Feedback loop

A thermostat keeps a room at 20 °C by cooling when the sensor reads high and warming when it reads low. A microphone held next to a speaker produces an eardrum-splitting howl because each bit of sound feeds back into itself. An ant trail gets stronger exactly where ants are walking, which makes more ants walk there, which makes it stronger. Three very different systems, one shared structure. The word for it is feedback loop: a circuit in which a system's output is routed back as part of its own input.

What is a feedback loop?

A feedback loop is any arrangement where a system's output affects its own subsequent input, making the system's behaviour depend on its own history. Two flavours dominate: negative feedback, where output reduces the deviation that caused it, pushes the system toward a stable point (thermostats, insulin, cruise control). Positive feedback, where output amplifies the deviation, pushes the system away from equilibrium (microphone howl, bank runs, ice-age cascades).

The structure has three moving parts: a measurement of current state, a comparison to some reference or prior state, and an action that changes the state for the next round. Take away any one of them and the loop breaks. Most real systems carry a delay between action and next measurement, which is why feedback rarely settles instantly. It oscillates, overshoots, hunts for equilibrium, and sometimes blows through it.

What a feedback loop is not

The phrase gets used loosely. A few things people call feedback loops that aren't, or that miss what the word actually names.

• Not always stabilizing. "Feedback loop" in business writing often means a gentle self-correcting process. Half of feedback does the opposite. Positive feedback runs away from equilibrium, not toward it. Microphone howl, viral growth, and ice-sheet collapse are all feedback loops.
• Not a single event. One person giving another person feedback once is not a loop, it is a message. A loop exists only once the response is itself fed back into the next round of behaviour and the cycle keeps repeating.
• Not linear. A loop has a delay and can compound. Output at step N changes input at step N+1, which changes output at step N+2. Small effects multiply; small delays invert them into oscillation. Straight cause-then-effect isn't it.
• Not a metaphor. It is a literal mechanism with a directional arrow: output routes back to input. If you can't draw the arrow, the system doesn't have a feedback loop. It has, at most, a correlation.

Where do you see feedback loops in the wild?

In every stable living system and in every collapse. Homeostasis in the body, the ~37 °C you hold without thinking about it, is a wall of negative feedback loops: hypothalamic thermoregulation, blood-glucose regulation, pH buffering. Climate ice-albedo feedback is positive: more ice reflects more sunlight, which cools the planet, which grows more ice. Predator-prey cycles are a mixed loop: prey increase, predators follow, prey crash, predators starve, prey rebound. The cycle repeats because each output becomes the next period's input.

Social systems run on them too. Network effects are positive feedback (more users make a network more useful, which attracts more users). Price signals in markets are largely negative feedback (high prices reduce demand and increase supply). Attention algorithms create positive feedback on engagement, which is why a small initial bias can grow into a runaway content trend. Scroll a little on one topic and the feed shows you more of that topic, which invites more of that scroll.

Why do feedback loops matter?

Because they are the cleanest language for regulation and runaway. Norbert Wiener laid out the idea in his 1948 book Cybernetics, which launched the field of the same name and gave control engineering, electronics, and systems biology a shared vocabulary. The hardware was older than the book: in 1788 James Watt fitted a centrifugal governor to the steam engine, spinning balls that pulled a valve shut as speed rose and opened it as speed fell. First mechanical negative-feedback controller in wide use, and the reason the Industrial Revolution had steady-speed power at all.

The framing also tells you where a system is dangerous. Every wall thermostat quietly runs the same rule Watt's governor did. And every climate tipping point is a positive feedback loop waiting to cross a threshold: ice melts, dark ocean absorbs more heat, more ice melts. Once the positive term outweighs the negative terms holding it in check, the system does not drift back. It commits to a new regime. Knowing which loops stabilise a system and which ones can flip it is often the difference between a fixable problem and one that runs away.

Try it in the sim

The Nuclear Reactor simulation is a feedback loop you can dial. Each fission sends two to three neutrons into the pool, every neutron either hits another atom and triggers the next fission or is absorbed by a control rod or the moderator. That is positive feedback (the chain amplifying itself) and negative feedback (absorption damping it) running against each other in every frame. The rod slider shifts the balance.

• Run defaults. Watch the colony settle onto a stable trail. That is negative feedback winning, evaporation killing off alternate paths faster than they can be reinforced.
• Push Trail strength way up, Evaporation way down. Positive feedback dominates. The first random path becomes a permanent highway whether or not it is the shortest. The colony locks in, then cannot escape.
• Invert: low trail strength, high evaporation. Negative feedback overwhelms everything. No path ever builds. The colony forages like a gas, never converging.

Where feedback loops connect on this site

Feedback loops are the mechanism under almost every concept in the library. Stigmergy is a spatial feedback loop: agents modify the environment, the environment modifies agents. Self-organization needs at least one positive feedback term to form structure and at least one negative term to keep the structure bounded. Cascades are positive feedback running on a connected network. Optimization methods such as ant colony search and gradient descent all route output back as a guide for next-step input. The library holds them all. Free to use in a class on systems thinking or cybernetics.