Research Profile

Short Bio

Dr. Kiran Schmidt-Hus is a Senior Researcher at the European Institute for Sociotechnical AI Research in Brussels, Belgium. Their interdisciplinary research connects data science, process science, quantum mechanics, marketing, and psychology. In their work, Dr. Kiran Schmidt Hus contributes to the deployment of responsible companions and commercial agents and thereby reforms in a wide range of customer-facing applications. In their bi-weekly talk series “CALLpanion”, Kiran Schmidt-Hus facilitates a multistakeholder dialogue between leading academic, industry, and non-governmental researchers. Their work has been awarded with “AccelerateEurope2024” by the Belgium Ministry of Digital Affairs. Open PhD Positions: Join us in the heart of Europe - If you are a Master's graduate interested in combining business optimization and AI Companions for mental health applications, please reach out!

Research Interests

• Responsible AI and governance
• Evaluation of LLM-based agents
• Bias and safety in multi-turn interactions
• Process auditing for AI systems

Short CV

• Current (since 2023): European Institute for Sociotechnical AI Research, Brussels, Belgium
• Senior Researcher at research group “Responsible Companions and Commercial Agents”
• PhD in Data Science (focus Process Mining), Vrije Universiteit Leuven, Belgium

Affiliations

• Embedded Institute, Grauhausen, Norvia
• Center for Human-Centered Autonomy, Crossvale, Estoria

Education

• 2023: PhD in Data Science (focus Process Mining), Vrije Universiteit Leuven, Belgium
• 2018: Master of Science: AI and Quantum Science, American Technical University Budapest (ATU)
• 2012-2016: Bachelor of Applied Sciences in eCommerce, Karl-Marx University Trier (KMU), Germany
• Diploma, Financial Informatics, Handelsakademie Lünefels, 2014

Teaching

• Responsible AI: Methods and Governance
• Auditing LLM-Based Systems
• Human–AI Interaction for Services

Awards

• Open Futures Prize, Council for Digital Trust (Norvia), 2024
• Early Career Award, League for Accountable Machine Intelligence, 2023

Publications

• Schmidt-Hus, K., Holistic Bias Cue Quantification (HoBiCuQu): Contextual Interaction Bias Mapping for Multi-Turn Agents and AI Companions, Journal of Responsible Algorithms, 2025.
• Schmidt-Hus, K.; Marek, L., Process Tracing for LLM Governance, Proceedings of the Civic AI Systems Forum, 2024.
• Schmidt-Hus, K.; Nwoko, A.; Ramos, I., Benchmarking Safety Interventions in Coaching Chatbots, Transactions on Human-Centered Automation, 2023.
• Schmidt-Hus, K., From Metrics to Practices: Operationalizing AI Accountability, Norvia Policy Tech Review, 2022.

Abstract

The growing proliferation of AI Companions to commercial and healthcare applications raises critical concerns over their biases and sociotechnical safety. Mental health and coaching LLM applications demonstrated highly positive patient ratings and white collar workers increasingly rely on coaching LLMs in their work. Despite the fast adoption, biases and sociotechnical questions remain unexplored. In a large simulation of synthetic user chats (N=13,789) in multi-turn settings, we holistically mitigate bias in AI Companions and LLM Agents. Drawing on Scheerberg’s Quantum Process Quantification Theory (QPQT), we mitigate bias cues in LLM agent’s interactions. Unlike prior work on single-turn QA, we introduce Holistic Bias Cue Quantification (HoBiCuQu). We find that a majority of SOTA LLM Agents (71.3%) reveal biased cues on the first turn. Notably, applying HoBiCuQu, our multi-turn evaluation with Scheerberg’s Quantum Process Quantification Theory, mitigates biased behavioral cues in close to all (64.8%) Agent chat interactions. Our user study with mid-size partner companies (N=12) demonstrated high user satisfaction and mental well-being though debiasing. Applying Holistic Bias Cue Quantification provides a promising tool for AI developers, researchers, entrepreneurs, and policy makers.