A technique used to teach a desired behavior by reinforcing increasingly accurate approximations of the behavior.

Complex behaviors can be difficult to teach. Shaping is a strategy whereby complex behaviors are broken down into smaller, simpler subbehaviors, and then taught one by one. The behaviors are reinforced (e.g., given food), and ultimately chained together to achieve a desired result. For example, to teach a mouse to press a lever, the mouse is first reinforced to move close to the lever; then reinforced only when it makes contact with the lever; and eventually only when it presses the lever.

Often, shaping occurs without awareness. For example, video games use shaping when initial game levels require simple inputs in order to “beat” the level (obtain the reinforcement), and then require increasingly difficult controller actions to master higher levels of the game. Salespeople use a form of shaping when they offer a prize to come to their location, provide food and drink to discuss the sale, and then offer a discount for making a purchase decision that day. Each action toward the goal behavior (making the sale) is reinforced.

During shaping, behaviors that have nothing to do with the desired behavior can get incidentally reinforced. For example, when training a mouse to press a lever, the mouse may incidentally press a lever with one foot in the air. The reinforcement for the lever press may also inadvertently reinforce the fact that the foot was in the air. This behavior then becomes an integrated, but unnecessary component of the desired behavior; the mouse lifts its foot whenever it presses the lever. The development of this kind of superstitious behavior is common with humans as well.

Use shaping to train complex behaviors in games, simulations, and learning environments. Shaping does not address the “hows” or “whys” of a task, and should, therefore, primarily be used to teach rote procedures and refine complex motor tasks. Shaping is being increasingly used to train complex behaviors in artificial beings, and should be considered when developing adaptive systems.

See also Classical Conditioning and Operant Conditioning.

Project

Pigeon Project Pigeon was a classified research-and-development program during World War II. It was developed at a time when electronic guidance systems did not exist, and the only compensation for the inaccuracy of bombs was dropping them in quantity. This ingenious application of shaping would have dramatically increased the accuracy of bombs and decreased civilian casualties. Despite favorable performance tests, however, the National Defense Research Committee ended the project—it seems they couldn’t get over the idea that pigeons would be guiding their bombs.

1. Pigeons were trained to peck at targets on aerial photographs. Once a certain level of proficiency was obtained, pigeons were jacketed and mounted inside tubes.

2. The pigeons in their tubes were inserted into the nosecone of the bomb. Each nosecone used three pigeons in a type of voting system, whereby the pigeon pecks of two birds in agreement would overrule the errant pigeon pecks of a single bird.

3. Sealed in the bomb, the pigeons could see through glass lenses in the nosecone.

4. Once the bomb was released, the pigeons would begin pecking at their view of the target. Their pecks shifted the glass lens off-center, which adjusted the bomb’s tail surfaces and, correspondingly, its trajectory.