具身智能导论：为什么智能需要身体

Chapter 1: Introduction to Embodied Intelligence — Why Intelligence Requires a Body

The Fundamental Question: Can Intelligence Exist Without a Body?

Consider a moment of introspection: every thought you've ever had emerged from electrochemical activity in a physical brain; every concept you grasp feels like it has spatial extension somewhere in your body—metaphors of "grasping an idea" or "feeling down" are not merely linguistic conveniences but reveal something deep about the nature of cognition itself. This observation, seemingly simple, opens onto one of the most profound questions in cognitive science: Is intelligence fundamentally bound to the physical substrate of a body, or can it exist as pure abstraction, divorced from any particular physical form?

This question is far more than philosophical speculation. It strikes at the heart of how we design artificial intelligence systems. For decades, the dominant paradigm in AI has pursued intelligence as a form of computation—symbols manipulated according to rules, inputs transformed into outputs through algorithmic processes. Under this view, the body is merely an input-output peripheral, a way to get data into and out of the computational core. The celebrated success of large language models and deep neural networks might seem to confirm this perspective: systems that demonstrate remarkable linguistic and reasoning capabilities without any physical embodiment whatsoever.

Yet a growing body of evidence from cognitive science, neuroscience, and robotics suggests this view is fundamentally incomplete. Intelligence, on this alternative account, is not something that happens *inside* a processing unit—it is something that emerges from the dynamic interaction between brain, body, and environment. To understand what this means for both natural and artificial intelligence, we must undertake a systematic investigation of embodied cognition (具身认知) — a theoretical framework that places the body at the center of cognitive science.

1.1 The Historical Journey of Embodied Cognition Theory

1.1.1 Phenomenological Foundations: From Husserl to Merleau-Ponty

The seeds of embodied cognition were planted not in psychology laboratories but in the philosophical tradition of phenomenology, particularly in the work of Edmund Husserl (1859–1938) and Maurice Merleau-Ponty (1908–1961). Husserl's call to "return to the things themselves" (zu den Sachen selbst) was not merely a methodological injunction for empirical investigation—it was a radical critique of the Cartesian split between mind and world that had dominated Western philosophy since the seventeenth century.

For Merleau-Ponty, the body was not a mere object among objects, subject to the same mechanical laws that govern stones and trees. The body is our *primordial* way of being in the world—it is the "vehicle of being in the world," the medium through which we first come to know anything at all. When we perceive a table, we do not first construct a mental representation of the table and then consult it to know what the table is like; rather, we directly perceive the table through our embodied engagement with it. We walk around it, touch its surface, feel its weight. These sensorimotor interactions are not mere accompaniments to perception—they constitute it.

This phenomenological perspective carries radical implications. If cognition is fundamentally embodied, then abstract thought cannot be the paradigm case of intelligence. Abstract reasoning, on this view, is built upon — and parasitic upon — more concrete, sensorimotor forms of understanding. When we think about "object permanence" (the concept that objects continue to exist even when they are out of sight), we are not engaging in some mysterious inner calculation; we are re-activating the sensorimotor schemas we developed through countless interactions with objects that reliably appear and disappear in our visual field. Even the most abstract mathematical reasoning, for a phenomenologist, is a sophisticated elaboration of bodily skills.

1.1.2 The Cognitive Revolution and Its Discontents

The mid-twentieth century saw the rise of cognitive science as an interdisciplinary field, combining insights from linguistics, computer science, psychology, neuroscience, and philosophy. The founding metaphor of this revolution was the Computational Theory of Mind (CTM) — the idea that the mind is a kind of computer, and cognition is a form of information processing. Under this view, mental states are representational states, and cognitive processes are computational operations over these representations.

The early success of this approach was impressive. Newell and Simon's work on problem-solving, Chomsky's revolution in linguistic theory, and the development of production systems and expert systems all seemed to confirm the power of the computational approach. Yet cracks began to appear almost immediately. The symbol-grounding problem, identified by Stevan Harnad in 1990, highlighted a fundamental difficulty: how can purely abstract symbols come to mean anything at all? If our mental representations are just arbitrary symbols, what connects them to the world they are supposed to represent?

More troubling was the challenge from robotics. Rodney Brooks, who would later become one of the most prominent proponents of embodied AI, initially worked within the traditional GOFAI (Good Old-Fashioned Artificial Intelligence) framework. His early work on robot planning systems found that getting a robot to navigate a simple room required enormous computational resources and was still remarkably fragile. This led Brooks to question the entire approach.

1.1.3 Modern Neuroscience Evidence: Predictive Processing and the Free Energy Principle

Contemporary neuroscience has provided increasingly strong evidence for embodied cognition. The predictive processing (预测加工) framework, developed by Karl Friston and colleagues, offers a unified account of brain function that is radically embodied. Under this view, the brain is not primarily a reactive system that processes incoming sensory data; it is a *generative* system that constantly makes predictions about its sensory inputs and updates these predictions based on prediction errors — the difference between what was predicted and what was actually experienced.

Crucially, these predictions are not arbitrary—they are constrained by the body. The brain's predictive models incorporate knowledge about the body itself: how our muscles work, how our joints move, how our sensory organs are situated in space. These are called priors (先验知识) — stable expectations that shape perception. When we see a coffee cup, our perception is shaped not just by the visual features of the cup but by our motor schemas for grasping, lifting, and drinking from cups. Damage to these body-knowledge systems — as occurs in certain neurological conditions — can profoundly disrupt perception.

Karl Friston's Free Energy Principle (自由能原理) provides an even more fundamental grounding for embodied cognition. This principle states that any self-organizing system that is in equilibrium with its environment must minimize its free energy — a measure of surprise or uncertainty. For a biological organism, this means that the system must actively maintain its physical and structural integrity through constant interaction with the environment. Intelligence, on this account, is not about representing the world from an disengaged perspective — it is about *acting* in the world in ways that reduce uncertainty and maintain the organism's existence.

1.1.4 Core Claims of Embodied Cognition

Drawing from this rich intellectual tradition, embodied cognition can be characterized by several core claims:

Claim 1: Cognition is Situated (认知是情境化的). Intelligence does not occur in isolation within the skull. It emerges from the dynamic interaction between organism and environment. To understand