Operant conditioning

A car beeps until you fasten your seatbelt. The beep stopping is a reward of a sort psychologists call negative reinforcement.

Classical conditioning explains reactions that happen to you, such as drooling or fear. It does not explain why a child learns to ride a bicycle or why you check your phone so often. For behaviour you choose to make, the key is what happens after it. That is operant conditioning: behaviour followed by good consequences becomes more likely, behaviour followed by bad ones less likely.

Edward Thorndike put cats in puzzle boxes with a latch inside and food outside. At first escape came by accident; over many trials the cats escaped faster and faster. Thorndike summed it up as the law of effect: responses followed by satisfying consequences are repeated, and those followed by unpleasant ones are not. B. F. Skinner built on this in the 1930s with the operant chamber, now called a Skinner box, where a rat could press a lever or a pigeon peck a disk for food while a recorder counted every response.

Plus and minus do not mean good and bad

In this vocabulary, reinforcement always increases a behaviour and punishment always decreases it. Positive means something is added; negative means something is taken away. So there are four combinations. Positive reinforcement: add something pleasant (a treat for a trick). Negative reinforcement: remove something unpleasant (the seatbelt beep stops). Positive punishment: add something unpleasant (a scolding). Negative punishment: remove something pleasant (losing phone time). Negative reinforcement is the one most often confused with punishment; remember that it makes the behaviour more likely.

Shaping and schedules

A pigeon will not play ping pong on its own, so Skinner used shaping: reward a rough approximation, then require something a little closer each time. Rewards can be primary, such as food, or secondary, such as praise, money or tokens that gain value through their link to primary ones. How often reward comes also matters. Fixed ratio pays after a set number of responses (piecework). Variable ratio pays after an unpredictable number (gambling) and produces high, steady responding that is slow to extinguish. Fixed interval pays for the first response after a set time; variable interval after an unpredictable time (checking for new messages).

Skinner claimed he could turn a pigeon into a pathological gambler with a variable ratio schedule. OpenStax notes that a 2019 study in rats (Laskowski and colleagues) failed to support the idea that such schedules alone cause problem gambling. Skinner also treated the mind as a black box, but Edward Tolman showed that rats allowed to explore a maze without food for 10 sessions found their way quickly once food appeared. They had built a cognitive map with no reward, which he called latent learning. Learning is not only what reinforcement stamps in.