Название: Embodied Intelligence: Multidisciplinary Perspectives on Natural, Artificial, and Hybrid Systems
Автор: Sheila L. Macrine, Jennifer M. B. Fugate, Arsen Abdulali, Josie Hughes
Издательство: The MIT Press
Год: 2026
Страниц: 459
Язык: английский
Формат: pdf (true)
Размер: 34.5 MB

An exploration of embodied intelligence that moves beyond the traditional focus on brains and code to the role of embodiment across various disciplines.

Intelligence research is undergoing a radical transformation, moving beyond the traditional focus on brains and code to increasingly recognize the role of embodiment, as well as our understanding of goal-directed behavior across different scales and substrates. In this edited collection, experts across fields including philosophy, phenomenology, neuroscience, cognitive psychology, robotics, Artificial Intelligence (AI), bio-inspired design, biology, and bioengineering initiate transdisciplinary dialogues and facilitate the sharing of insights on embodiment, enabling a new understanding of embodied intelligence and how intelligence manifests across diverse substrates.

By embracing a broad definition of embodied intelligence, the contributors transcend the traditional divide between the biological and the artificial, recognizing the potential for intelligence to emerge in unexpected forms. This perspective challenges us to reconsider our assumptions about the nature of intelligence and to appreciate the remarkable diversity of intelligent behavior in the world around us. This broadened concept holds immense promise to profoundly reshape our understanding of ourselves, the technologies we create, and the very nature of intelligence itself.

Part I examines the philosophical and theoretical foundations of Embodied Intelligence (EI). It begins by highlighting a significant convergence between human-­embodied cognition (HEC) and Embodied Artificial Intelligence (EAI), emphasizing a paradigm shift that recognizes the crucial and reciprocal influence of the body and environment in shaping intelligent be­hav­ior, moving away from traditional brain-­centric perspectives. Furthermore, it proposes a “pattern theory of self” as a dynamic and interconnected system for both ­ humans and robots, introducing the concept of “robodiment” for robots, characterized by physical continuity, self-­movement, and intelligible be­hav­ior, alongside outlining capacities for “ipse identity.” This part also addresses the ongoing debate surrounding brain repre­sen­ta­tion, introducing active inference with generative models as a framework for understanding repre­sen­ta­tional pro­cesses like belief updating, perception, and action through approximate Bayesian inference.

Part II explores Embodied Artificial Intelligence (EAI). EAI is modeled based on collective intelligence and draws on the princi­ple of active inference and the ­free energy princi­ple (FEP). This part discusses the importance of equipping artificial embodied agents (AEAs) with a perceptuo-­motor apparatus for perceiving and acting, which is crucial for grounding, embodiment, and situatedness. It also highlights the importance of body morphology, sensor morphology, and sensorimotor coordination for control and perception, criticizing the prioritization of the “robot brain.” The concept of “strong” versus “weak” AI is tackled, defining the former as exhibiting human-­like flexibility and creativity. The debate hinges on the definition of concepts and the necessity of embodiment for meaning, suggesting that strong AI with human-­like concepts remains a possibility despite dif­fer­ent sensorim- otor systems. This part discusses ­ whether large language models (LLMs) in EAI are merely “weakly” embodied and therefore inherit aspects of Good Old-­Fashioned AI’s (GOFAI’s) symbol-­grounding prob­lem. The part ends with research directions for stronger EAI, including bio-­inspired design, hybrid simulation to bridge the real­ity gap, and computational methods for studying embodied interactions, emphasizing standardization, accessible tools, and cross-­disciplinary collaboration to advance the field.