Artificial intelligence (AI) has become central to digital innovation, yet many AI initiatives stall before deployment due to unclear problem framing, data-readiness gaps, and cross-role misalignment rather than technical limitations. This study examines how such early-phase challenges emerge in AI-enabled Software-as-a-Service (SaaS) development and explores how service design can support alignment. Five SaaS providers in a national innovation program were studied through semi-structured group interviews and reflexive thematic analysis. The findings identify seven recurring challenges related to data conditions, feasibility negotiation, and semantic divergence across roles, reflecting structural characteristics of AI development. The study further discusses how selected service design principles may help surface assumptions and improve early-phase coordination, contributing empirical insight into a critical yet underexplored stage of AI-enabled system development.
This study explores how junior high school students conceptualize AI across multimodal representations, and how gendered biases emerge in their mental models. A total of 111 ninth-grade students (50 males, 61 females) from a public junior high school in Taiwan participated in a 45-min session involving three tasks: a brief textual description of AI, a hand-drawn representation, and an AI-generated image created via ChatGPT based on their textual input. Using thematic coding and within-participant cross-modal analysis, we investigated how students’ conceptions of AI shifted across representational modalities and how implicit gendered biases became visible through representational translation. Findings show that students’ textual descriptions across genders were largely functional and utility-oriented, reflecting a shared instrumental view of AI. However, subtle gendered biases were evident in emphasis: male students more frequently framed AI in technical and task-oriented terms, whereas female students foregrounded everyday use and social interaction. These latent biases became more pronounced in visual modalities. In both hand-drawn and AI-generated images, female students more often depicted AI as anthropomorphic, interactive, and socially embedded, while male students tended to represent AI through mechanistic, system-oriented, and human–machine relational framings. Cross-modal discrepancies further indicate that bias did not reside as a fixed belief in students’ textual definitions, but emerged through representational translation. Male students’ representations showed greater internal consistency across modalities, whereas female students’ representations more often diverged, introducing social or anthropomorphic features not explicit in text. Notably, AI-generated images functioned as mediated artifacts that amplified culturally familiar and gender-coded visual templates. This study contributes to AI literacy and human–AI interaction research by conceptualizing AI bias as a dynamic, representational, and sociotechnical process, highlighting the value of multimodal analysis for surfacing implicit biases that remain obscured in text-only approaches. Based on these findings, the current study underscores the importance of addressing not only technical understanding but also the socio-emotional and representational dimensions in inclusive AI education and design.
As Generative Artificial Intelligence (GenAI) continues to evolve, we need to revisit the role of AI in the design ecosystem. AI not only serves as a tool but also acts as an agent of human intent. Based on the different stages of the design process, we developed RolePlayAI, a multi-agent role-playing framework, which utilizes AI agents to simulate diverse roles in a human design team. By introducing role-playing and debate mechanisms, we explore the impact of adversarial interactions on creativity and decision-making transparency. Our user study shows that RolePlayAI not only enriches the creative space, but also allows users to better understand AI’s decision-making and enhances their trust in the system compared to stand-alone AI tools. This emerging model of multi-agent collaboration brings important ethical and practical implications for the role of AI in design, providing guidelines for more responsible and transparent AI-augmented creativity.
The rapid deployment of artificial intelligence (AI) across domains has highlighted the critical challenges of ensuring safe, transparent, and effective human-AI integration. Despite the growing body of work examining individual aspects of human-AI interaction, such as trust, usability, and performance, these factors are often studied in isolation, limiting the collective understanding of how human and AI characteristics jointly shape outcomes over time in a team setting. The lack of a comprehensive framework for studying human-AI interactions makes it difficult to understand what factors are of most importance. In order to address this gap, the Human-AI Interaction Framework, grounded in the Inputs-Mediators-Outputs-Inputs (IMOI) process model, is outlined and explained. In this framework, the initial inputs would be organized into AI system characteristics (e.g., the sources and algorithms the AI utilizes, and the observable features related to its information output) and human characteristics (i.e., the individual differences the user brings into the interaction). These inputs are influenced by moderators (i.e., task and situation factors) and mediated by processes and emergent states to influence outputs of the framework, which are classified into human-AI performance and human perceptions factors. In this framework, these outputs feed into the subsequent performance episode’s inputs through feedback: the AI adaptations and human adaptations that commence. Overall, this framework seeks to guide researchers in studying human-AI interactions in a way that provides a structured approach to increase generalizability of findings and support the design of safer, more transparent, and effective human-AI teams.
The study presents key findings to evaluate and ensure the readiness of GenAI tools for use in the professional VFX, film and media industry. Building on previous insights, we present themes which will support rather than hinder artistic practices as well as considerations on ethical guidelines to ensure sustainable and fair use. The study suggest skills to develop in order to incorporate GenAI tools efficiently such as: adaptability of tools and workflow, knowledge in sufficient input quality, prompting, and to be able to apply human ideation and filtering as well as discipline specific competence to the GenAI tool assisted process throughout the whole production workflow. The study suggests that HAIC tools, while being highly cost- and time-effective, are less valuable to incorporate in post-production and final production as they lack sufficient control, high resolution output, nor an option for personalization.
