Abstract
Subjects tracking a moving target are thought of as controlling their behaviour predictively wherever possible. This means that they work from the sensory input only indirectly, using it to build up and trigger off running ‘ hypotheses ’ of future events, which are then used to control behaviour. It is suggested that wherever redundancy in the external world, or learning by subjects, allows them to switch to a predictive mode of control, they do so. They respond directly to the input only where they realise that prediction is impossible. Where predictions are accurate, subjects can continue appropriately through gaps in sensory data. Where predictions are inaccurate, subjects may generate larger errors than they would with a simple following response. These features are illustrated in three demonstrations of predictive control; two where appropriate tracking movements were made with the target invisible, and one where an inappropriate movement was made to an unexpected event.