Assistance or Distraction? A Cognitive Ergonomics Perspective on Cognitive Load During AI-Assisted Post-Editing

The integration of generative artificial intelligence (GenAI) into professional workflows is reshaping human-computer interaction (HCI), requiring a deeper understanding of how these systems affect human cognition from a cognitive ergonomics perspective. This study examines AI-assisted post-editing (AIPE), a translation workflow utilizing large language models to provide suggestions and accompanying rationales. Drawing on Cognitive Load Theory, we investigated the impact of AIPE on cognitive load across varying task conditions using a mixed-methods approach of eye-tracking, key-logging, subjective ratings, and retrospective interviews. Results demonstrate a cognitive paradox in AI assistance: although AIPE reduces overall cognitive load—especially in L1–L2 translation and light PE scenarios—it concurrently increases localized cognitive engagement. These findings characterize AIPE as a form of human-AI collaboration that can offload routine diagnostic work while reconfiguring users’ effort at decision-making, and they highlight the importance of task context when designing cognitively ergonomic, human-centered AI support for expert workflows.