We have a deep, almost religious obsession with the central controller. When we watch a robot navigate a room, or when we look at our own choices, we assume there is a command deck. A captain sits at the helm. In cognitive neuroscience and AI, this has been the default settings for decades: sensory information comes in, a central engine weighs the options to make a 'decision', and a command goes out to the motor system. We call it intelligence. But what if this command center is a neat fiction?
A new framework by Indiana University professor Tom James, published in the Journal of Cognitive Neuroscience under "Sensorimotor Mechanisms of Decisions and Actions", suggests we are looking at the brain all wrong. James argues that the traditional "sandwich model" of cognition—where decision-making is a high-level cognitive engine wedged between perception and action—is a scientific fallacy. The physical brain does not actually possess a localized, discrete neural engine for making decisions. There is no board room inside your prefrontal cortex debating options. Instead, what we call a choice is an emergent property of decentralized, simultaneous, and circular interactions between sensory systems, physical movement, and the environment.
As someone who spent years evaluating venture funding for enterprise AI startups, I find this neurophilosophy overhaul incredibly satisfying. We have spent billions building massive "cognitive models" that mimic this exact three-stage sandwich model, believing it is the only way to generate intentionality. James’s work suggests that intent is not a top-down decision. It is a loop.
Dismantling the Sandwich Model
The "sandwich model" is easy to understand because it matches our everyday intuitions. You feel thirsty. You perceive a glass of water on the table. You decide to grab it. You move your hand. It is a linear, clean, step-by-step causal sequence. Your brain moves from sensory input, to a cognitive decision, to motor action. It feels like common sense. Our everyday language is built entirely around this idea. We say, "I decided to take the job," or "I chose the coffee over the tea."
Model-based cognitive neuroscience has historically re-verified this assumption. By structuring experiments around discrete choice tasks—press button A or button B—researchers design models that isolate a specific "decision phase" in time and space. They look for the exact moment the scale tips in the brain.
But James points out a massive, gaping hole in this paradigm. While we can easily point to specialized, physical neural networks in the brain that handle sensory incoming data, and we can point to specific structures that execute physical movement, we cannot find the middle of the sandwich. There is no "decision region" that coordinates them. Instead of a discrete, localized controller, the brain runs on a continuous mixture of sensory, sensorimotor, and motor processes. What we think of as decision-making is actually "action selection," a dynamic, circular process that occurs across the entire system simultaneously. The sandwich doesn't have a filling.
The Mystery of the Missing Neural Center
Why have we spent centuries searching for a central decision-making neural engine that doesn't exist? Part of the answer lies in how we study the brain. When researchers set up fMRI or EEG tests, they look for active zones. But activity is not a command.
If you map the brain during a decision-making task, you will see a flurry of activity in various regions. You might see the prefrontal cortex light up, or the basal ganglia engage. But these areas do not function as a localized, top-down executive switchboard. Instead, sensory inputs and motor outputs are constantly talking to each other in parallel loops. The motor cortex is already preparing movements before a "decision" is formally reached, and sensory regions are shifting their focus based on motor intentions.
James argues that "action selection" is a much more accurate term. In this framework, the brain does not decide and then act. Rather, multiple potential actions compete for execution in real time. This competition is influenced by sensory inputs and internal states, but the winner is selected through the physical dynamics of the sensorimotor loop itself, not by a separate cognitive arbiter. This is a decentralized network, not a top-down hierarchy. For anyone trying to build efficient edge AI or autonomous robotics, this is a massive design clue. You don't need a central brain to get purposeful activity.
Dennett, Physicalism, and the Center of Mass
To explain why we feel like we are making choices when the brain is just running physical loops, James adopts a physicalist framework. He draws heavily on the work of philosopher Daniel Dennett. Under a strict physicalist view, only physical things can cause other physical things. The firing of a neuron is a physical event; it can cause a muscle contraction. The physical world is closed.
Decisions, however, are mental, nonphysical abstractions. A decision cannot physically push a neuron to fire any more than a shadow can push a physical wall. So how do we reconcile this?
James offers Dennett’s analogy of an object's "center of mass" (CoM). The center of mass is an incredibly useful mathematical concept. We use it to calculate how objects balance, spin, and fall. But the center of mass has no physical existence. It is an abstract property. You cannot touch it, and it cannot exert a physical force independent of the object. If you move a baseball, its center of mass moves, not the other way around.
In the same way, James argues, a "decision" is a conceptual abstraction that describes the collective behavior of the system. It is a highly useful summary of why we moved our hands, but it has no independent physical force. It is the result of the brain's physical processes, not the cause of them.
Consider how we talk about institutions. We say, "The university decided to build a new library." We use this abstract shorthand to simplify a complex reality. In reality, there was no single physical entity called "the university" that made a decision. Instead, there was a series of emails, board meetings, donor conversations, architectural sketches, and budget allocations. The decision is a description of this distributed, physical process. Looking for a "decision" in the brain is like walking around a campus looking for the physical entity called "the university." You will only find buildings, people, and actions.
The Cartesian Theater and Infinite Regress
The assumption of a central controller doesn't just lack neural evidence; it is a logical trap. Philosophers call this the Cartesian Theater paradox, a concept coined by Dennett.
If you argue that the brain works by having a central cognitive controller that receives sensory input, decodes it, decides what to do, and sends instructions to the motor systems, you have just introduced a homunculus. You've placed a tiny person inside the brain who watches the sensory screen and pulls the motor levers.
But how does that tiny person's brain work? It would require an even smaller person inside its head to watch its sensory screen and make its decisions. This leads to an infinite regress, an endless loop of smaller and smaller controllers without ever explaining how decision-making actually works. It is a logical dead end.
By abandoning the central controller, we solve the paradox. The brain doesn't need an internal observer. It is a physical system that responds to its environment through direct, circular sensorimotor loops. Intent is not something calculated in a control room and shown to the mind. It is the path the system takes as it interacts with the world.
Braitenberg's Legacy: The Wall-Following Robot
To prove that complex, intentional-looking behavior can emerge without a central decision-maker, James highlights a classic thought experiment in robotics: the wall-following vehicle.
Imagine a simple robot equipped with two sensors and two motors. One sensor detects distance to a wall on its right, and the other detects obstacles in front. The wiring is incredibly basic: if the right sensor gets too far from the wall, the left wheel speeds up to steer closer. If the front sensor detects an obstacle, both wheels reverse.
When you put this robot in a room, it will navigate along the walls, turn corners, avoid obstacles, and look like it is exploring the space with intent. An outside observer might say, "The robot decided to turn right because it saw a wall," or "The robot wants to follow the wall."
But the robot has no internal planner. It has no map of the room, no memory of where it has been, and no cognitive "decision engine" to weigh options. It is just a set of physical sensorimotor loops interacting with the environment. The strategic, goal-driven behavior is not inside the robot. It is an emergent property of the robot, its physical body, and the shape of the room.
This is a vital lesson for modern AI. In the VC world, we see startup founders pitching complex LLM-based planning systems for simple robotic tasks. They are building massive, expensive "cognitive engines" in the cloud to tell a robot arm how to pick up a box. It is the sandwich model all over again. What James’s work suggests is that we should be building tighter, faster, physical loops that let the behavior emerge naturally from the machine's interaction with the physical world.
Active Sensing and the Embodied Future
If we accept that the sandwich model is dead, how do we study cognition? James argues that cognitive neuroscience needs a major shift in experimental methods.
Historically, we have studied decision-making by putting participants in highly artificial, static environments. We strap them into an fMRI scanner, show them a flash on a screen, and tell them to press a button. This setup purposely breaks the circular loops between brain, body, and environment because it freezes the body and controls the input. It forces a linear process, which then artificially confirms the linear sandwich model.
To map how the brain actually works, we need to study active testing environments where participants have physical agency. We must allow them to dynamically update their sensorimotor processes through active sensing—moving their eyes, adjusting their hand position, and exploring their surroundings in real time.
This model of embodied cognition and ecological psychology is not just an academic debate. It is a technological path forward. In autonomous vehicles, defense tech, and industrial robotics, the limit is often latency and power. If a drone has to send video to a central cloud processor, wait for a decision, and then send a command to the rotors, it crashes. If we instead design systems where sensing and action are direct, circular, and decentralized, we get faster, more resilient, and much cheaper systems.
The brain works because it doesn't need to make decisions to act. It just moves, senses, and adapts. It's time our models—and our machines—did the same.