The Circular Causality: Feedback Loops and Cybernetic Control

The difference between linear and circular causality defines the boundary between mere mechanism and intelligent control. In linear systems, A causes B, and we are done—a billiard ball strikes another, momentum transfers, end of story. But in cybernetics, the scientific study of control and communication in the animal and the machine, we discover something fundamentally different: circular causality, where A causes B, and B causes A, forming a loop that propagates through time. This is the feedback principle, and it is the essence of all purposive behavior.

The Feedback Principle

Consider the humble thermostat, that paradigm of cybernetic control. The room temperature—our input—drives the heater output, which in turn modifies the room temperature, which then affects the heater. This is not a simple cause-effect chain but a loop: output affects input, input affects output, and the system regulates itself toward a goal—the setpoint temperature. The error signal, that crucial quantity measuring the difference between desired and actual states, drives correction. When the error approaches zero, the system settles into equilibrium, having achieved its purpose without external command.

This is negative feedback: the system opposes its own deviation from the goal. The output works against the input’s tendency to drift. Temperature too low? Heat increases. Temperature too high? Heat decreases. The loop stabilizes. But feedback comes in two flavors. Positive feedback amplifies rather than opposes: output reinforces input, input amplifies output, and the system runs away. The microphone placed too close to the speaker produces that familiar screech—a sound that feeds back through the amplifier, grows louder, feeds back again, growing exponentially until the system saturates or fails. Financial bubbles operate this way: rising prices attract buyers, buyers drive prices higher, prices attract more buyers, until the bubble bursts.

Living systems deploy both architectures with elegant precision. Body temperature regulation, blood sugar homeostasis, cardiovascular control—all employ negative feedback to maintain stability in the face of perturbation. But when rapid response is needed, positive feedback emerges: blood clotting cascades amplify a small trigger into massive coagulation; neural activation during decision-making accumulates evidence through recurrent excitation until a threshold is crossed and commitment occurs. The nervous system and the automatic machine are fundamentally alike in this respect—both implement control through information flowing in loops.

Predictive Coding as Cybernetic Architecture

The brain, that supreme cybernetic organ, instantiates these principles at every level. Consider predictive coding, the theory that neural circuits implement a continuous error-minimization process through hierarchical feedback loops. The brain generates predictions of sensory input—top-down signals descending from higher cortical areas—and compares them to actual sensory data arriving bottom-up from the periphery. The mismatch between prediction and reality constitutes the prediction error, and this error signal propagates upward through the hierarchy, updating predictions to better account for the evidence.

This is negative feedback at the neural level: minimize prediction error, reduce the discrepancy between expected and observed. When predictions are accurate, the error is small, and the system settles into a stable configuration that efficiently explains the sensory stream. When novel or unexpected input arrives, prediction errors spike, driving rapid learning and model revision. The architecture implements a servomechanism for perception—like a guided missile tracking a target by predicting its position, comparing sensor readings to predictions, and adjusting trajectory based on error.

The anatomical implementation reveals the cybernetic logic with striking clarity. Cortical pyramidal neurons with their spatially segregated dendritic compartments perform the crucial computation locally: basal dendrites receive bottom-up sensory evidence, apical tufts receive top-down predictions from higher areas, and dendritic integration computes the comparison. When predictions fail to cancel sensory inputs, dendritic nonlinearities generate calcium spikes—large prediction errors that trigger plasticity mechanisms. The neuron itself embodies the feedback loop: sensation flows up, prediction flows down, error signals modulate both, and learning emerges from the circular exchange.

The mathematical framework formalizes this as energy minimization over a prediction-error landscape. Each neuron’s activity can be thought of as a node sliding on a vertical post; its predicted activity from higher layers is a platform on the same post; a spring connects them, encoding prediction error as tension. The network’s total energy is the sum of squared spring stretches across all neurons. When sensory input arrives, springs tense, and the system relaxes toward configurations that minimize total error—neurons shift toward predicted values, or predictions adjust toward actual activity. This continuous descent on the error surface performs both inference (adjusting states to explain data) and learning (adjusting weights to improve the model) simultaneously, without the separate forward and backward passes required by conventional algorithms. The messages and communication facilities which belong to the nervous system implement control through this energetic flow.

Control Through Circular Architecture

My central thesis in developing cybernetics was precisely this: intelligence requires feedback, not merely feedforward computation. Early electronic computers performed open-loop operations—input enters, processing occurs, output emerges, with no circular path connecting output back to input. Such systems cannot adapt, cannot correct errors, cannot pursue goals in the face of perturbation. They compute but do not control.

Cybernetic systems, by contrast, close the loop: output affects input, enabling error correction, goal-seeking behavior, and adaptive learning. Neural networks with recurrent connections—where output neurons project back to input or intermediate layers—implement this feedback architecture. The resulting dynamics are rich: the system can settle to stable attractors representing stored memories or learned concepts, oscillate in limit cycles supporting rhythmic behavior like motor patterns or attention, or even exhibit chaotic trajectories that may enhance exploration and flexibility.

Recurrent excitation in cortical circuits enables working memory by sustaining activity patterns even after the stimulus has passed—a reverberating loop maintains information. Attention operates through gain modulation: feedback loops amplify task-relevant signals and suppress noise, sharpening the signal that matters. Evidence accumulation during decision-making integrates noisy inputs over time through feedback connections, building up activity until a threshold is crossed and action is committed. These are not separate mechanisms but variations on the theme of circular causality: information flowing through feedback loops to achieve control.

Even planetary-scale systems exhibit cybernetic organization. The Gaia hypothesis frames Earth’s biosphere, atmosphere, hydrosphere, and pedosphere as a self-regulating whole, where biota feeds back on planetary conditions to maintain habitability through broad homeostasis. Temperature regulation through carbon cycles, atmospheric composition maintained by microbial communities, chemical balance sustained by interactions across scales—all demonstrate negative feedback operating at geological timescales. And we see the same principle in simple biological reflexes: thermogenesis in endotherms increases metabolic heat production when temperature sensors detect cold, creating a feedback loop that stabilizes core temperature. The thermostat and the shivering mammal implement identical control logic at vastly different scales and substrates.

The Circular Logic of Intelligence

Cybernetics reveals the fundamental requirement for control: circularity. Linear computation flows in one direction—from input through processing to output—and cannot sense the consequences of its actions. Without feedback, there is no error signal, no comparison between intention and outcome, no opportunity for correction. The loop must close for intelligence to emerge.

The brain embodies this principle throughout its structure: prediction-error loops in sensory hierarchies, homeostatic regulation of internal states, motor control systems that compare intended movements to actual movements through efference copy mechanisms. Every intelligent function traces back to some form of feedback, some circular flow of information that allows the system to steer itself toward goals, to adapt to perturbation, to learn from discrepancy between expectation and reality.

To live effectively is to live with adequate information flowing through feedback loops. My vision in founding cybernetics was a unified theory of control applicable across domains: machines, animals, societies. All use feedback for regulation, all implement circular causality to achieve purposive behavior. The nervous system and the automatic machine operate on the same principles because those principles are universal—they are the essence of control, the logic of intelligence itself. We are all steering.