Paper Session 1: AI, Expertise and Changing Practice
Date: Monday 2 Nov 2026
Time: 11:00–12:00
Add this session to your calendar, or click a talk’s time to add just that talk.
| Time | Presentation | Abstract | Authors | Contribution type |
|---|---|---|---|---|
| 11:00-11:10 | Between Thought and Expression: How Generative AI and Brain Data Shape the Future of Creative Practice | This paper investigates an exploratory approach to EEG-derived Generative AI (GenAI) co-creation, focusing on how EEG-derived signal variation can be used as low-level modulation input for generative visual output. We performed three studies: Study 1 used co-design workshops to elicit expectations and concerns among early-career creative practitioners. Study 2 used expert interviews to contextualise GenAI within established creative practice. Study 3 implemented and evaluated an EEG-modulated prototype generative system in a pilot study. The first two studies informed design requirements; only the third study evaluated the prototype interaction. Our findings from three studies reveal variations in participants’ experiences but show that shared creative perspectives can strengthen the robustness of AI-driven creative experiences through a Shared Neural Streams concept. We identify design opportunities and methodological constraints for using EEG-derived signals transparently within GenAI-supported co-creative interaction. | Pham, Holmquist | Paper |
| 11:10-11:20 | Craft Knowledge and Embodied Labelling in ML-Assisted Motion Capture | The StyleCap project developed a Machine Learning (ML) model for styling full body animation from sparse motion capture and text. The development contributes an approach to Artificial Intelligence (AI) in HCI research prioritising subjective expertise and embodied knowledge. Interviews with animators documented their values, expertise and use of language in animation. The interviews included a live motion capture session and were coded to produce a database of words describing style. Inverting the traditional process of using language to label data, data was created by expressing language through bodily movement. An ML model was trained on this dataset to produce stylised animations which were evaluated by professional animators. These evaluations were then used to measure the effectiveness of GPT 4.1 as a translator of the evaluators’ language to the specific labels utilised in training. The accuracy of these translations of the craft knowledge and vocabulary used by individual practitioners and enacted through embodied performance demonstrates the potential of human-centred approaches to HCI research with AI. | Koterwas, Morrow, Braghetto, Vaughan | Paper |
| 11:20-11:30 | Beyond the Dichotomy of Tool vs. Teammate: Functional, Normative, and Experiential Perspectives on AI in Organizations | The dichotomous debate coined “Team member vs. Tool” addresses the question of whether artificial intelligence (AI) is, can, or should be a team member or a tool in an organizational context. In this paper, we challenge the assumption that the ”Team member vs. Tool” positions are in tension at all and posit that apparent discrepancies between these two positions mainly arise from a failure to distinguish between different analytical levels: functional descriptions of what AI does, normative claims about what AI should be, and social-psychological accounts of how people experience and interpret AI at work. We disentangle these three perspectives and their respective contributions to the conceptualization of AI in the work and organizational context. Building on this distinction, we further demonstrate that the relations between these perspectives are interdependent but asymmetrical. Functional properties of AI systems more directly shape social–psychological experiences, which in turn inform organizational practices, while normative perspectives primarily influence both functional design and interpretive frames through processes of translation and implementation. We argue that many tensions in the literature stem from overlooking these uneven pathways of influence. | Ontrup, Alizadeh, Berretta, Georganta, Kerstan, Rieth, Tausch, Tschopp, Ulfert, Wischnewski | Paper |
| 11:30-11:40 | Beyond AI Adoption: Trust as a Task-Level Appraisal in Academic LLM Use | Large language models (LLMs) have been positioned as tools that could reduce workload and support productivity in academic work, yet adoption remains selective. Prior research attributes this to differences in trust but offers limited insight into how trust is enacted in practice. We report a qualitative study with 18 UK academics, examining how they decide when and how to trust LLMs in their work. Trust is not a stable attitude but a situated, task-level appraisal, governed primarily by whether outputs can be verified and at what cost. We identify verification burden to describe how, in academic knowledge work, the effort required to validate LLM outputs can exceed the benefit of using them, making restraint the reasonable choice. LLMs also serve as emotional infrastructure for managing the psychological demands of academic work, while creating tensions with academic identity. At the institutional level, we identify institutional paralysis: unclear policies, absent norms, and a lack of endorsed tools constrain adoption even among willing users. Together, these reframe cautious LLM use as a nuanced interplay of constrained trust calibration, emotional ambivalence, and institutional paralysis. We translate these findings into design concepts that make verification effort legible and treat emotional support as core function, as well as describing implications for institutional policy. | Chang, Mella, Dowthwaite, Ahmetoglu, Cox | Paper |
| 11:40-11:50 | Appropriate Automation in AI-Assisted Usability Heuristic Generation: Comparing Human-AI Collaborative Workflows | Domain-specific usability heuristics can better capture context-sensitive usability issues than general heuristics, but their development and refinement remain resource-intensive. This study examined UX practitioners’ experiences of three human-AI workflows for generating and refining heuristics for automotive smart cockpit interfaces. Thirty user experience (UX) practitioners used an artificial intelligence (AI) tool to generate an initial heuristic set and were assigned to one of three refinement workflows: continuous AI-assisted feedback and refinement (G1), human expert feedback with manual refinement (G2), or human expert feedback with AI-assisted refinement (G3). No significant between-group differences were found in overall workload, trust or technology acceptance. However, G3 participants reported higher mental demand than G2 participants and higher temporal demand than G1 participants. Interview findings indicated that AI supported early ideation and routine structuring, whereas AI-assisted refinement required prompt formulation, output checking and decision validation. Participants valued workflows that preserved expert control, enabled transparent review and retained human final judgement. These findings suggest that AI can support early-stage heuristic generation, while judgement-intensive refinement benefits from human control, task-relevant transparency and expert review. | Li, Hermawati, Maior | Paper |
| 11:50-12:00 | From Maker to Orchestrator: Navigating the UI/UX Designer Role Shift in AI-Augmented Design Practices | Generative AI is reshaping digital design, yet how practitioners integrate these tools into professional workflows remains understudied. Through 13 expert interviews, complemented by contextual inquiry with a subset of participants, we examine how UI/UX designers describe and demonstrate their use of AI prototyping tools during the realisation/implementation-facing stages of the design process. Our findings suggest a shift in reported practice: designers describe moving beyond divergent ideation to using AI for convergent, implementation-facing work. With the emergence of vibe coding, where designers leverage code-aware agents to participate more directly in codebases, participants described their role evolving from visual ”makers” towards technical ”orchestrators” of AI outputs. This shift has implications for the traditional design-engineering divide, pointing towards an emerging hybrid ”product engineer” role capable of contributing functional code. We discuss friction points in this reported workflow and derive implications and guidelines for future AI tools, workflows, and designers’ roles that prioritise inspectability and closer code-based collaboration. | Cheung, Lee | Paper |