Resumo: Human beings not only entertain a conversation with other beings, like an LLM does, but based on their sensory organs, are able to make sense of their surrounding environment, having feelings and experiences while interacting with it. They are able to truly understand the meaning of exchanged communications and evolve high-level thinking before acting in the world. In this talk, we investigate the role of Understanding, what it is, according to different thinkers, and how this capability might be made available in a cognitive architecture. We discuss whether LLMs (and Transformers, in a general sense) can be afforded to exhibit true understanding, or if they just simulate this capability, acting like “probabilistic parrots”, without a true understanding of what they say. Finally, we propose an ontology of reality to support the development of Cognitive Architectures, which we believe might enable future artificial agents to attain a human-comparable understanding of reality, and possibly consciousness.

O link para participar é meet.google.com/obe-spbe-zsc