Tag: LLMs

Posted on

Can MLLMs Distinguish Human Laughter?

The 18th International Conference on Social Robotics (ICSR + Art 2026) is currently taking place in London from July 1–4, bringing together researchers, academics, and industry professionals from around the world to explore the latest developments in social robotics. The conference serves as an international platform for exchanging ideas on how intelligent systems can better understand, interact with, and support people in everyday life. On the second day of the conference, Sahan Hatemo, a student at the FHNW School of Computer Science, presented the paper “Reading Between the Laughs: A Human-Referenced Audio Evaluation of MLLMs for Social Robotics”, co-authored with Dr. Katharina Kühne (University of Potsdam) and Prof. Dr. Oliver Bendel (FHNW School of Business). The study investigates whether today’s leading multimodal large language models (MLLMs) can distinguish authentic from non-authentic laughter using audio signals alone. As laughter is an important social cue, the ability to recognize its authenticity could significantly improve how robots and AI systems communicate with people in social settings. The researchers found notable differences in how the evaluated AI models interpreted laughter. OpenAI models showed a clear tendency to classify most laughter as genuine, while Gemini models were generally more skeptical in their assessments. Despite these contrasting biases, several models performed significantly better than chance, with Gemini 2.5 Pro achieving the strongest overall performance. A closer analysis also revealed qualitative differences in the models’ decision-making. Less capable models appeared to rely on superficial acoustic features, such as pitch, and were more likely to classify higher-pitched laughter as less authentic. In contrast, the best-performing model seemed to focus on more sophisticated aspects of voice quality, indicating a deeper understanding of the characteristics that distinguish genuine from non-authentic laughter. The findings demonstrate the growing potential of multimodal AI for social robotics. As robots increasingly become part of everyday environments, the ability to accurately interpret subtle social signals such as laughter could play a crucial role in fostering trust, improving communication, and strengthening human-robot relationships. Further information is available at icsr2026.uk.

Posted on

Chatbots for Endangered Languages

On April 8, 2026, the article “Chatbots for Dead, Endangered, and Extinct Languages: Possibilities and Limitations of Generative AI for Continuing Education” by Oliver Bendel was published in Wiley Industry News. The focus is on how chatbots based on generative AI can contribute to the preservation and promotion of dead, endangered, and extinct languages in continuing education (as well as in vocational training). Following an introduction to the technical and conceptual foundations, several projects at the FHNW School of Business are presented and discussed from technical, ethical, and didactic perspectives. These dimensions are revisited in the next section and expanded into general and overarching considerations. Finally, possible and necessary steps are outlined that go beyond the purely technological discourse. Additionally, an outlook is provided on future possibilities related to new versions of large language models. This article provides the first comprehensive overview of the projects initiated by Oliver Bendel that are dedicated to dead, endangered, and extinct languages, including @ve (for Latin), @llegra (for Vallader), and kAIxo (for Basque), as well as Cleop@tr@ (Egyptian). It can be accessed via the publisher’s website or downloaded here as a PDF.

Posted on

Biases and Stereotypes in LLMs

The paper “Revisiting the Trolley Problem for AI: Biases and Stereotypes in Large Language Models and their Impact on Ethical Decision-Making” by Sahan Hatemo, Christof Weickhardt, Luca Gisler (FHNW School of Computer Science), and Oliver Bendel (FHNW School of Business) was accepted at the AAAI 2025 Spring Symposium “Human-Compatible AI for Well-being: Harnessing Potential of GenAI for AI-Powered Science”. A year ago, Sahan Hatemo had already dedicated himself to the topic of “ETHICAL DECISION MAKING OF AI: An Investigation Using a Stereotyped Persona Approach in the Trolley Problem” in a so-called mini-challenge in the Data Science degree program. His supervisor, Oliver Bendel, had told the other scientists about the idea at the AAAI 2025 Spring Symposium “Impact of GenAI on Social and Individual Well-being” at Stanford University. This led to a lively discussion. The student recruited two colleagues, Christof Weickhardt and Luca Gisler, and worked on the topic in a much more complex form in a so-called Challenge X. This time, three different open-source language models were applied to the trolley problem. In each case, personalities were created with nationality, gender, and age. In addition, the data was compared with that of the MIT Moral Machine project. Sahan Hatemo, Christof Weickhardt, and Luca Gisler will present their results at the end of March or beginning of April 2025 in San Francisco, the venue of this year’s event.