Being There

1. The Mind is Not a Disembodied Logic Machine

"Mind, intellect," "ideas": these are the things that make the difference. But how should they be understood? Such words conjure nebulous realms. We talk of "pure intellect," and we describe the savant as "lost in thought." All too soon we are seduced by Descartes' vision: a vision of mind as a realm quite distinct from body and world.

Challenging tradition. For centuries, Western thought, heavily influenced by Descartes, has envisioned the mind as a separate, ethereal entity—a "ghost in the machine"—distinct from the physical body and the external world. This perspective often portrays intelligence as pure intellect, a realm of abstract thought and logical reasoning, detached from the messy realities of physical existence. This traditional view has profoundly shaped how we approach the study of cognition, leading to models that prioritize symbolic manipulation and internal, abstract problem-solving.

The practical reality. However, this book argues that such a disembodied view fundamentally misconstrues the nature of intelligence. Biological minds, including our own, did not evolve to be abstract logic machines. Instead, they are, first and foremost, organs for controlling biological bodies, designed to make things happen rapidly and effectively in a dynamic, often hostile, real-world environment. The emphasis on "pure thought" has led cognitive science down a path of modeling intelligence as the production of symbolically coded solutions to symbolically expressed puzzles, often ignoring the crucial roles of the body and the local environment.

Beyond the filing cabinet. This traditional approach, exemplified by projects like CYC, which aimed to instill common sense in a computer by encoding vast amounts of explicit knowledge, is fundamentally flawed. It assumes that intelligence is primarily about accumulating and manipulating a "filing cabinet" of explicit, language-like data. Yet, even simple creatures like cockroaches exhibit robust, flexible, practical intelligence without commanding large stores of explicit knowledge. The true challenge is not knowledge accumulation, but real-time, real-world responsiveness, rooted in the fluent coupling of organisms and their environment.

2. Brains are Controllers for Embodied Action

Might it not be more fruitful to think of brains as controllers for embodied activity? That small shift in perspective has large implications for how we construct a science of the mind.

A new perspective. Instead of viewing brains as isolated computational devices, this book proposes a radical shift: brains are primarily sophisticated control systems for embodied activity. Their evolutionary purpose is to generate appropriate actions rapidly, based on continuous interaction between the body and its changing environment. This perspective demands abandoning the Cartesian idea of the mental as distinct from the body, and the notion of neat dividing lines between perception, cognition, and action.

Practical implications. This shift has profound practical consequences. Consider the problem of assembling tight-fitting components with a computer-controlled machine. The "Pure Thought" solution involves complex feedback loops and precise computational adjustments. The "Embodied Thought" solution, however, is simpler: mount the assembler arms on rubber joints. This allows the parts to "jiggle and slide into place as if millions of tiny feedback adjustments to a rigid system were being continuously computed," effectively offloading computational complexity onto the body's physical properties.

Beyond abstract reasoning. The traditional focus on abstract problem-solving, like chess-playing programs, often overlooks the fundamental biological imperative: survival through fluent, real-world action. Brains are exquisitely geared to the production of actions, laid out in local space and real time. This means that understanding intelligence requires studying the brain not in isolation, but as part of a dynamic system that includes the body and its physical interactions with the world.

3. The Environment is an Active Partner in Cognition

The world can thus function as much more than just external memory. It can provide an arena in which special classes of external operations systematically transform the problems posed to individual brains.

Beyond passive backdrop. The environment is not merely a passive stage for cognitive activity; it is an active, integral partner in problem-solving. This concept, termed "embedded cognition," highlights how external structures and cues can profoundly simplify or transform the computational tasks confronting our brains. We actively "lean on the environment," exploiting its properties to reduce internal processing load, a principle dubbed the "007 Principle": "Do not develop explanations requiring expenditure of metabolic energy... until simple physical effects... are ruled out."

Niche-dependent sensing. Organisms are highly sensitized to specific aspects of their environment, forming an "Umwelt"—the effective environment defined by parameters relevant to their lifestyle. For example:

• A tick responds only to butyric acid, heat, and tactile contact to find a host.
• Herbert, a robot, uses low-resolution cues to find tables and then special-purpose routines to locate cans, ignoring other details.
  This selective sensitivity reduces information-processing load by focusing on ecologically significant cues.

Epistemic actions. We don't just perceive the world; we actively manipulate it to simplify our mental tasks. These "epistemic actions" are interventions whose primary purpose is to alter the nature of our own cognitive problems. Examples include:

• Physically rotating Scrabble tiles to prompt word recall.
• Grouping jigsaw puzzle pieces by color or edge type.
• Rearranging items on a work surface to simplify grocery packing.
  These actions transform complex internal computations into simpler perceptual or motor tasks, making problems tractable for our pattern-completing brains.

4. Cognition Emerges from Coupled Systems

The overall (ship-level) behavior is not controlled by a detailed plan in the head of the captain. The captain may set the goals, but the sequence of information gatherings and information transformations which implement the goals need not be explicitly represented anywhere.

Beyond central control. Complex intelligent behavior often "emerges" from the interactions of multiple simple components, without the need for a central planner, leader, or explicit blueprint. This "self-organization" is a hallmark of biological systems, from slime molds aggregating to form a mobile slug, to termite colonies building intricate nests. The collective pattern arises from local interactions, not from a global design.

Stigmergic coordination. Many emergent phenomena rely on "stigmergic algorithms," where individual actions modify the local environment, which in turn triggers further actions by other individuals or the same individual at a later time. This environment-based coordination is highly efficient:

• Termites deposit mud balls with chemical traces, attracting others to build columns and arches without a master plan.
• Ship navigation teams coordinate complex tasks by responding to local environmental cues (e.g., a chart on a table, a verbal request) rather than following a single, explicit plan.
  This distributed problem-solving offloads cognitive burden from individuals to the collective system.

Dynamical systems perspective. Understanding these emergent behaviors often requires a "Dynamical Systems" approach, which describes the continuous evolution of system states over time. This framework is adept at modeling the complex, coupled interactions between an agent and its environment, treating them as a single, integrated system. It focuses on collective variables and control parameters that capture higher-level patterns, rather than solely on individual components, providing a powerful lens for explaining how complex adaptive behaviors arise from simple, interacting parts.

5. Action Loops Simplify Cognitive Tasks

In many cases, perception should not, it seems, be viewed as a process in which environmental data is passively gathered. Instead, perception may be geared, from the outset, to specific action routines.

Perception as active. The traditional view of perception as a passive intake of information, followed by cognitive processing and then action, is challenged. Instead, perception is deeply intertwined with action, forming continuous "action loops" that criss-cross the organism and its environment. Our perceptual systems are not designed to build a detailed, objective model of the world, but rather to extract information relevant to immediate action and to guide further exploration.

Action-specific knowledge. Learning itself is often action-specific. Infants, for example, learn about slopes differently when crawling versus walking. Knowledge acquired through crawling about slopes does not automatically transfer to walking, suggesting that understanding is deeply tied to the specific motor routines used to engage with the environment. Similarly, adults adapting to distorting lenses show adaptation specific to particular motor loops (e.g., overhand dart throwing), not a general perceptual correction.

Animate vision. Research in "animate vision" further illustrates this active role of perception. Instead of constructing a full 3D internal model, our visual system uses rapid eye movements (saccades) to sample the world as needed, treating "the world as its own best model." This strategy avoids the computational cost of maintaining a detailed internal representation, relying instead on quick, repeated interactions with the external scene. This suggests that our subjective experience of a rich visual world is partly an "illusion" supported by our ability to rapidly access detailed information on demand.

6. Brains Operate as Pattern-Completing Engines

The major lesson of neural network research, I believe, has been to thus expand our vision of the ways a physical system like the brain might encode and exploit information and knowledge.

Beyond symbolic manipulation. Artificial neural networks (connectionist models) offer a powerful alternative to classical AI's rule-and-symbol paradigm. Inspired by the brain's architecture, these models consist of many simple processing units linked in parallel. Knowledge is stored in weighted connections between units, and processing involves the spreading of activation. These systems excel at "pattern completion"—recreating whole patterns from partial cues—and associative memory, rather than explicit logical inference.

Strengths and weaknesses. Neural networks are robust, fast, and tolerant of noisy or incomplete data, making them ideal for tasks like motor control, face recognition, and reading handwritten zip codes. However, they are not intrinsically well-suited for sequential reasoning or long-term planning. This profile of "good at Frisbee, bad at logic" suggests that our brains, at their core, are powerful pattern-recognition devices, which then leverage external tools for more abstract, sequential tasks.

Scaffolding pattern completion. This inherent computational profile makes brains ideal candidates for extensive external scaffolding. For complex problems like long multiplication, we offload intermediate results onto external media (e.g., pen and paper), reducing the problem to a sequence of simpler pattern-completion tasks. This suggests that classical AI may have mistakenly projected the cognitive profile of an agent plus its environment onto the "naked brain," overlooking how external structures amplify our basic neural capacities.

7. Evolution is a Tinkerer, Not an Engineer

Jacob likens evolution to a tinkerer rather than an engineer. An engineer sits down at a blank drawing board and designs a solution to a new problem from scratch; a tinkerer takes an existing device and tries to adapt it to some new purpose.

Opportunistic design. Natural evolution rarely designs solutions from scratch; instead, it acts as a "tinkerer," adapting existing structures for new purposes. This evolutionary holism means that complex adaptive designs are built incrementally, with each intermediate form being a robust, functional system. This process often leads to solutions that appear "messy" or non-intuitive from an engineer's perspective, but are highly effective given their historical constraints.

Path dependence. The constraints of evolutionary history mean that current solutions are heavily "path-dependent." For example, our lungs are thought to have evolved from fish swim bladders, leading to features that, while functional for breathing, might seem suboptimal from an ideal design standpoint. This historical baggage means that biological systems often exploit gross physical features like friction, elasticity, or noise, alongside computational strategies, to achieve robust adaptive behavior.

Simulated evolution. To understand these opaque, historically contingent solutions, researchers use "genetic algorithms" to simulate evolution. By evolving neural network controllers for robots (e.g., insect locomotion), these simulations reveal robust and often non-obvious strategies that rely on close, continuous interactions between the controller and the environment. This approach helps uncover solutions that human designers might overlook due to their tendency to seek neat, decomposable problem solutions.

8. Internal Representations are Action-Oriented and Partial

The internal representations the mind uses to guide actions may thus be best understood as action-and-context-specific control structures rather than as passive recapitulations of external reality.

Reconceptualizing representation. The traditional view of internal representations as objective, action-neutral models of the world is too narrow. Instead, biological brains often employ "action-oriented" or "personalized" representations. These are internal states that simultaneously encode aspects of the world and prescribe appropriate actions. For example, a robot's internal map might directly specify the motor activity needed to link different locations, blurring the line between description and control.

Local and partial. These representations are often local and partial, focusing on idiosyncratic, locally effective features rather than universal, defining properties. In animate vision, for instance, a visual search for a coffee cup might rely on its specific color ("my cup is yellow") rather than its abstract shape or function. This computationally cheap approach is effective in ecologically normal circumstances, especially when agents actively structure their environment (e.g., using brightly colored mugs) to simplify perceptual tasks.

Beyond decoupling. While some representations can be "decoupled" from immediate sensory input (allowing thought about absent or counterfactual states), many are deeply coupled to ongoing sensorimotor activity. The adaptive oscillator, for example, entrains its internal rhythms to external temporal patterns, effectively representing periodicity through a process with intrinsic temporal properties, rather than an arbitrary symbol. This highlights that representation can be a dynamic process, not just a static data structure, and can be intrinsically "wide," spanning brain, body, and environment.

9. Language is the Ultimate External Scaffolding

Public language is in many ways the ultimate artifact. Not only does it confer on us added powers of communication; it also enables us to reshape a variety of difficult but important tasks into formats better suited to the basic computational capacities of the human brain.

Beyond communication. Language's role extends far beyond mere communication. It acts as a powerful cognitive tool, an "ultimate artifact," that fundamentally alters the nature of computational tasks. Like physical tools such as scissors, language exhibits a "double adaptation": it fits our manipulative capacities (e.g., vocalization, writing) and confers new powers (e.g., making neat cuts in thought). It allows us to reshape complex problems into formats more amenable to our pattern-completing brains.

Vygotskian insights. Developmental psychology, particularly Vygotsky's work, highlights language's role in scaffolding thought and action. "Private speech" (self-directed talk) helps children guide their own behavior, focus attention, and overcome difficult tasks. This internal rehearsal of language acts as an "extra control loop," modulating the brain's use of its own resources. It's not just expressing pre-formed thoughts, but actively shaping and structuring them, allowing us to "think about thinking."

Trading spaces. Language enables us to "trade culturally achieved representation against individual computational effort." By externalizing thoughts into words, sentences, and texts, we create stable, inspectable objects that:

• Offload memory: Diaries, notes, and labels act as external storage.
• Simplify learning: Labels provide powerful cues, shrinking search spaces for concept acquisition.
• Coordinate action: Explicit plans and schedules reduce on-line deliberation.
• Facilitate complex reasoning: Written text allows us to reorganize, compare, and refine ideas in ways impossible for the unaugmented brain.
  This externalization transforms problem spaces, making otherwise intractable cognitive tasks manageable.

10. Human Reason is a Distributed, Extended Process

Much of what we commonly identify as our mental capacities may likewise, I suspect, turn out to be properties of the wider, environmentally extended systems of which human brains are just one (important) part.

Fuzzy boundaries of mind. The insights of embodied and embedded cognition challenge the traditional notion of the mind as solely residing within the skull. Cognitive processes often extend beyond the brain, involving the body, tools, and the environment in complex, integrated computational and dynamic wholes. This "leakage" of mind into the world suggests that the intelligent system is a spatio-temporally extended process, not limited by the skin.

The scaffolded brain. Human brains, while not fundamentally different in their basic pattern-completing nature from other animals, excel in creating and exploiting external scaffolding. We build "designer environments"—cultures, institutions, languages, and physical tools—that augment our individual cognitive profiles. These external structures constrain and amplify our problem-solving, allowing us to achieve complex behaviors with less individual computational effort. Our brains make the world smart, so we can be "dumb in peace."

Implications for self and study. This distributed view of cognition has profound implications. While individual consciousness may supervene on the brain, the "flow of reason and thoughts, and the temporal evolution of ideas and attitudes," are determined by the intimate interplay of brain, body, and world. This calls for a wider methodological approach in cognitive science, integrating neuroscience, robotics, developmental studies, and cultural analyses to understand these hybrid, extended cognitive systems. The question of "where the mind stops and the rest of the world begins" becomes a pragmatic one, focusing on the functional role of reliable, accessible, and trusted external resources.

Last updated: December 29, 2025