| To promote sustainability in an era of global shortages, the democratisation of AI through low threshold Machine Learning Operations (MLOps) is a critical necessity. In this paper, the DeKIOps research project addresses the gap between complex Machine Learning (ML) systems and domain experts who lack profound AI skills to operate them. In a prior work, we showed how the energy usage of multi-sensor platforms in an industrial setting can be optimised by providing robust ML models. This research emphasises UI & UX as catalysts for the democratisation of AI. By streamlining interaction with complex ML models, we aim to empower non-experts to operate these systems autonomously and independently. The core contribution is an interdisciplinary framework that translates Guidelines for Human-AI Interaction (HAX) as established by Amershi et al. along with findings from expert interviews into actionable UI & UX principles for MLOps. We evaluated these guidelines through a mixed-methods study using a functional prototype designed for autonomous model management including automated retraining, data augmentation and multilayered explainable visualisations. In this context, we achieved an excellent System Usability Scale (SUS) score of 84.5 (Grade A+). This proves the framework’s capability to enable even laymen to steer complex ML systems, fostering true AI democratisation. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.