Feedback is the arrangement in which part of a system’s output is routed back to affect its own input. In negative feedback, the system measures the difference between where it is and where it should be and acts to reduce that gap, the way a thermostat switches a heater off as the room approaches the target temperature, or a ship’s autopilot nudges the rudder to hold a course. In positive feedback, the output reinforces itself, which can produce rapid growth or runaway change. The concept was placed at the center of a general science of control by Norbert Wiener in his 1948 book “Cybernetics,” the primary source used here.
The power of the idea is that the same simple loop, sense, compare to a goal, correct, explains purposeful, goal-seeking behavior in systems that have nothing else in common. A governor on a steam engine, a body regulating its temperature, an animal reaching for food, and an electronic circuit holding a voltage steady can all be described in the same terms. This let engineers and scientists treat goal-directed behavior, long thought to be special to living minds, as something that could be built and analyzed mathematically.
Feedback is foundational to the prehistory of artificial intelligence because it offered a mechanical account of purpose and adaptation. Grey Walter’s tortoises and Ashby’s Homeostat were feedback machines, and the broader cybernetics movement that grew up around the concept fed directly into early thinking about brains, learning, and machine intelligence. The same principle underlies how modern systems learn: they adjust based on the difference between their output and a desired result.
For a general reader, feedback is one of the most useful ideas to carry around, because it explains how systems, from a heating thermostat to a learning algorithm to a business adjusting to its market, can correct themselves and pursue a goal without anyone steering them step by step.