Can random experimental choice lead to better theories?