Diversity-Aware Social Robots for Education and Social Assistance

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

This thesis asks what it means for a social robot to be competent with respect to the diversity of its users, and argues that metaphor is one of the key cognitive mechanisms underlying this competence. The starting point is the concept of interdiversity: diversity is treated not as a set of isolated variables, but as a relational condition in which user, technology, and research method co-determine one another. Within this framework, culture occupies a privileged position, not because it is the only relevant dimension, but because it represents the socially organized form of a more general cognitive strategy: the ability to select, from an indefinitely large encyclopedia of knowledge, the locally relevant portion that enables action under conditions of limited information. The thesis proposes that this capacity depends on metaphor and analogy as mechanisms of flexible categorization. The theoretical framework draws on semiotics, philosophy of language and mind, embodied and situated cognition, theories of analogy, and cognitive architectures. On this basis, the thesis develops a formal model that combines modal epistemic logic with a culturally informed knowledge graph, and tests it through a proof-of-concept implementation. On the empirical side, the thesis presents META, a multiplatform training system for metaphorical competence that integrates a humanoid robot (NAO), a web-based chatbot using a two-level prompting architecture with GPT-4o, and eight virtual reality scenarios. Three studies were completed with neurotypical adults, middle-school students, and children with Autism Spectrum Disorder; a fourth, a dedicated ASD pilot, was completed but is still under analysis. Across the completed studies, explicit metaphor comprehension tended to improve across conditions and populations. The findings also suggest that text-based interaction better supports verbal-reflective tasks, whereas embodiment is not uniformly beneficial with middle-school pupils; VR functions less as a score-raising device than as a semantic scaffold that shapes interpretation qualitatively; and that the local encyclopedias used in the prompt architecture, by working reliably across populations, languages, and modalities, offer a partial empirical response to the claim that key aspects of metaphor cannot be algorithmized.

Links and sources

Comments