As a European and global society, we are challenged to use artificial intelligence for societal, business and individual benefit; to build knowledge around what sustainability means in business and at work, and translate such knowledge into practice. Our research asks:
How can conversational AI help build knowledge about sustainability in business and work, and connect such knowledge to practical tasks?
Based on the fundamental motivation for this question and its examination in the domain of sustainabiltiy in business, work and everyday practice, we ask questions that generalise to overarching questions around generative AI, the value of knowledge, and how humans can best build both knowledge and competent practice in this age of AI.
GPT-generated visual for one of the business cases, created by the Enchatted team.
It clearly feels different to talk than to write.
As voice-based interaction powered by modern Large Language Modles (LLMs) is rapidly increasing in quality every few months (better AI-generated voices, less delays), a fundamental perceived difference between "talking to an AI agent" and "texting with an AI agent" remains. We ask:
How does interaction modality matter?
In an exploratory user study, doctoral student Alina Kopkow and colleagues found that users perceived voice-based interaction as a more natural conversation, and indicated that it allowed them to focus more on the document they were discussing with an AI agent. On the other hand, users reported feeling more in control of the conversation when the agent participated via both voice and text in the conversation (Kopkow et al., 2026 - Balancing Control and Agency @ACM MuC).
Now, Kopkow studies the impact of interaction modality when users discuss a (fictive) business case in terms of sustainability strategies and practices either via text or via voice with an AI agent. As we design agents with the goal to bring sustainability-focused knowledge into business and practice, two questions matter in particular to us:
1) How does interaction modality affect cognitive load and learning?
2) How does interaction modality affect users' perception of the agent, and of their interaction?
When studying AI, it is tempting to focus on the impact of the AI - but really, it is always used in a specific socio-technical context. Therefore, we ask:
How does the socio-technical context shape whether and how AI agents designed to stimulate knowledge-building are effective?
In a user study set in entrepreneurship education, Dr. Regalado and colleagues explored how students from a university's entrepreneurship course developed their sustainability-focused practice ventures both with an educational AI agent (Gaia) and general-purpose genAI agents (e.g., ChatGPT). The study highlights that students live in an AI-infused world, with easy access to a wide range of AI tools. In this world, AI agents designed to "make you think about sustainability" compete with general-purpose agents designed to fulfill their users' wishes. Students have both the freedom and responsibility to orchestrate their AI tool usage: deciding which tools to use, how to use and move between them, and how to combine the outputs (Regalado et al., 2026 - Students' Agency in Educational vs. General-Purpose GenAI Agents @AIFE, best paper nomination).
In settings where users are asked to develop creative, innovative and at the same time well-grounded, feasible and viable ideas for considering sustainability in business, products, and practice; the question then develops into one around knowledge, agency, and creativity. We ask:
How do users' prior knowledge and AI design characteristics shape usage, agency, and creativity?
In ongoing studies, Dr. Regalado investigates practices of generative artificial intelligence usage, knowledge, and creativity in Hackathons focused on developing computer- and AI-based products that foster sustainability.
The responsible core team within the HCI & EdTech research group at the TU Graz' Institute of Human-Centred Computing was:
International collaborators help us shape our research:
Prof. Dr. Aleksandra Kekkonen, Estonian Business School
Prof. Dr. Michael Liut, University of Toronto
Prof. Dr. Nikolaus Obwegeser, Bern University of Applied Sciences
Prof. Dr. Thiemo Wambsganss, Bern University of Applied Sciences
This research has been conducted on the backdrop of and with substantial funding by the L2BGreen project (Erasmus + grant 101139849).
Regalado, Glinik, Kroller, Obwegeser, Wambsganss, Kekkonen & Pammer-Schindler (2026). Students' Agency in Educational vs. General-Purpose GenAI Agents: An Exploratory Study in Entrepreneurship Education. AI for Education Conference, Singapore. Best paper nomination. Publication link upcoming.
Kopkow, Bangerl, Regalado, Liut & Pammer-Schindler (2026). Balancing Control for Agency: An Exploratory Study with a Voice-Based Self-Explanation Agent. Mensch und Computer 2026 - Tagungsband. https://dl.gi.de/items/cf85a128-d7f7-4b05-a341-e4f72912e5b3