Why so serious? Generative AI in Knowledge-Work from a Text-Workers’ Perspective

Generative AI (genAI) systems have become an integral part of scientific text-producing practice. They can conduct literature-related research, organize content and information, proofread text, and assist with writing itself. Above all, they are credited with reducing workload by taking on or taking over these routine tasks. Drawing on an internal, interdisciplinary KIT-ITAS project that deployed the genAI-supported research assistance system 'Undermind.ai', we demonstrate that the potential to reduce workload is offset by the need to reorganize basic scientific competencies in response to the changing workflow requirements of genAI-assisted knowledge work. GenAI systems produce texts that closely resemble stylistic conventions and appear both well-formed and plausible. However, the closer these texts adhere to researchers’ expectations, the more challenging it becomes to verify their content using established academic methods. This generates higher-order efforts that must be weighed against existing working standards for assessing text-based results. The constant validation and ongoing negotiation of plausibility with the system’s output, both during and beyond the writing process, places additional demands on researchers’ critical competence and judgement. Using genAI in knowledge work thus shifts the focus of familiar textual and intellectual labour practices and procedures away from independent creation and toward the verification of textual output, along with the corresponding demands on epistemic attention, responsibility, authorization, and authority.