Leonardi’s affordance theory as well as the triangle of design responsibility, an industrial prototype of a compound LLM engineering system is developed and verified. The prototype integrates reasoning and tool agents with transparency mechanisms, enabling engineers to observe actions, data sources, and decision pathways. Based on empirical insights and combined with a literature analysis, a concept for validation and transparency evaluation is derived to assess whether hybrid intelligence systems are not only able to deliver value by solving required tasks but also remain interpretable and accountable. Building upon established verification and validation strategies the concept provides a foundation for the evaluation of transparency as a distinct quality dimension in AI-driven engineering systems.
Statistik
  • Das PDF-Dokument wurde 8 mal heruntergeladen.
Lizenz-/Rechtehinweis