Richard J. Robertson in The Early Days of Perceptual Control Theory: (Vol 4 #1 of Closed Loop)
During the tournaments, a number of us noted a curious change that would quite regularly occur in the play of a contestant when he began to recognize that he was clearly outmatched. He would first concentrate very hard, then begin to alternate between wild shots and cautious play. It occurred to me that these periods of variability, if we could graph them, would look like the patterns of RT variation preceding a new plateau in the Powers Game experiment. The participants themselves acknowledged this aspect of play as part of their experiments to obtain eventual increases in skill. In this view, what would have seemed a lapse into sloppiness on the part of a losing player took on an opposite significance.
Compare with what a commenter wrote over at Cal Newport’s blog:
It’s too hard to simultaneously do and diagnose, at first, and the only way to actually succeed at things at the boundary of one’s ability is by accident. So make sure to get many, many attempts at them, to have more chances to do them right by accident. Once that happens, you can get a sense of “what did I do differently then, that made that finally work?”
The key phrase here being “do them right by accident”.
Robertson again:
Powers made the implication that the organism does not know exactly what change must occur. Random excitation caused by the reorganizing system results in various alterations of action. Then the action that begins to bring the desired objective under control becomes the basis of a new control system.