Errorless learning is a procedure introduced by Herbert Terrace (1963) which allows discrimination learning to occur with few or even with no responses to the negative stimulus (abbreviated S-). A negative stimulus is a stimulus associated with undesirable consequences (e.g., absence of reinforcement). In discrimination learning, an error is a response to the S-, and according to Terrace errors are not required for successful discrimination performance.
A simple discrimination learning procedure is one in which a subject learns to associate one stimulus, S+ (positive stimulus), with reinforcement (e.g. food) and another, S- (negative stimulus), with extinction (e.g. absence of food). For example, a pigeon can learn to peck a red key (S+), and avoid a green key (S-). Using traditional procedures, a pigeon would be initially trained to peck a red key (S+). When the pigeon was responding consistently to the red key (S+), a green key (S-) would be introduced. At first the pigeon would also respond to the green key (S+) but gradually responses to this key would decrease, because they are not followed by food, so that they occurred only a few times or even never.
Terrace (1963) found that discrimination learning could occur without errors when the training begins early in operant conditioning and visual stimuli (S+ and S-) like colors are used that differ in terms of brightness, duration and wavelength. He used a fading procedure in which the brightness and duration differences between the S+ and the S- were decreased progressively leaving only the difference in wavelength. In other words, the S+ and S- were initially presented with different brightness and duration, i.e., the S+ would appear during 5 s and fully red, and the S- would appear during 0.5 s and dark. Gradually, over successive presentations, the duration of the S- and its brightness were gradually increased until the keylight was fully green during 5 s.
The errorless learning procedure is highly effective in reducing the number of responses to the S- during training. In Terrace’s (1963) experiment, subjects trained with the conventional discrimination procedure averaged over 3000 S- (errors) responses during 28 sessions of training; whereas subjects trained with the errorless procedure averaged only 25 S- (errors) responses in the same number of sessions.
Later, Terrace (1972) claimed not only that the errorless learning procedure improves long-term discrimination performance, but also that: 1) S- does not become aversive and so does not elicit "aggressive" behaviors, as it often does with conventional training; 2) S- does not develop inhibitory properties; 3) positive behavioral contrast to S+ does not occur. In other words, Terrace has claimed that the "by-products" of conventional discrimination learning do not occur with the errorless procedure.
However, some evidence suggests that errorless learning may not be as qualitatively different from conventional training as Terrace initially claimed. For example, Rilling (1977) demonstrated in a series of experiments that these "by-products" can occur after errorless learning, but that their effects may not be as large as in the conventional procedure; and Marsh and Johnson (1968) found that subjects given errorless training were very slow to make a discrimination reversal.
Interest from psychologists studying basic research on declined after the 1970’s. However, errorless learning attracted the interest of researchers in applied psychology, and studies have been conducted with both children (e.g., educational settings) and adults (e.g. Parkinson’s patients).