Challenges and Opportunities of Using LLMs
Monday, August 26, 2024
Full-day hybrid workshop in conjuction with RO-MAN 2024
The main topic of our workshop will be challenges and opportunities of using large language models (LLMs) in robotic applications for well-being, which is strongly in line with the RO-MAN 2024 theme of “Embracing Human-Centered HRI”, which seeks to design human-robot interactions “towards building a promising and excellent human quality of life, all for our wellness”.
In the last year, the use of Large Language Models (LLMs) has spread rapidly in our society, from information retrieval to text classification, as well as robotic applications. Given these rapid advancements, recent studies have attempted to identify challenges and opportunities for using LLMs in various contexts, such as in healthcare and medicine, among others. These works have highlighted that the use of LLMs in healthcare and well-being pose risks in terms of data privacy (i.e., leaking confidential and sensitive patient information), questionable credibility and accuracy of information (i.e., perpetuating inaccurate information from open internet sources, aka “hallucination effect”), data bias (i.e., perpetuating the bias of data, such as gender, ethnicity etc., that feeds into the training of LLMs), interpretability of LLMs (i.e., lack of interpretability of the generated results and the decision-making process behind the choice of the generated text), role of LLMs (i.e., understanding the role of LLMs in providing responses in healthcare contexts, for instance, is it an assistant or clinician?), deployment of LLMs (i.e., trade-off between costs and resources needed to deploy the model). However, there is a lack of analysis and understanding of the challenges and opportunities when using LLMs for robotic applications.
Previous HRI research that leverages the use of LLMs has focused on applications for planning, control, and manipulation in lab environments, but not yet for well-being purposes. For example, Huang et al. (2023) have used LLMs to decompose high-level tasks into mid-level plans and they proposed a procedure that semantically translates these plans into admissible actions.
While robots for well-being are becoming an increasingly relevant line of HRI research, as people have shown increased interest in using digital tools to improve their well-being, the use of LLMs in such applications is still unexplored. Therefore, there is a need for discussing and identifying the risks and potentials of using LLMs in human-robot interaction for well-being.
Exploring the challenges and opportunities of using LLMs in robots for well-being is extremely important to gain insights into how these models affect the users’ perception of robots in well-being applications. The conversations held in this workshop could serve as the first stepping stone to guide the HRI community in designing robotic applications for well-being that leverage the use of LLMs in a responsible and ethical manner.
Topics included in the workshop, but not limited to, are the following:
Affective robotics for well-being
Socially assistive robots for well-being
Robotics assistant/companion/coach for well-being
Robotic design for well-being
User studies for well-being in the real world
Adaptation and personalization for well-being applications
Machine learning for well-being
Concept papers on the role of robots for well-being
Methods to measure well-being
Ethics, Privacy, and Data Security Considerations
Linguistic Communication and Dialogue
LLMs for well-being robotics