Conference Agenda

This paper analyzes the epistemic status of generative artificial intelligence systems, with an emphasis on large language models (LLMs), from a perspective explicitly oriented toward engineering design, evaluation, and education. Adopting an operational definition of knowledge as the capacity to construct predictive representations that guide successful action , the paper examines the epistemic limitations of purely textual LLMs arising from their lack of sensorimotor grounding , and contrasts this condition with the potential—and current limits—of multimodal and embodied AI systems to acquire functional world models through interaction with the environment ]. Drawing on theoretical frameworks from embodied cognition and predictive neuroscience], it is argued that multimodal systems can develop operational knowledge relevant to applications in robotics, simulation, and autonomous systems, while remaining conceptually distinct from human knowledge associated with subjective experience and self-representation . As its main contribution, the paper proposes an engineering-oriented epistemic taxonomy and derives from it a set of empirical evaluation criteria intended to guide the design, certification, and training of AI systems in educational and technological contexts.