SynopsisThe training of human operators for skilled tasks may be regarded as the synthesis of a specific controller from a general-purpose adaptive device, by influencing its adaption through selection and variation of the learning environment. Selection of environments to maximise the rate of learning is itself a control problem, and an automatic feedback training system is proposed which feeds back information about the operator's performance to control the parameters of his environment. The stability and performance of the trainer have been investigated both theoretically and experimentally, and its utility has been tested in a fairly realistic training situation using as trainees both human operators and computer-simulated learning machines.