Neural Network Architecture of Embodied Intelligence

Abstract

In recent years, advances in artificial intelligence and machine learning have been driven by advances in the development of large language models (LLMs) that are based on deep neural networks. At the same time, in spite of their substantial capabilities, LLMs have fundamental limitations, such as their spontaneous unreliability in facts and judgments; commission of simple errors that are dissonant with high competence in general; credulity, manifested by a willingness to accept a user’s false claims as true; and lack of knowledge concerning events occurring after training has been completed. Probably the key reason for these limitations is that bioinspired intelligence learning takes place through an assimilation of implicit knowledge in terms of an embodied form of intelligence to solve interactive real-world physical problems. Bioinspired studies of the nervous systems of organisms suggest that the cerebellum, which coordinates movement and maintains balance in human beings, is a prime candidate for uncovering methods of realizing embodied physical intelligence. Its simple, repetitive structure and ability to control complex movements offer hope for the possibility of creating an analog to adaptive neural networks. This paper explores the bioinspired architecture of the cerebellum as a form of analog computational networks that are capable of modeling complex, real-world physical systems. For a simple example, a realization of embodied AI in the form of a multicomponent model of an octopus tentacle is presented that demonstrates the potential for creating adaptive physical systems that learn from and interact with the environment.