Responsible AI in Manufacturing: The Case of Accountability in AI Systems

Abstract

Manufacturing enterprises increasingly integrate Artificial Intelligence (AI) into production processes to enhance efficiency and competitiveness. However, Responsible AI (RAI) practices are often overlooked, leading to potential ethical, legal, and operational risks such as biased decision-making and lack of transparency and trust. Existing maturity models, which evaluate the development stages of systems or practices, primarily address either Industrial AI (IAI) applications or general aspects of RAI. AI sandboxes, as controlled environments for testing and validating AI systems under regulatory supervision, offer a structured approach to assess RAI practices. Yet the frameworks lack a targeted model for determining RAI maturity specifically within manufacturing. To address the gap, this paper introduces a RAI-oriented maturity model adopted to the context of smart manufacturing. The maturity model evaluates AI systems across six key dimensions: Accountability, Fairness, Human-Centricity, Privacy/Security, Sustainability, and Explainability. In particular, this paper delves into the dimension of accountability, exploring its indicators. A structured literature review defines the indicators, and a pairwise comparison determines their relative importance. The model is applied to an industrial case study involving a Large Language Model (LLM)-based chatbot for maintenance in the semiconductor industry. Preliminary results demonstrate the model's practicality and effectiveness, providing actionable insights, i.e. the need for auditing mechanisms, to enhance accountable AI design and deployment in industrial settings.