A single light sensor can’t see motion — it only knows bright or dark, now. But take TWO neighbouring sensors, delay one, and multiply: the signal is big only when something moved from the first to the second in exactly that delay. Delay-and-compare is how a fly (and your retina) reads direction. Slide the motion and watch the detector swing positive one way, negative the other.
The Hassenstein–Reichardt detector (1956): two adjacent photoreceptors A and B; one signal is temporally DELAYED then correlated (multiplied) with the other, and two mirror subunits subtract. A bump crossing A→B in the delay time gives a large POSITIVE output; B→A gives NEGATIVE — direction selectivity built from two non-directional inputs. A fail-loud self-check throws unless rightward yields positive and leftward negative. ◆ real neuroscience, node-verified.
A minimal two-subunit correlator on a Gaussian bump; real motion vision stacks many across space and speed (a motion-energy model). The delay-and-correlate giving opposite signs for opposite directions is the exact defining behaviour.