Feedback is information about the effect of an action, process or system state that is returned so future behaviour can be adjusted. If you are asking “what is feedback?”, the useful answer ranges from a teacher commenting on a student’s work to a thermostat measuring room temperature, a controller correcting a machine, a biological system maintaining homeostasis and a computer algorithm updating from error. Feedback closes a loop: something acts, the result is observed, the difference matters, and the next action changes in response.
A complete explanation of feedback must distinguish everyday comments from feedback loops. In a conversation, feedback may be a reply, expression or correction. In control systems, feedback means measured output is returned to a controller and compared with a target. In learning, feedback is information that helps a learner or teacher decide what needs to change next. Across these domains, the central mechanism is similar: the system uses consequences of previous behaviour to modify future behaviour.
This guide explains feedback from first principles: feedback loops, positive and negative feedback, reinforcing and balancing loops, control systems, sensors, setpoints, error signals, delays, gain, stability, oscillation, proportional and PID control, feedforward, homeostasis, biological regulation, climate feedback, organisations, communication, education, learning, artificial intelligence and common feedback failures. The goal is a durable model: identify what is measured, what target or reference matters, how the difference is interpreted, what action follows and whether the loop actually moves the system toward the desired state.
What Is Feedback? A Short Definition
Feedback is information about what happened after an action that is used to influence what happens next. The defining feature is return. Output from one stage becomes input to a later stage.
A microphone feeding a loudspeaker can create acoustic feedback. A thermostat measures temperature and changes heating. A learner sees which answer was wrong and changes the method used next time.
Feedback therefore turns a one-way process into a loop. Without feedback, a system acts without using evidence about the consequences of its own behaviour.
The Anatomy of a Feedback Loop
Most feedback loops contain several elements: a system or process, an output, a measurement, a reference or goal, a comparison and an action that changes future output.
In a heating system, the room is the system, temperature is the output, a sensor measures it, the thermostat contains a target, the controller compares actual with desired temperature and the heater changes heat input.
The exact vocabulary differs across fields, but the logic is stable: observe → compare → correct → observe again.
Feedback Versus Information
Not every piece of information is feedback. A weather report is information; it becomes feedback if it is used to adjust an ongoing process such as irrigation or flight operations.
Feedback is therefore relational. It has meaning because it is connected to a previous action or system state and to a future adjustment.
The same observation can be feedback for one system and mere information for another.
Feedback Versus Advice
Advice tells someone what to do. Feedback describes or interprets the consequences of what has already happened.
The two can be combined: “Your introduction states the topic but not the argument; add a clear thesis before the evidence.” The first clause is feedback; the second is advice.
Keeping them distinct helps because useful correction depends on accurate diagnosis before prescription.
Positive and Negative Feedback
In systems language, positive and negative do not mean good and bad. Positive feedback reinforces change; negative feedback counteracts deviation.
A reinforcing loop can produce growth or runaway failure. A balancing loop can produce stability or suppress necessary change.
The sign describes the direction of the loop’s effect, not its moral value.
Positive Feedback: Reinforcing Loops
Positive feedback amplifies an initial change. An increase causes effects that produce further increase, or a decrease produces effects that drive further decrease.
Examples include microphone squeal, bank runs, network effects and some biological cascades. More adoption can make a platform more useful, which attracts still more adoption.
Reinforcing loops can create rapid growth, tipping points and lock-in. They often require a limiting process eventually because unbounded exponential growth is rarely physically sustainable.
Negative Feedback: Balancing Loops
Negative feedback reduces the gap between a current state and a target or stable range.
Thermostats, cruise control and many biological homeostatic systems use balancing feedback.
Negative feedback is the foundation of regulation, but poor timing or excessive correction can create oscillation rather than smooth stability.
Feedback Is Not Automatically Correction
A feedback signal can exist without producing a useful adjustment. A warning light may be ignored. A student may receive comments but not understand how to act on them.
The loop closes only when the information changes behaviour or system input.
This is why “providing feedback” and “using feedback” should be treated as different events.
Open-Loop and Closed-Loop Systems
An open-loop system acts without measuring the result. A basic toaster may heat for a fixed time regardless of how brown the bread becomes.
A closed-loop system measures output and adjusts future action. A temperature controller changes heating based on measured temperature.
Closed-loop control can compensate for disturbances, but it also depends on sensor quality, timing and controller design.
The Setpoint
A setpoint is a reference value that a control system tries to maintain or approach.
A thermostat might use 24 °C as its target. Blood glucose regulation operates around ranges rather than one perfectly fixed point.
Not every feedback system has an explicit numerical setpoint. Some systems regulate relative differences, constraints or moving targets.
Error Signals
An error signal is the difference between the desired and observed state.
If a motor should rotate at 1,000 revolutions per minute but is measured at 950, the error is the deviation that guides correction.
The word “error” here is technical. It means discrepancy, not necessarily a mistake.
Sensors
Sensors measure the state needed for feedback. They can detect temperature, speed, pressure, position, concentration, light or many other variables.
Bad sensor data produce bad feedback. Noise, bias, delay and calibration error can cause the controller to correct the wrong problem.
Good feedback therefore depends on measurement quality before it depends on response quality.
Controllers
A controller decides how to respond to the feedback signal.
The controller may be mechanical, electronic, biological, computational or human. It translates discrepancy into action.
Controller design determines whether the system corrects gently, reacts aggressively, oscillates or becomes unstable.
Actuators
An actuator changes the system in response to control signals. A heater, valve, motor or muscle can function as an actuator.
Actuators have limits. They cannot respond infinitely fast or with unlimited force.
Saturation matters because a controller can demand more correction than the actuator can physically deliver.
The Plant or Controlled Process
In control theory, the “plant” is the system being controlled. It could be a motor, aircraft, chemical process or room temperature.
The plant has its own dynamics: inertia, delays, nonlinearities and disturbances.
A controller that works for one plant may fail for another because the response characteristics differ.
Disturbances
A disturbance is an external influence that pushes the system away from its intended state.
Opening a window disturbs room temperature. A hill disturbs vehicle speed. A sudden demand surge disturbs inventory.
Feedback is valuable because it can correct for disturbances that were not known perfectly in advance.
Delay
Feedback takes time. Sensors measure, signals travel, controllers calculate and systems respond.
Delay can destabilise a loop because the controller acts on an old state. By the time the correction arrives, the system may already have moved past the target.
Many oscillations in organisations, supply chains and biological systems arise from feedback combined with delay.
Gain
Gain describes how strongly the controller responds to an error.
Low gain can make correction slow. Excessively high gain can cause overshoot or oscillation.
The right gain depends on system dynamics, noise and the cost of being slow versus unstable.
Stability
A stable feedback system returns toward a bounded operating state after disturbance rather than diverging uncontrollably.
Stability is one of control theory’s central questions.
A system can respond quickly yet be unstable, or respond slowly yet remain robust. Good control balances speed, accuracy and stability.
Overshoot
Overshoot occurs when correction carries the system beyond the target.
A heating system may continue warming after the heater switches off because stored heat remains. A learner may overcorrect style after receiving criticism and make writing unnaturally formal.
Understanding overshoot reminds us that system momentum matters after the feedback decision is made.
Oscillation
Oscillation occurs when a system repeatedly moves above and below a target.
It can arise from delay, strong feedback, interacting loops or poor damping.
Oscillation is not always failure—some systems are naturally rhythmic—but unwanted oscillation reduces control quality and can increase wear or instability.
Damping
Damping reduces oscillation by dissipating or counteracting motion.
Mechanical shock absorbers damp vibration. Control systems can include terms that respond to rate of change rather than only current error.
Damping illustrates a broader feedback principle: correcting the present state may be insufficient if momentum predicts where the system is heading next.
Proportional Control
Proportional control makes corrective action proportional to current error.
Large error creates a strong response; small error creates a weaker response.
Proportional control is intuitive and useful, but some systems retain a steady residual error unless additional control terms are added.
Integral Control
Integral control accumulates error over time.
If a small discrepancy persists, the integral term grows until enough corrective action is applied.
This can remove steady-state error, but excessive integration can contribute to overshoot and windup when actuators saturate.
Derivative Control
Derivative control responds to the rate at which error is changing.
If the system is approaching the target rapidly, derivative action can reduce correction before overshoot occurs.
Derivative terms are sensitive to noisy measurements, so filtering and practical implementation matter.
PID Control
PID control combines proportional, integral and derivative actions.
It is widely used because the three terms address present error, accumulated error and error trend.
PID is not universally optimal, but it provides a flexible practical framework for many industrial systems.
Feedforward Versus Feedback
Feedforward acts before the error appears by using information about expected disturbances or required input.
Feedback acts after measuring the system’s response.
The two complement each other. A vehicle can anticipate a hill from a map while still using speed feedback to correct unexpected differences.
Why Feedback Is Powerful
Feedback makes systems adaptive to uncertainty. Instead of needing a perfect prediction of every disturbance, the system observes what actually happened and adjusts.
This allows control even when the plant or environment is only partly known.
The price is dependence on timely, trustworthy measurement and a controller whose corrections do not create new instability.
Cybernetics
Cybernetics is the interdisciplinary study of control and communication in animals, machines and organisations.
It emphasises feedback, regulation, information and goal-directed behaviour.
Cybernetic thinking helped connect engineering control with biological homeostasis and organisational regulation.
Homeostasis
Homeostasis is the regulation of internal conditions within viable ranges.
Body temperature, blood glucose, fluid balance and many physiological variables involve feedback mechanisms.
Homeostasis does not mean perfect constancy. Biological systems fluctuate and often regulate around ranges that depend on context.
Thermoregulation
Human thermoregulation uses sensors, neural control and effectors such as sweating, shivering and blood-vessel changes.
Deviation from a viable temperature range produces responses that alter heat loss or production.
The system contains multiple loops and environmental interactions, making it richer than a simple household thermostat.
Hormonal Feedback
Endocrine systems often use feedback to regulate hormone levels.
A downstream hormone can suppress upstream signalling, creating negative feedback across several organs.
Hormonal loops include delays and nonlinear responses, which is why dynamics matter as much as one static concentration.
Ecological Feedback
Ecological systems contain feedback among populations, resources and environments.
Predator abundance affects prey, prey abundance affects predator reproduction and vegetation can alter habitat conditions that later affect species.
Ecological feedback can stabilise some relationships or amplify change when thresholds are crossed.
Climate Feedback
Climate feedback describes processes that amplify or reduce an initial climate change.
For example, warming can reduce reflective ice cover, increasing absorbed solar energy and reinforcing warming. Other processes can provide balancing effects.
Climate feedbacks are quantified scientifically and should not be reduced to the everyday idea of “feedback comments”.
Feedback in Earth Systems
Earth systems connect atmosphere, ocean, land, ice and life through flows of energy and matter.
Feedback loops operate across timescales from days to centuries.
NASA systems-thinking materials use feedback loops to help learners trace how changes in one Earth subsystem can propagate through others.
Feedback in Economics
Economic systems contain feedback through prices, expectations and behaviour.
Higher prices can reduce demand, providing balancing feedback. Rising asset prices can attract buyers expecting further rises, creating reinforcing feedback.
Because economic agents adapt strategically, feedback can change as participants learn.
Feedback in Markets
Markets use prices as signals that influence future supply and demand.
The signal can coordinate decentralised decisions, but it may also create reinforcing speculation or delayed investment cycles.
Market feedback is therefore a system mechanism, not proof that every price is perfectly informative.
Feedback in Organisations
Organisations use reports, reviews, customer data and performance measures as feedback.
The usefulness of that feedback depends on whether measures represent the real goal and whether people can act on the information.
If incentives reward one metric strongly, people may optimise the metric instead of the underlying purpose.
Feedback and Goodhart’s Law
Goodhart’s Law warns that when a measure becomes a target, it can lose value as a measure.
A customer-service team rewarded only for call duration may shorten calls without solving problems.
The feedback loop changed behaviour, but the measured variable no longer tracked the real objective well.
Feedback in Communication
Conversation is a feedback-rich process. Speakers observe responses, facial expressions, questions and misunderstandings, then adjust wording.
Without feedback, communication becomes broadcasting rather than interaction.
Useful communicative feedback tests meaning: did the receiver reconstruct the intended message?
Feedback in Learning
In learning, feedback gives information about performance relative to a goal, criterion or strategy.
It can identify errors, clarify standards and guide the next attempt.
Feedback is most useful when the learner can understand it and has an opportunity to act on it. Comments without uptake do not complete the learning loop.
Formative Feedback
Formative feedback is intended to improve performance while learning is still ongoing.
It answers questions such as: Where am I going? How am I doing? What should I do next?
Its value depends on timing, specificity and whether the next action remains cognitively useful rather than simply revealing the answer.
Summative Results as Feedback
A final grade or test score can function as feedback, but it is often too compressed to diagnose what should change.
An 82 does not reveal whether lost marks came from conceptual misunderstanding, method selection, carelessness or time management.
Useful improvement often requires finer-grained evidence than the final score alone.
Specificity
Specific feedback identifies the part of performance that matters.
“Good work” may encourage but provides little diagnostic information. “Your conclusion answers the question, but the evidence does not yet support the causal claim” identifies a repairable gap.
Specificity should target the task and process rather than turn one performance into a judgment of the person.
Actionability
Actionable feedback makes the next move visible.
A learner should know what to revise, practise, compare or check.
Actionability does not require prescribing every step. The goal is enough direction to support independent correction.
Timing
Feedback timing changes its effect. Immediate feedback can prevent error repetition, while delayed feedback can force retrieval and self-evaluation before correction.
The best timing depends on task complexity, learner knowledge and the risk of practising an error.
There is no universal rule that faster feedback is always better.
Feedback Uptake
Feedback uptake is the learner’s or system’s actual response to feedback.
A comment can be accurate and still fail if it is ignored, misunderstood or impossible to implement.
Measuring uptake requires observing changed behaviour or performance, not merely confirming that the message was delivered.
Feedback Literacy
Feedback literacy is the ability to interpret feedback, judge its relevance, manage emotional response and use it productively.
Learners need to distinguish comments about task quality from judgments about identity.
They also need enough subject knowledge to understand what the feedback means.
Peer Feedback
Peers can provide feedback by evaluating examples, explaining criteria and identifying unclear reasoning.
Giving feedback can itself be educational because the evaluator must apply standards explicitly.
Peer feedback requires structure and examples so inaccurate advice does not simply circulate.
Self-Feedback
Self-feedback occurs when a learner or operator compares performance with criteria without waiting for another person.
Checklists, worked examples and self-testing can support this process.
Expertise includes internalising standards so external correction becomes less necessary over time.
Feedback and Metacognition
Feedback can improve metacognition by revealing the gap between predicted and actual performance.
A student who expected 95% but scored 70% receives evidence about both subject performance and confidence calibration.
Repeated prediction and feedback can improve judgment about what is genuinely known.
Feedback and Motivation
Feedback affects motivation because it changes beliefs about competence, progress and controllability.
Feedback focused on specific strategies and improvement can support agency more effectively than global labels.
However, motivational effects vary by person, context and relationship. Feedback should not be engineered through praise alone while ignoring accuracy.
Feedback and Error
Error is useful when it contains information about the system.
A failed attempt can reveal a missing concept, wrong assumption or poor procedure.
The goal is not to celebrate errors indiscriminately; it is to extract information and prevent the same mechanism from recurring.
Feedback in Engineering
Engineering systems use sensors and controllers to maintain position, speed, temperature and other variables.
Feedback allows systems to reject disturbances and compensate for model uncertainty.
MIT’s control-system materials use feedback as a foundational principle linking signals, stability and correction.
Feedback in Software
Software development uses feedback through tests, logs, monitoring, user reports and deployment metrics.
Fast feedback helps localise bugs before changes accumulate.
But metrics must represent meaningful behaviour; otherwise teams can optimise dashboards while user experience deteriorates.
Debugging as Feedback
Debugging is a feedback process: change code, run the system, observe behaviour and update the hypothesis.
A good test isolates one possible mechanism so the result discriminates among explanations.
Random changes provide feedback too, but weakly, because many variables change at once and causality becomes unclear.
Feedback in Artificial Intelligence
AI systems can learn through feedback in several ways. Supervised learning uses labelled examples; reinforcement learning uses reward signals; deployed systems may receive user interaction data.
Feedback signals shape optimisation, so poorly specified rewards can produce behaviour that satisfies the metric without satisfying the real goal.
This is another version of Goodhart’s problem: the proxy becomes a target.
Reinforcement Learning
Reinforcement learning trains an agent to choose actions that maximise expected cumulative reward.
The reward is feedback about consequences, but it may be delayed and sparse.
Designing reward functions is difficult because the system can discover strategies the designer did not anticipate.
Human Feedback in AI
Human preferences can be used to adjust AI behaviour through ranking, critique or reward modelling.
This can improve usefulness, but human feedback is itself noisy, culturally variable and incomplete.
The process therefore requires evaluation beyond whether annotators liked the output in development.
Feedback Loops in Recommendation Systems
Recommendation systems influence what users see, and user responses then become future training signals.
This creates endogenous data: the system helps generate the behaviour it later learns from.
Feedback loops can amplify popularity, narrow exposure or stabilise preferences, so evaluation must consider system-level effects over time.
Feedback and Adaptation
Feedback enables adaptation by connecting consequences to future behaviour.
Biological organisms learn, organisations revise policy and control systems correct disturbances because information returns from action.
Adaptation is therefore feedback with memory: the response influences future rules or parameters, not just the next moment.
Feedback and Resilience
Resilient systems detect disturbance and respond before failure spreads.
Monitoring, redundancy and local correction can provide feedback that contains problems early.
Too much centralisation can slow feedback, while too much local autonomy can fragment coordination. Resilience depends on architecture.
Feedback and Learning Rates
An adaptive system can update too slowly or too quickly.
Slow updating ignores useful evidence. Fast updating can overreact to noise and destabilise behaviour.
The appropriate learning rate depends on how noisy the environment is and how quickly the true process changes.
Noise
Noise is variation in feedback measurements unrelated to the underlying state of interest.
Reacting strongly to noise creates unnecessary correction.
Filtering, repeated measurement and robust statistics can improve feedback quality, but excessive smoothing can hide real change.
Signal-to-Noise Ratio
Signal-to-noise ratio describes how distinguishable the meaningful signal is from background variation.
Low signal-to-noise feedback requires caution because one measurement may be misleading.
In learning, one bad test can be an outlier; repeated patterns across tasks provide stronger evidence.
Measurement Bias
Biased feedback consistently misrepresents the true state.
A miscalibrated sensor can drive a controller to the wrong setpoint. A rubric that rewards superficial features can push students toward superficial writing.
Feedback quality therefore begins with valid measurement rather than eloquent commentary.
Feedback Delay in Organisations
Organisations often make decisions using lagging indicators such as quarterly results.
If the delay is long, leaders can continue an ineffective policy before consequences become visible.
Leading indicators can shorten feedback, but they must be validated so they do not become convenient proxies detached from real outcomes.
Runaway Feedback
Runaway feedback occurs when reinforcing effects grow faster than limiting processes can contain them.
Acoustic squeal is a simple physical example. Financial bubbles and viral spread can show analogous reinforcing dynamics.
Runaway behaviour often ends when a constraint, saturation effect or balancing loop becomes dominant.
Saturation
Saturation occurs when further input cannot produce proportional additional output.
An actuator reaches its maximum. A market becomes fully penetrated. A learner’s working time cannot increase beyond available hours.
Controllers that ignore saturation can accumulate excessive corrective demand and behave badly when the limit clears.
Windup
Integral windup occurs when an integral controller keeps accumulating error while the actuator is saturated.
When the actuator finally regains control authority, the accumulated correction can cause severe overshoot.
The example generalises: systems that keep escalating pressure while action capacity is capped can overcorrect once the bottleneck disappears.
Competing Feedback Loops
Real systems contain multiple loops operating simultaneously.
A growth loop may increase demand while a balancing resource constraint slows it. A learner may gain confidence from success while fatigue reduces performance.
System behaviour comes from the interaction among loops, not from identifying one loop in isolation.
Nested Feedback
Feedback loops can sit inside larger loops. A motor controller regulates speed inside a vehicle whose driver regulates route and arrival time.
Organisations similarly contain personal, team and system-level feedback.
Conflicts occur when lower-level optimisation harms higher-level goals, so nested loops need aligned objectives.
Feedback and Targets
A feedback loop is only as good as the target it regulates.
If the setpoint is wrong, perfect control produces the wrong result reliably.
This matters in organisations and education because choosing a metric or benchmark embeds a value judgment about what success means.
Feedback Without a Fixed Target
Some feedback systems adapt without one permanent setpoint.
Evolution, markets and learning systems may update relative to changing environments rather than converge on one fixed value.
The feedback objective can therefore be viability, performance improvement or error reduction rather than exact target maintenance.
Common Feedback Failures
- Bad measurement: the signal does not represent the state that matters.
- Delay: corrections arrive after the system has already changed.
- Excess gain: response is too strong and produces overshoot or oscillation.
- Weak gain: correction is too small to overcome disturbances.
- Wrong target: the loop regulates an inappropriate metric.
- Ignored feedback: information is produced but does not change action.
- Proxy gaming: agents optimise the measure rather than the real goal.
- Loop blindness: decision-makers respond to one pathway while another feedback loop dominates behaviour.
A Practical Feedback Checklist
Identify the system, output, measurement, reference and correction. Ask what information is returned and who or what can act on it.
Check timing. How long does measurement take? How long before the action affects output? Could the system overshoot before the feedback arrives?
Then test quality. Is the signal accurate, specific and actionable? What noise, bias, saturation or unintended incentive could distort the loop?
How to Design Better Feedback
Start with the real objective, not the easiest metric. Select a measurement that represents meaningful progress toward that objective.
Make the loop short enough to support correction but not so reactive that it chases noise.
Provide a next action and then observe the next performance. Feedback becomes useful only when the system receives a new receipt showing whether the adjustment worked.
How to Use Feedback for Learning
Before feedback arrives, attempt the task and record your reasoning. This preserves evidence about what you could do independently.
After feedback, identify the smallest repair: concept, method, evidence, explanation or checking routine. Practise that repair on a fresh example rather than simply editing the original answer.
Return later without the feedback visible. The delayed attempt shows whether the correction became part of independent capability.
How to Learn Feedback Properly
Begin with a thermostat. Draw the target, sensor, error, controller, actuator and room.
Then compare a reinforcing loop such as network adoption. Notice how the same idea—output returning as input—can stabilise one system and amplify another.
Finally, analyse a human feedback process. Separate observation, interpretation, advice, action and later evidence. The loop becomes clearer once each stage is named.
Frequently Asked Questions About Feedback
What is feedback in simple terms?
Feedback is information about the result of an action or process that is used to influence what happens next.
What is a feedback loop?
A feedback loop occurs when a system’s output is measured and returned as input to affect future behaviour.
What is positive feedback?
Positive feedback reinforces change, causing an initial increase or decrease to produce more change in the same direction.
What is negative feedback?
Negative feedback counteracts deviation and tends to move a system toward a target or stable range.
Why can negative feedback oscillate?
Delay, excessive gain and system inertia can make corrections arrive too late or too strongly, producing repeated overshoot.
What is the difference between feedback and feedforward?
Feedback reacts to measured outcomes; feedforward anticipates disturbances or required action before the error occurs.
Why does feedback sometimes fail in education?
Comments may be vague, poorly timed, misunderstood, emotionally rejected or impossible to act on. Feedback also fails when the underlying measurement does not represent the learning goal.
How does AI use feedback?
AI can use labelled examples, reward signals, human preferences and deployment data to update model behaviour. The quality of the feedback signal strongly shapes what the system optimises.
Authoritative Starting Points
For control systems, MIT OpenCourseWare’s Feedback and Control materials introduce feedback, system response and control. MIT’s system-dynamics resource An Introduction to Feedback develops reinforcing and balancing loops. NASA’s Systems Thinking About the Earth System provides accessible examples of feedback loops across interacting Earth processes.
What Is Feedback? The Complete Idea
Feedback is the mechanism by which consequences return to shape future behaviour. It connects action with measurement, measurement with comparison and comparison with adjustment.
The strongest understanding keeps several distinctions clear: positive feedback is not automatically good; negative feedback is not automatically bad; information is not feedback unless it influences the loop; comments are not useful merely because they were delivered; and measurement quality determines the quality of the correction.
Once those distinctions are stable, feedback becomes a general language for regulation and learning. Define the goal, observe the state, measure the gap, respond with appropriate strength, account for delay and then measure again. Whether the system is a thermostat, organism, classroom, organisation or AI model, feedback turns consequences into information that can change what happens next.